Hazel Duvall's Blog

This Is Why My Latest iOS CI Check Failed

Wed, 21 Jan 2026 12:00:00 GMT

Video courtesy of my coworker. This is running inside the iOS simulator:

Yeah ok sure whatever.

Video description in case it doesn't load: A blue circle fades into view at the top, indicating a tap on a chosen element. Right as the tap is about to occur, however, a system notification reading "Ready for Apple Intelligence? Time to experience the new personal intelligence system." swoops in from the top, intercepting the tap. This takes the simulator away from the app being tested into settings, causing the test to fail.

Click-To-Open In Neovim With Kitty

Fri, 14 Nov 2025 12:00:00 GMT

Recently, I saw my coworker using VSCode, and they had this amazing feature where they could:

Run ripgrep in the command pane
Click on a filename it outputted
To immediately open it in a new tab in the editor pane

This, to me, was magical. However, I thought it out-of-reach for a lowly terminal user like myself, until I saw what jyn was able to do with their terminal. This inspired me to create my own customized setup, which I shall now explain in excruciating detail.

The Basics

The Shell

In order to actually have anything clickable, we’ll first need to set up our command line tools to output Terminal Hyperlinks that point to the full path of the filename displayed. The tools I uses for this are rg (searching by file contents), fd (searching by filename), & eza (listing directories).

# ~/.zshrc
alias rg="command rg --hyperlink-format=default"
alias fd="command fd --hyperlink=auto"
# TODO: use --hyperlink=auto once eza supports it
function ls {
  if [ -t 1 ]; then
    command eza --hyperlink "$@"
  else
    command eza "$@"
  fi
}

Because Terminal Hyperlinks are just extra text that’s interpreted + hidden by the terminal (similar to how terminal colors work), we need to be careful that this extra text is only written to the terminal, not to any pipes (|) or file redirection (>). Fortunately, most programs supporting colors/hyperlinks already have ways to detect if they’re writing to the terminal or not. Unfortunately, “most” is not “all”, and for some reason eza still only has an unconditional --hyperlink flag despite a PR about it being open for some time¹, so we need to detect this manually. Overall not too hard to do tho. Next!!

The Terminal

I’ve been using Kitty as my terminal for some time now, and have become quite comfortable with it. Fortunately, it supports scripting what happens when you click a Terminal Hyperlink, so we’ll do just that! The manual page lists a lot more things we can do with these clicks, but for now we just care about text editing.

# ~/.config/kitty/open-actions.conf
# Note this is _not_ included from the main kitty.conf, and is its own file
protocol file
action launch --type=window --cwd=current -- zsh -c "source ~/.zshrc; nvim -- $FILE_PATH"

Then, any time you use rg or any of the other hyperlinked commands, all relevant filenames can be Ctrl+Shift+Left-Click-ed to open them in a new window inside the current Kitty tab. protocol file tells Kitty “any link starting with file://, use the following action”, and action launch tells Kitty exactly how to launch our editor.

You may be wondering, “why not just use $EDITOR -- $FILE_PATH, why is $SHELL involved at all?” And that’s a great question. See, there’s a lot of stuff going on in my .zshrc, from homebrew/nix-darwin globally installed packages, to per-directory workspaces made with direnv/mise-en-place.

Needless to say, my system is pretty bare without its $PATH filled to the brim with all these goodies. In fact, many of my Neovim plugins require it: I need rg & fd for snacks.nvim, language servers for blink.cmp, git for fugitive.vim, y’know, the works. However (for myriad reasons beyond the scope of this post), when launching a new window, Kitty isn’t quite smart enough to copy environment variables like $PATH from the currently active window². So, we just have to re-compute our $PATH before launching our editor. The reason I do -c "source ~/.zshrc" instead of -i is for speed; launching zsh first means there will be a brief flash of shell before the editor launches, and the former approach seems to minimize that flash.

The Complications

“Well golly gee willikers,” I hear you say, “Being able to open inside a new Kitty window sure is dandy and all, but what I really want to do is open files I click on inside a new Neovim tab, in an existing Kitty window ” Well fear not, dear reader, because I thought those same thoughts & have just the solution for you! This requires a just a bit more setup, which I have put together in the form of a “simple” shell script:

#!/usr/bin/env bash

# Ignore the ~/.local/bin directory when searching for nvim
# (assumes this script has been placed at ~/.local/bin/nvim)
export PATH=$(echo $PATH | sed -E -e "s|${HOME}/.local/bin(:\|$)||")

if [ ! -z "${KITTY_TAB_ID}" ]; then
  # If running in Kitty, use the tab ID to get the pipe for the Neovim instance for that tab.
  pipe="${XDG_CACHE_HOME:-${HOME}/.cache}/nvim/server-${KITTY_TAB_ID}.pipe"
  if [ ! -S "${pipe}" ]; then
    # Launch instance if there isn't one already
    if [ "$1" = "-k" ] || [ "$1" = "--kitty-remote" ]; then
      shift
    fi
    nvim --listen "${pipe}" "$@"
  else
    # Pipe already exists
    if [ "$1" = "-k" ] || [ "$1" = "--kitty-remote" ]; then
      # Open files in existing instance
      shift
      nvim --server "${pipe}" --remote-tab "$@"
    else
      # Launch new, non-remote instance
      nvim "$@"
    fi
  fi
else
  # If not running in kitty, pass through arguments unchanged
  nvim "$@"
fi

There are comments inline explaining most of it, but to summarize:

Use a unique-per-tab ID to construct a path inside a directory for Neovim to listen on
If such a path doesn’t already exist, starts Neovim listening on that path
If such a path does exist, only connect to it when a special command-line flag is given.

I chose the -k flag (shorthand for --kitty-remote), which also happens to be unused for all of nvim, rg, fd, and eza surprisingly, that I then use in my updated Kitty open-actions.conf:

# ~/.config/kitty/open-actions.conf
protocol file
action launch --type=window --cwd=current -- $SHELL -c "source ~/.zshrc; nvim --kitty-remote $FILE_PATH"

This replaces the protocol/action block we had previously. We get rid of the -- because the --remote-tab used in our script already treats all remaining arguments as filenames.

The nice part about having this as a full-blown script, as opposed to just a shell function, is that we can use it with xargs, leading to some even crazier aliases:

# ~/.zshrc
function rgim {
  extra_args=""
  if [ "$1" = "-k" ]; then
    shift
    extra_args="-k"
  fi
  rg -l "$@" | xargs nvim ${extra_args}
}
alias rgkm="rgim -k"

Meaning, rgkm foobar will open all files containing the string “foobar” in separate tabs in an existing Neovim pane. Or, you can rg foobar to see the matches before clicking on the individual filesnames, and they will open in the existing Neovim just the same. Magical!!

Notes On Getting `$KITTY_TAB_ID`

Unfortunately, dear reader, I may have pulled a fast one on you in the previous section; we are not done, there is one block remaining in this tower of shell scripts. Kitty only provides a unique-per-window-ID, $KITTY_WINDOW_ID, where a “window” is not an OS window but rather a split within a given tab. We require a unique-per-tab-ID instead, thus we are forced to calculate it by hand:

# ~/.zshrc
if [[ ! -v KITTY_TAB_ID ]] && (( $+commands[kitten] )) then
  # If not already calculated, define KITTY_TAB_ID from our built-in KITTY_WINDOW_ID
  export KITTY_TAB_ID="$(kitten @ ls --match "id:${KITTY_WINDOW_ID}" 2> /dev/null | jq '.[0].tabs[0].id')"
fi

For performance, this uses some zsh-isms to check if the variable hasn’t been defined already & we have the kitten command, in order to keep that shell startup time nice & fast. This requires you to have allow_remote_control yes in your ~/.config/kitty/kitty.conf for the kitten @ ls command to work. A little bit of extra gnarliness to an already-pretty-gnarly setup, but in the end it is just a few lines so whatever (:

Conclusion

These instructions have been tailored for the combination of Kitty + zsh + Neovim, but I’m fairly confident they can be adapted for most other shells/editors running in Kitty too. Not too sure about other terminals though, you might have a harder time scripting the mouse click than in Kitty. I do also recommend checking out jyn’s original post for further inspiration on terminal-agnostic approaches, doing the scripting inside tmux instead perhaps. Though whatever you end up doing, I highly recommend writing some of it yourself, because it is a very good learning experience³.

And that, my friends, is all I have for you today. Take care & stay frosty 👍

Footnotes

Though maybe there’s hope with this latest PR… ↩
Trust me, I tried hard. There is sort of a way to copy environments from existing windows using a kitten @ launch, but it didn’t work for launching commands in new windows for whatever reason. The next-best thing is to recreate the environment that I actually care about from scratch. ↩
Use of time, however? Who’s to say :) ↩

It's Very Hard To Get LLMs To _Not_ Respond To Messages

Thu, 21 Aug 2025 12:00:00 GMT

Being upfront: all these experiments were done by my coworker Jacob Fancher (GitHub), I’m just writing it up because I think it’s cool :)

At Roam, we have an AI assistant called “On-It!” that, among other things, can open a chat between two users, with itself in the middle to close out any tasks discussed:

Unfortunately, this can lead to some pretty stilted conversations; the AI butts in after every message as if it were addressed to it directly, even when the users are pretty clearly just talking to each other. The above example is mostly fine, but for longer conversations it gets really annoying. Turns out, our code had no way for an AI to refuse to answer a message, so we set out to fix that.

Things That Don’t Work

Sentinel string: Telling the AI, to effect, “if you think you shouldn’t respond, reply ‘No comment’”, or other such strings. The AI’s response would be compared against the sentinel, and the message not sent if it matched.
Optional tool call: Telling the AI, “If you think you shouldn’t respond, use this dont_respond tool call instead of sending a message.” The AI already knew about a bunch of other tool calls (for scheduling meetings, creating new tasks, etc.), so if we saw it make this one, we would not send the message.
Enforced tool call: Making it so there are both respond & dont_respond tool calls, and only by using the respond tool call can the AI send messages back to the user.

Unfortunately, none of these approaches worked! No matter what, the AI would do whatever it could to send a message back to a user, even entirely useless ones as before.

Hypothesis For Why

Our AI is built on top of OpenAI’s ChatGPT API, which is LLM-based². Modern LLMs innately expect to complete a message by responding to it, and get very angsty when they can’t respond. This is because the process for turning a “base” model (just trained on raw text) into something like ChatGPT involves a lot of fine-tuning on chat-specific data, enforcing “roles” like user (for messages input by the user) and assistant (for responses created by the AI). OpenAI’s API even encodes these roles directly so you don’t have to manually format the input, it’s all done under the hood for you.

So, the fine-tuning makes an LLM always want to put an assistant message after every user message, no matter what. No amount of prompt engineering can counteract an LLM’s innate drive to respond to a user; even the “Enforced Tool Call” approach is treated as merely a fancy response format. In order get the AI to not respond, we need a different prompting strategy entirely.

The Thing That Does Work

We landed on a 2-step approach: first, we have an “evaluator” prompt with instructions to decide whether or not the AI should respond, outputting into a structured JSON object containing just a boolean should_respond field. Then, we have the normal “responder” prompt that does the rest of the work, crafting a response & making tool calls. Separating out the prompts like this frees each part to think about only what it needs to; the “evaluator” is happy saying true/false as a complete response, and the “responder” only has to think about what it should respond with, which usually overpowers any thought as to if it should respond.

Unfortunately, doing it in two stages like this is a wee bit more expensive, and requires a non-trivial change to our backend, so we decided not to go through with it for now. Still, good to know what we should do whenever this comes up again!

Footnotes

Solmi, Will’s dog, is also featured in another post of mine talking about a Chrome rendering bug (now fixed). Thank you for helping us debug our software, Solmi <3 ↩
Really, what AI isn’t LLM-based nowadays… ↩

CSS Flex, Grid, and Containing Blocks

Wed, 25 Jun 2025 12:00:00 GMT

TL;DR: Changing from display: flex; to display: grid; can break certain designs due to differences in how containing blocks work between them. If you read that and went “what does that even mean??”, keep reading.

The Bug

This toy example is based on a real bug that completely slipped under my radar in code review, so I am highly motivated to write this blog post & understand in meticulous detail exactly what’s going on with it:

<style>
.Container {
  border: solid black 2px;
  padding: 8px;
  contain: strict;
  height: 200px;
  display: flex;
  flex-direction: column;
}
.ChatContainer {
  width: 100%;
  height: 100%;
  display: flex;
  flex-direction: column;
  gap: 8px;
}
.Chats {
  background: lightgray;
  flex: 1 1 200px;
  min-height: 1lh;
  overflow-y: auto;
  display: flex;
  flex-direction: column;
}
.Composer {
  border: dashed grey 1px;
  padding: 4px;
  max-height: 200px;
  overflow-y: auto;
}
</style>
<div class="Container">
  <div class="ChatContainer">
    <div class="Chats">
      <p>There’s a bunch of text here</p><br>
      <p>Just so that it scrolls</p><br>
      <p>Mostly line breaks perhaps</p><br>
      <p>yeah</p><br>
      <p>line breaks</p><br>
      <p>:)</p>
    </div>
    <div class="Composer" contenteditable="true">
      <p>You can type in this box, try it out!</p>
    </div>
  </div>
</div>

There’s a bunch of text here

Just so that it scrolls

Mostly line breaks perhaps

yeah

line breaks

You can type in this box, try it out!

It’s been simplified a lot, but it’s representative of the situation we found ourselves in: some editable text below some static text, where both texts are scrollable, and the amount of editable text changes how much room the static text is given. There’s nothing¹ wrong with it yet, just setting the scene/giving a baseline.

Next, let’s add a sidebar, changing .Container from display: flex; to display: grid; so as to not add another nested div:

<style>
.Container.v2 {
  display: grid;
  grid-template-columns: 1fr 100px;
  gap: 8px;
}
.Sidebar {
  background: orange;
}
</style>

There’s a bunch of text here

Just so that it scrolls

Mostly line breaks perhaps

yeah

line breaks

You can type in this box, try it out!

Uh-oh, big trouble!! There’s some bad overflow going on here, and we can no longer see .Composer! Why is this? We set width: 100%; height: 100%; on .ChatContainer after all, so it should be the exact same size as its parent .Container, right??

Time To Do Some Reading

Wait a minute… width: 100%;… so why can we still see .Sidebar?. Checking the MDN documentation for width:

<percentage>: Defines the width as a percentage of the containing block’s width

OK, and a containing block?

Many developers believe that the containing block of an element is always the content area of its parent, but that isn’t necessarily true.

[…]

If the position property is static, relative, or sticky, the containing block is formed by the edge of the content box of the nearest ancestor element that is either a block container (such as an inline-block, block, or list-item element) or establishes a formatting context (such as a table container, flex container, grid container, or the block container itself).

Well, we have a flex/grid container, so looks like it’s spec-reading time for us. Er, more like spec-Ctrl-F-ing for “containing block”, but hey :P

From the CSS Flexible Box Layout Module Level 1, Paragraph 3:

Flex containers form a containing block for their contents exactly like block containers do.

The containing block for the display: flex; is exactly the area of its content element, a.k.a. “the normal thing”. So, doing width: 100%; height: 100% in a child element was doing what we expected: setting our width/height to exactly that of a parent.

Next, reading from CSS Grid Layout Module Level 2, Paragraph 3.3:

A grid item’s grid area forms the containing block into which it is laid out.

Aha! As we observed before, there’s no longer one big containing block that spans the full width of the div, but rather individual containing blocks created for each grid item. Once we get into layouts that span multiple rows/columns, I can definitely see why this was done: it makes laying out the item within the grid area a lot easier; knowing the width/height of the area is much more important than knowing the width/height of the parent grid.

But for now, I mostly just care about one thing: why the height do that??

Synthesis

Let’s recap what we know:

.ChatContainer has height: 100%;, meaning it will be as tall as its containing block.
Because it’s a grid item, .ChatContainer’s containing block will be its grid area.

Solving This The Correct Way

If not specified, the default values for grid rows are:

.Container.v2 {
  grid-template-rows: none;
  grid-auto-rows: auto;
}

Meaning, all the rows are automatically inserted and have height auto. Meaning, the containing block will try to resolve its height based on the content. Meaning, our height: 100% now does nothing, effectively the same as if it were height: auto itself. Whoops!²

To fix this, we just need to force the grid rows to have a specified height. At the level of grid-template-rows:

<percentage>: Is a non-negative <percentage> value, relative to the block size of the grid container.

So a row of height 100% will make the grid area’s containing block be the correct size, making our height: 100% propagate thru as intended, and the design will be fixed:

<style>
.Container.v3 {
  display: grid;
  grid-template-columns: 1fr 100px;
  grid-template-rows: 100%;
  gap: 8px;
}
</style>

There’s a bunch of text here

Just so that it scrolls

Mostly line breaks perhaps

yeah

line breaks

You can type in this box, try it out!

Solving This The Cool Way

I do love the contain property, so I couldn’t resist showing off a solution that uses it too :)

Thinking a bit, the problem isn’t necessarily that .ChatContainer gets height: auto;, the problem is what happens after, when the content inside forces .ChatContainer to become taller than its parent to contain all that content. But what if there were a way to say “only size yourself based on elements above you, ignoring input from all those below you”? Turns out that’s exactly what contain: size; does!!

<style>
.ChatContainer.v4 {
  width: auto;
  height: auto;
  contain: size;
}
</style>
<div class="Container v2">
  <div class="ChatContainer v4">
    <div class="Chats">
      <p>There’s a bunch of text here</p><br>
      <p>Just so that it scrolls</p><br>
      <p>Mostly line breaks perhaps</p><br>
      <p>yeah</p><br>
      <p>line breaks</p><br>
      <p>:)</p>
    </div>
    <div class="Composer" contenteditable="true">
      <p>You can type in this box, try it out!</p>
    </div>
  </div>
  <div class="Sidebar"></div>
</div>

There’s a bunch of text here

Just so that it scrolls

Mostly line breaks perhaps

yeah

line breaks

You can type in this box, try it out!

This approach is likely a little more brittle than the explicit grid approach, but I think it’s cooler, and that’s what matters 😎

OK that’s it, hope you enjoyed & learned something today ^w^

Footnotes

OK that’s a lie, I would definitely not recommend using anything like this in production, but it works well enough for demonstration purposes here :) ↩
This is often a big problem with height: 100% bounds in general: if you introduce any layer of height: auto indirection in between, they tend to break like this. Switching from display: flex; to display: grid; is just an extreme example of this I guess :P ↩

The AIs Seem To Like One Of My Posts; Why Is That?

Sat, 07 Jun 2025 12:00:00 GMT

One of my past posts, “A Safari aspect-ratio CSS Bug”, has had very steadily growing traffic to it for a few months now. Mostly from Google, as a given, but ChatGPT, Claude, & Perplexity are all in there as referring domains too.

To be clear, I’m pretty pleased with this! I’m happy that people are continuing to find my work useful, and glad that I bothered to write down my solution 9 months ago. But, it’s not like I really do any marketing for this thing? This (LinkedIn wall sorry) doesn’t count. So, it’s all just the algorithms out there deciding my words are tasty & serving it up to people of their own accord. My curiousity cannot be sated. Why?

Hypotheses For Why Algorithms Like It

It’s very short
It’s very structured
It efficiently answers “How Do I Do X?” which may be the exact question the user has typed in

I didn’t really optimize for these things when writing that post, it just sorta ended up that way. But! Those turned out to be pretty good properties to have in a post regardless, even if you’re not an algorithm. To test these hypotheses, I would need to write more blog posts, but alas, i am merely a channel for the energies of the divine muses, & their whims r fickle. anyways

Why I Like It

It’s interactive
It includes not only the what, but a plausible explanation for the why, helping you learn something about browsers/CSS after you finish reading/experimenting

Neither of these are things you get by reading the AI output¹. Which is why I’m glad that people actually clicked thru to the source! I like writing these posts for people, so I hope the state of things will be “peope use AI to find the posts” and not “the AIs read the posts & tell the people what was in the posts”, tho sometimes it seems we’re headed in the latter direction, ah well.

Footnotes

In a perfect world: yes the perfect AI assistant could do both. But we do not live in a perfect world. The common assistants don’t do arbitrary HTML/CSS-injected demos (good!!!), and the summaries will miss the subtleties of what’s going on in this article. Please just read the original it’s so short. ↩

node-webrtc v0.9.0 Released!

Fri, 16 May 2025 12:00:00 GMT

Wow!! I can’t believe it’s been over 1.5 years since I became the maintainer of node-webrtc. Feels a lot longer than that somehow :) Finally at a point where it feels like “my” codebase that I know every square inch of.

This release has been a long time coming, and contains a few major improvements over the previous v0.8.0 release:

No More Compiler Warnings

It had always bothered me when Clang spat out a bunch of compiler warnings when building the project. It was a lot of work over many other commits, but I recorded this commit as the last one in the series. Now, when you build node-webrtc, you don’t get any compiler warnings!¹ Speaking of building:

Unified Build Environments With Nix

Previously, I had to do some pretty bespoke stuff on each machine to set up a proper build environment for node-webrtc. Given the number of machines I had, this sucked. This past year, however, I’ve been slowly becoming more familiar with Nix, to the point where I could create a mildly complex build envionment definition to share between my Linux and MacOS machines. I’m most proud of the fact that it’s completely gotten rid of any dependency on the system XCode on MacOS, managing that with Nix is so much nicer holy wow.

Cross-compilation is still a bit tricky, even with this setup. Here’s what I can do so far:

MacOS arm64 ➡️ MacOS x64
MacOS x64 ➡️ ️MacOS arm64
Linux x64 ➡️ Linux arm64

Yeah, not very impressive I know sorry :P My stretch goal is to be able to target anything in {MacOS x64, Linux x64, Linux arm64, Windows x64} all from MacOS arm64, because my M4 Macbook Pro chews through compilation tasks like no one’s business. That’s where I did my main development for:

Reducing Memory Errors With Rust-Inspired Lifetime Management

Garbage collection in C++ sucks, man. Or more specifically, interfacing with Node’s garbage collection through their API sucks. There’s a lot of stuff to uphold, and because the API was originally written for C, not C++, a lot of stuff really just wants to be a pointer. I do not know why I had to write my own RefPtr class, but I did, and I know too much about C++ constructors because of it.

That’s really the thing that was missing from the codebase: a concept of ownership, à la Rust. Not properly tracking that would cause destructors to not fire at all (memory leak/hang), or fire too early (segfault), both of which would be caught by some test cases (massive shoutouts to the previous maintainer Mark for writing these tests in the first place!!). My solution for this was to introduce an OwnedWrap class, clarifying the existing semantics of the Wrap class used to create weak references to cached objects.

Future Plans?

Really, other than a few tricky GC errors & the usual fighting with the build system, I didn’t have too much trouble updating node-webrtc to M98 and Node.JS 22. My hope is that, with a lot of annoyances cleaned up and a lot of knowledge gained, I can finally start making progress on catching up to the latest libwebrtc version. I know there will be some link errors along the way, and I still don’t have a lot of free time to dedicate to this project, but damnit it feels good to finally get out a release. Here’s hoping the next one doesn’t also take a whole year!

Footnotes

GCC still gives compiler warnings, unfortunately. I think they’re wrong, though, so shouldn’t be too much trouble to turn them off if I need someday. ↩

More Photography Updates (and RSS)

Tue, 25 Feb 2025 12:00:00 GMT

Hello again! I promise this will be the last time I announce a photography collection here: 2024: Empire State Building is my latest set as mentioned in my previous post.

Also no, “last time” is not because I’m quitting taking photos or anything like that. It’s just because I’ve finally set up an RSS feed for my photography! Previously, I was worried people might miss when I released those unless I updated here, but now that I have RSS set up for it, any new photography collections I release will show up under the photography feed.

Speaking of RSS, I also did a little digging and found my surfacing of it in my page’s metadata was sorely lacking. I was only advertising it on my main page, not on any of the blog posts themselves, so people wanting to quickly add my site into their feed reader may have been out-of-luck. No longer! I’ve been using RSS a lot more myself recently, so I’m more attuned to things like this now. Most peoples’ sites actually have pretty good RSS support turns out!! You should use RSS, it is excellent, also subscribe to my blog with it :) kthxbye

NYC Photography

Sat, 11 Jan 2025 12:00:00 GMT

Hi! I just released the first of my 2024 NYC photo collections (yes there is another coming). More details on the page itself, check it out!

All Javascript Keyboard Shortcut Libraries Are Broken

Fri, 10 Jan 2025 12:00:00 GMT

Either subtly (by using key), or not-so-subtly (by using code). There is no way to fix the more subtle variant, and the only solution is to Give Up (on supporting a large subset of keyboard shortcuts).

So for work, I was tasked with making it so users could rebind the keyboard shortcuts used in our app. Not too complicated, I just needed to consolidate all the shortcuts into one place so I can split apart the keyboard shortcut -> action and action -> code that gets run maps, create a definition for a keyboard shortcut, let users to override the first map in settings, and bam. Don’t even need to use any external libraries, the existing code shows the browser-provided information is more than sufficient! Except wait a minute, I feel like I’m forgetting something…

What Is A Keyboard Shortcut?

Ah. Those pesky definitional questions. Others may have different, more complex definitions to account for different, more complex scenarios, but for our purposes in our app, a keyboard shortcut is simple:

A keyboard shortcut is one or more modifier keys being pressed when exactly one non-modifier key gets pressed.

And while I’m at it, I should probably also define what a modifier key is:

A modifier key is any one of the following keys:

Shift, aka ⇧

Control, aka ⌃

Alt, aka Option, aka ⌥

Meta, aka Super, aka Windows Key, aka Command, aka ⌘

I’m sure we’ll have to revisit this for Vim mode, whenever a critical number of nerds starts using our platform, but good enough for now.

We choose this definition because it exactly lines up with the fields in KeyboardEvent: altKey, ctrlKey, metaKey, and shiftKey. And then one other field for the non-modifier key. One other field… Wait, there’s more than one??

More Than One Way To Describe A Key

Ignoring all the deprecated ways, there are exactly two fields that describing which non-modifier key was pressed: code (deprecated: keyCode, which) and key (deprecated: keyIdentifier). The linked MDN documentation does a pretty good job at explaining the differences, but i also like to think i am also good at explaining things, so here I go:

code is for when you want to write a game, where the position of the keys relative to each other matters.
key is for literally everything else. Why? Because otherwise, users on alternative or international layouts would read the incorrect thing when using code.

Great, easy! I’m sure that all the popular libraries will be aware of this and all use key, then :)

Oh No All The Libraries Use `code`

Let’s stroll down the top “shortcut” packages on NPM, sorted by monthly downloads, shall we?

react-hotkeys-hook: uses both code and key in a way that will lead to shortcuts triggering more often than they should.
mousetrap: uses which which is basically just code but more deprecated.
hotkeys-js: uses keyCode which is also basically code but more deprecated, but at least it’s less deprecated than which.
tinykeys: supports both code and key but defaults to key; the user has to explicitly specify a string like KeyA if they want to match on code.
- Wow, it’s our first good one! I honestly really like this choice.
combokeys: uses which, with a fallback to keyCode.
react-hot-keys: just uses hotkeys-js internally, which used keyCode.
react-shortcuts: just uses combokeys internally, which used which.
@discourse/itsatrap: is a fork of mousetrap and still uses which.
v-hotkey: uses keyCode
ng-keyboard-shortcuts: uses which with a fallback to keyCode and is also the one of the most horrifying Javascript codebases I’ve seen.
@ngneat/hotkeys: uses key. Not quite as horrifying as the previous, but comes close. Angular projects man…
shortcuts: Somehow has more downloads than its underlying library, shosho, which uses code with a fallback to key.
@sofie-automation/sorensen: uses code, but with an interesting twist we’ll get to later.
use-keyboard-shortcuts: uses both code and key in a way that makes me think it will break under specific circumstances on alternative layouts.

A list is a bit hard to parse, let’s look at it in a table too:

Package	Uses `which`	Uses `keyCode`	Uses `code`	Uses `key`
react-hotkeys-hook	❌	❌	✅ (bad)
mousetrap	✅	❌	❌	❌
hotkeys-js	❌	✅	❌	❌
tinykeys	❌	❌	✅ (good)
combokeys	✅	❌	❌	❌
react-hot-keys	❌	✅	❌	❌
react-shortcuts	✅	❌	❌	❌
@discourse/itsatrap	✅	❌	❌	❌
v-hotkey	❌	✅	❌	❌
ng-keyboard-shortcuts	✅	✅	❌	❌
@ngneat/hotkeys	❌	❌	❌	✅
shortcuts	❌	❌	✅ (bad)
@sofie-automation/sorensen	❌	❌	✅ (good?)	❌
use-keyboard-shortcuts	❌	❌	✅ (bad)

Holy cow. I knew it was bad but I had no idea it was this dire. Now obviously, “most downloaded packages under a keyword” is not necessarily representative of the ecosystem as a whole. lexical for example, which is not primarily a keyboard shortcut library but includes some anyways, uses key. Still, it boggles the mind how bad the popular purpose-built ones tend to be.

But Wait There’s More

Unfortunately, “just use key” isn’t all sunshine & roses like I may have led you to believe. There’s a reason library authors might shy away from it, even though it’s been supported in all major browsers for almost 8 years (or 5, if you include Edge), and choose to use code or its variants instead: the Shift modifier modifies the value produced by key 🙀

Wait, that’s it? Like sure, maybe upper/lowercase doesn’t always round-trip, and doing Shift+2 will be different from Shift+Meta+2, and all this might change based on the user locale, and oh no yeah that’s a big problem isn’t it. By using key, you might be able to support DVORAK a bit better, but German keyboards will have Shift+2 give you " instead of @ like you’d get on a US layout. As far as I can tell, only the roman alphabet (A-Z) is safe.

Even More Actually

The event that sparked this post was honestly something a bit more silly: It’s the fact that, on MacOS, the Option (⌥) key also modifies keys, even on US QWERTY! Alt+c? Surely you mean Alt+ç 🙂. That is what made me switch my initial prototype to use code instead of key, which led me to discover this incosistency, which led me down the rabbithole & out the other side with this big corkboard full of red string.

What The Good Libraries Have To Do

When searching NPM for “shortcut” libraries sorted by most recently updated, I came across keyboard-i18n, which has an interesting solution to this problem: it uses the experimental Keyboard API (only supported on Chrome) to read the KeyboardLayoutMap, which maps code -> key with no modifiers. From this, you can build a reverse mapping key -> code, which lets you specify your keyboard shortcuts using key, but read them using code, getting you the best of both worlds! @sofie-automation/sorensen also uses this approach afaict.

This is not quite the end of it, however; there are certain keys we might want to bind, like Ctrl+[, that won’t be present on all layouts. keyboard-i18n has some useful maps for certain keys like this that will work on international layouts without them, by mapping to non-roman characters in a similar location.

Unfortunately, this comes with a downside: the Keyboard API is only supported on Chrome. Which maybe isn’t a problem if the rest of your app only supports Chrome! But is still a problem if you want to support Safari¹, or that other browser everyone forgets about (Firefox).

What We Cross-Browser Plebians Have To Do

So, if we want to support all major browsers, and all major keyboard layouts, our keyboard shortcut definitions must:

Use key
Only use roman alphabet characters (A-Z) and arabic numbers (0-9)
Use .toLowerCase() or .toUpperCase() to normalize case when checking what character was pressed
Only allow Shift as a modifier with alphabet characters
Never allow Alt as a modifier

As far as I can tell, there is no general-purpose shortcut library that gets this right. And maybe it’s the state of web search, but I can’t find anyone talking about this either. This lexical PR is about the closest I could get. I’ll probably just copy them, enforcing these rules within my own proprietary framework.

Everything is broken, and everything is fine 🔥

Footnotes

Safari is especially problematic, because it has a very poor track record of adopting new web features in a timely manner. anyways ↩

Rochester Photography

Sat, 07 Dec 2024 12:00:00 GMT

Figured I’d keep my photography-posting “streak” up, if you can call it that :P

This time I have pictures from a trip I took to Rochester, on my own time, to visit a friend who’s at RIT there. Very fun, got a lot more “personal” pics than I did “artsy” ones. Still, got a few at least! Most happy with the last one, think I have a good thru-line for them as well. Check it out! 2024-10: Rochester

Montreal Photography

Tue, 03 Dec 2024 12:00:00 GMT

Wow I’ve been slacking on this one!! I have like 2? other albums I need to curate & get up here as well. Hoping to get those out of the way so I can focus all my photography attention on my New York City project (sitll very WIP). Then again I have so many other plates to juggle that idk if I’ll get a proper “2024 NYC” out by the end of this year…

Oh well! At least we have 2024-09: Montreal to look at in the meantime :) I’m pretty happy with these, had to cut quite a lot of cruft to get them tho. Def leaning into being a Digital Photographer with that approach… good thing I love digital photography!!

Tracking Down a "Heisenbug" in Chrome

Tue, 19 Nov 2024 12:00:00 GMT

You know the drill. Go on hiatus because life is busy, catch COVID again even tho I wear a mask, and uncover yet another browser bug while still somewhat sick (getting better tho!).

The bug I have for you today is very special to me. We’ve been encountering it on-and-off at Roam for a while now, since January I think? Customers would report “hey something happened and now my entire screen is shifted up”, and we would say “can we see?”, and they would send screenshots showing sure enough, their entire screen was shifted up. It even happened to me once, though it was less dramatic a shift than some:

Names/pfps redacted for privacy. Note the top of the window being cut off, and the bottom of the map having a misaligned border

Invariably, the reproduction (“someone sends me a chat message while I have Roam open on my external monitor”, “it happens when my computer wakes up from sleep”) wouldn’t work every time for them, and would never work on our computers, so we’d be stumped, and that would be the end of it. It drove me crazy. Not having access to a consistent reproduction meant I was up a creek with no way to solve it.

Once, months later mind you, we got lucky that one customer was willing to screenshare during their bug encounter, and I got to poke around second-hand in devtools. What I found was, uh,

The Entire `<body>` Element Was Shifted Up

Well that’s just impossible! Ok well clearly it is possible, given we’re seeing it live, but our mess of like 10 container divs shouldn’t leave any room for the body to scroll at all. And it’s not the <body> that’s scrolling, it’s something else entirely!

Looking at the situation more, I wasn’t entirely convinced that our mess of 10 container divs always behaved correctly. So I rolled up my sleeves and worked to reduce that number to 3 container divs. Everything worked fine, and to top it off, we got no more reports from them or any other customer about their screen shifting! The end :) What a happy tale about reducing technical debt paying off :)

Except, that’s not the end. That can’t be the end. No way did I witness something impossible, that only happens in extremely specific circumstances, only on customer machines, for it to be our fault. My nose is attuned to the smell of browser bugs, and oh brother does it stink.

Shifting 2 - Electric Boogaloo

There’s a longstanding issue with the drag-n-drop library we use, @hello-pangea/dnd, not working in any Electron window besides the main window. So, my coworker Klas Leino recently put in a herculean amout of effort to switch us over to dnd-kit. Natually, changing drag-n-drop libraries involves lots of reshuffling of divs, with no guarantee that behavior is unaffected. So we did what we normally do: feature-flag it for internal use only, and test it by dog-fooding our own software daily.

And it’s a good thing we do this too, because sure enough our bug-testing processes caught a bug: another coworker, Theo Ouzhinski, noticed that whenever he clicked an unread message on his external monitor, his screen would shift up.

HM THAT SOUNDS FAMILIAR

It was a bit different than before, though. Instead of the entire <body> being shifted, it was just the part inside the floating inbox. And instead of being finicky to trigger, it was very reliable: get a new unread message, click on it, boom. And, the most crucial part of all, it reproduced on my machine:

Debugging Time ^w^

The next few hours were a flurry of hypotheses and seeing where it reproduced and where it didn’t. In the end, I was left with some required conditions for the bug to reproduce in our app:

Reproduces across platforms (Mac, Windows), on many different types of displays
Depends on the display; e.x. on my Mac’s built-in display, it won’t trigger, but on an external display plugged into the same Mac, it will.
Reproduces only in Chrome/Electron, not Safari or Firefox
Must click an unread channel; normal navigation is unaffected.

A CSS Footgun?

After more digging & some wild hunches, I found out what was actually going on: there was an outer container with overflow: hidden and a scrollHeight greater than its own height. When an unread channel is clicked, we load in the messages, then call element.scrollIntoView() on the earliest unread (incidentally, this is why it only happened for unread messages; normal navigation didn’t make this call). When that earliest unread is “the latest message”, this can cause all containers with room to scroll to scroll to their bottoms, including the overflow: hidden one. I did not expect this since the overflow: hidden container should not have scroll on it.

Initially, I thought it was impossible for overflow: hidden elements to scroll at all. However, MDN docs for scrollHeight falsify this hypothesis: elements with overflow: hidden can have a larger scrollHeight and are expected to be able to scroll.

<div class="outer" style="height: 400px; overflow: hidden;">
  <div class="inner" style="height: 400px; overflow: hidden;">
    <div class="content" style="height: 600px; position: relative;">
      <span id="top">Top</span>
      <span id="bottom" style="position: absolute; bottom: 0;">Bottom</span>
    </div>
  </div>
</div>

With this structure, .inner will be the one to scroll when we call bottom.scrollIntoView(). And .outer will have a scrollHeight of 400, because .inner doesn’t overflow it at all.

A CSS Bug

But apparently, under some Sufficiently Complex CSS Conditions, .outer does have a scrollHeight much greater than .inner, even when both have overflow: hidden. Based on reading the spec, that’s a bug: outer.scrollHeight should not be larger than the box of its only descendant. With that in mind, I’d like to present my hand-minimized reproduction of these Sufficiently Complext CSS Conditions¹, whose <div> ids I’ll be referencing in the coming explanation:

https://www.hazelduvall.dev/static/2024-11-19/scrollHeightBug/index.html

Source code:

<!doctype html>
<html>
  <head>
    <title>scrollHeight bug</title>
    <style>
      * {
        box-sizing: border-box;
      }
      body {
        min-height: 100vh;
        text-rendering: optimizespeed;
        margin: 0px;
      }
      #root {
        background-color: white;
        color: black;
      }
      #outer {
        will-change: transform;
        overflow: hidden;
      }
      #inner {
        overflow: hidden;
      }
      #padding {
        margin: 1px;
      }
      #scroll {
        height: 600px;
        position: relative;
        overflow: auto;
      }
      #spacer {
        position: relative;
        height: 1500px;
        background-color: lightblue;
      }
      #pixel {
        position: fixed;
        width: 1px;
        height: 1px;
        margin: -1px;
        border: 0px;
        padding: 0px;
        overflow: hidden;
      }
      #output {
        position: fixed;
        bottom: 0;
        right: 0;
      }
    </style>
  </head>
  <body>
    <div id="root">
      <div id="outer">
        <div id="inner">
          <div id="padding">
            <div id="scroll">
              <div id="spacer">
                <span style="position: absolute; top: 0">Top</span>
                <span style="position: absolute; bottom: 0">Bottom</span>
              </div>
              <div id="pixel"></div>
            </div>
          </div>
        </div>
      </div>
      <div id="output"></div>
    </div>
    <script>
      const recalc = () => {
        const outer = document.getElementById("outer");
        const inner = document.getElementById("inner");
        const output = document.getElementById("output");
        output.innerHTML = `
outer.scrollHeight=${outer.scrollHeight} outer.clientHeight=${outer.clientHeight}<br>
inner.scrollHeight=${inner.scrollHeight} inner.clientHeight=${inner.clientHeight}
`;
        setTimeout(recalc, 1000);
      };
      setTimeout(recalc, 0);
    </script>
  </body>
</html>

If the bug occurs, outer.scrollHeight will be 1502, not matching any of the other values displayed, which are all 602. Because I read the spec, I am confident in calling this a bug, even though both Chrome and Firefox seem to have the same behavior²:

Chrome	Firefox	Safari

I am also very confident in calling it a bug because removing the will-change: transform; property from #outer causes the bug to go away on both Chrome and Firefox³. Additionally, on Chrome only, doing any of the following will also cause the bug to go away:

Setting left: 1px; on #pixel (it already had this as a computed value from the default left: auto;)
Removing margin: 1px; from #padding (??? in Firefox it doesn’t seem to matter if this div is present)

I highly encourage you to play around with this example in DevTools to get a feel for just how weird it is. That’s the thing with browser bugs: you thought you were standing atop a rock-solid platform, only to find out it’s made of crumbling sand.

Resolution

The best fix I’ve found for this bug is to just add top: 0; to #pixel. This works on both Chrome and Firefox, I think because setting an explicit top property moves the #pixel element firmly inside the container it’s supposed to be in, causing no more weird scrollHeight calculations to happen. I’ll patch dnd-kit locally, open a PR upstream, and send some P4 bug reports to Chrome and Firefox that will never get fixed. But hey, maybe I’ll be proven wrong! After all, this other bug I independently found eventually got fixed! In any case, just knowing enough about the bug to fix it w/o waiting for Chrome is a huge weight off my shoulders. Thanks for reading, until next time!

Footnotes

One unexpected benefit of making a minimal repro is that the bug no longer depends on what monitor you use, only what browser. Lucky, I guess! ↩
All of these screenshots are from my work MacBook, running MacOS 15.1 & the latest versions of every browser. The Chrome and FireFox examples repro on my Windows 11 laptop too. ↩
And I know it’s not an interaction with stacking contexts, because replacing that with position: relative; z-index: 0; or will-change: opacity; (which will also create stacking contexts) makes the bug go away. Something about the transform optimization path is making things go awry. ↩

I Made A Webring

Sun, 22 Sep 2024 12:00:00 GMT

Recently, both I and a friend of mine (James) have dove back into making our blogs better. I don’t have a whole lot of graphical changes to show for mine, so his is much more impressive. When he showed it off in the Discord we share with the rest of our CMU friends, it garnered a lot of attention, and one thing led to another, and suddenly the talk about “wouldn’t it be neat if we were all in a webring together?” turned into me having a domain name in my cart & plans to write the darn thing from scratch the next morning.

https://cmuwebr.ing/ is the result. I’m pretty proud of it; took 2:30hr from “ok let’s bang this thing out” to “website is live & ready to have sites added to it”. I read some other webring implementations before trying this, but chose not to use any because it really is too darn simple to not just write your own that you fully control, imo. There is a file https://cmuwebr.ing/static/sites.json that the server (written in Python using Sanic) reads, and each site on the ring indexes into it w/ their ID, and the server takes care of redirecting the prev/next links. I also plan on writing a Javascript snippet that will parse said JSON & replace the links on the page so redirects don’t have to go thru my server, but that’s for later.

As the founder of the webring, of course I’m already on it! You can see it at the bottom of every page. I think this is a very good use of the footer space; I do want to encourage more “low-tech” ways of socializing and discovery on the internet. RSS feeds, webrings, emails, it’s like I’ve travelled back to the 90s!! Not to be all RETVRN (HTML is way better than it was back then) but yeah I enjoy this a lot more.

If you’re a CMU friend & want to join the webring, view the instructions on GitHub. Or send me an email! If you’re not a robot I’m sure you can find where…

Better Static File Handling

Sat, 21 Sep 2024 12:00:00 GMT

For a while now, I’ve had a photography page on my blog. When I made it, I wasn’t thinking too hard and just followed along with the Astro Image guide which said “put ur images somewhere in src/ and import them”. One thing led to another, and I ended up with ~800MiB of data in my git repository. :P

To be clear, git can handle this much data w/o much stress. GitHub gets a little bit sad on the upload, but hey, at least that only happens once, right? But consider: the GitHub Action to deploy my site needs to download the repo in order to run. So every run, it was downloading ~800MiB, only to restore what it needed from a cache anyways. It got so bad, this repo download step was half the build time when caches were warm.

Needless to say, I can do better. So I did! All large static assets are now served from www.hazelduvall.dev/static, and I’ve rebased by git history to remove all the large objects. Astro is quite smart; it supports loading images from remote URLs while still putting optimized versions in its cache, w/ TTLs reflecting that returned by the Cache-Control header.

I think this, plus caching node_modules, will be enough to put me under 30s for full site rebuilds. Which is pretty solid! That’s as far as I can realistically push it w/ Astro, because it does need to rebuild the entire site every time. Thinking about building my own CMS, but that’s a project for another day…

A Safari aspect-ratio CSS Bug

Sat, 14 Sep 2024 12:00:00 GMT

Wow it’s been a hot minute, huh? I did catch COVID again recently, but that was only for a week, and really the reason I’ve not been posting is because I feel broadcasting all my hot takes about how terrible Go is (from more experience working on a Go service that I’ve had to redo many of the abstractions for) isn’t really productive. But you know what is productive? That’s right, more CSS bugs! And this time we have something special: a Safari bug! Sort of surprising that I haven’t shown off more of these, Safari is a pretty busted browser when you get right down to it. There’s another bug (related to this one) with an inconsistent repro that I won’t spoil just yet, so for now here’s just a taste for how weird it can get:

Problem Statement

Consider the following problem: you have an arbitrarily-sized parent container, and you want to make a child container with a fixed aspect ratio that fits flush with the bounds of the parent¹. Seems simple enough, no?

Solution 1: Simple Boundaries

<style>
.container {
  background-color: #888;
}
.take1.container {
  display: flex;
  align-items: center;
}
.layer1 {
  background-color: var(--color-orange);
}
.take1 .layer1 {
  width: 100%;
  max-height: 100%;
  aspect-ratio: 1;
  display: flex;
  justify-content: center;
}
.layer2 {
  background-color: var(--color-blue);
}
.take1 .layer2 {
  height: 100%;
  aspect-ratio: 1;
}
</style>
<div><label for="w1">Width:</label><input id="w1" type="range" min="120" max="600" value="200"><label for="h1">Height:</label><input id="h1" type="range" min="120" max="600" value="200"></div>
<div id="c1" class="container take1">
  <div class="layer1">
    <div class="layer2"></div>
  </div>
</div>
<script>
const register = (inputId, divId, what) => {
  const input = document.querySelector(inputId);
  const div = document.querySelector(divId);
  const set = () => {
    div.style[what] = input.value + "px";
  }
  set();
  input.addEventListener("change", set);
}
register("#w1", "#c1", "width");
register("#h1", "#c1", "height");
</script>

Width:Height:

Simple! Play around with the sliders to get a feel for how this works; there’s an blue box inside an orange box inside a gray parent container. How this works is:

Layer 1 (orange) sets:
1. width: 100% (be as wide as the parent)
2. aspect-ratio: 1 (make the implicit height: auto match the width)
3. max-height: 100% (clip height to that of its parent)

That gives us an orange box that can be at most as tall as it is wide. Then:

Layer 2 (blue) sets:
1. height: 100% (be as tall as the parent)
2. aspect-ratio: 1 (make the implicit width: auto match the height)

Because Layer 1 has height <= width, Layer 2 will be a perfect square, every time!

…except in Safari

Safari’s Problem

From what I understand, the problem happens at Layer 1. Instead of doing the procedure outlined above, it instead seems to do something like this:

Layer 1 (orange) sets:
1. width: 100% (be as wide as parent)
2. max-height: 100% (clip height: auto to that of parent)
3. aspect-ratio: 1 (make height match width)

Step 2 doesn’t do anything, because the default height: auto for a container is 0. It works if you swap around the width/height constraints between the two layers, (because width: auto is 100% so does get clipped properly), but this feels like a hack, and besides there’s a different solution that only uses one layer.

Solution 2: Good ol’ `margin: auto`

<style>
.take2.container {
  position: relative;
}
.take2 .layer1 {
  position: absolute;
  inset: 0;
  margin: auto;
  max-height: 100%;
  aspect-ratio: 1;
}
</style>
<div><label for="w2">Width:</label><input id="w2" type="range" min="120" max="600" value="200"><label for="h2">Height:</label><input id="h2" type="range" min="120" max="600" value="200"></div>
<div id="c2" class="container take2">
  <div class="layer1"></div>
</div>
<script>
register("#w2", "#c2", "width");
register("#h2", "#c2", "height");
</script>

Width:Height:

To me, it’s a bit more magical how this works. This is what I think is happening:

position: absolute; inset: 0 inside a position: relative sets the child container’s width and height to be that of the parent
- but not the width and height properties directly! those are both still auto so it can participate in aspect-ratio resolution.
aspect-ratio: 1 is applied, defaulting to setting height based on width
max-height: 100% is applied, clipping height
aspect-ratio: 1 is applied again, this time setting width based on height, because width is still auto

This explanation checks out with all my previous ones: some sort of “auto-resolution loop” happens, and we never saw it because earlier, one of the width/height was non-auto. It also now makes sense why this works in all browsers, because the order of aspect-ratio/max-height no longer matters; aspect-ratio gets applied twice anyways.

What margin: auto does then, is center the div in both axes (and in Safari’s case, make it so the vertical margin can no longer be negative, because of course Safari is still special). You can pin the container to a given side by setting margin-<side>: 0 afterwards.

Phew!! Box sizing is hard, man.

Footnotes

There’s an even harder variant of this problem that I had for https://ro.am/howard/: you have an arbitrary-sized parent container, and you want the child to overflow the parent in only one dimension with a fixed aspect ratio, but not just use object-fit: cover on an image because there are elements that need to be positioned relative to that image. I won’t go into it here, you can read that page’s source if you’d like tho :) ↩

A Third Browser Bug...

Fri, 12 Jul 2024 12:00:00 GMT

So at work, I had the opportunity to write a small tool that dumps the current state of the document into a single static HTML file, making no network requests. For the most part, it turned out to be a lot simpler than I expected! DOM APIs are really good nowadays (all hail querySelectorAll). It should have taken me like 3 hours tops, from starting up the editor to pushing the PR, but instead it took all day, because of one very weird bug.

Virtual StyleSheets

One thing I learned poking around in our app for a bit is that, not all <style> elements have to be present in the document! That’s right, you can define CSS rules entirely in JavaScript, never having their state show up in the DOM at all. You do this thru the CSSStyleSheet and document.adoptedStyleSheets APIs, supported on all major browsers:

const styleSheet = new CSSStyleSheet();
styleSheet.insertRule("* { color: red !important; }");
document.adoptedStyleSheets.push(styleSheet);

This example code will turn all text on the page red. Try it out in your browser console now!!

In any case, I knew about this restriction beforehand, and was prepared. The document.styleSheets objects can also be modified in-place, so really the only way to properly dump everything was to run something like this:

// Create a new empty document:
// <html><head><style></style></head></html>
const outputDocument = new Document();
const html = outputDocument.createElement("html");
outputDocument.appendChild(html);
const head = outputDocument.createElement("head");
html.appendChild(head);
const style = outputDocument.createElement("style");
head.appendChild(style);
// Add all rules to the style element
const rules = [...document.styleSheets, ...document.adoptedStyleSheets].flatMap(
  (styleSheet) => [...styleSheet.cssRules].map((cssRule) => cssRule.cssText),
);
style.sheet.replaceSync(rules.join("\n"));

Virtual Documents

As it turns out, this doesn’t work! Why?? As it turns out, browsers are very finicky about element state; many properties are defined only when they are actively being rendered, which this virtually-created Document is not. This extends for everything created in the “context”¹ of that virtual Document too, so style.sheet will be undefined in the above example.

No problem, though! Instead of modifying thru style.sheet.replaceSync, we can do what MDN recommends on another page somewhere else, and use innerHTML:

style.innerHTML = rules.join("\n");

and boom, everything is working! Except wait a minute, certain styles got applied right, but other styles are wonky. Diffing between the two, we see that no direct child selectors (that is, ones like .a > .b) are working. And looking in the rendered source, we find the culprit: they’re being written as .a > .b. Huh??

`innerHTML`

Some background on innerHTML: it has some simple character escapes built-in, from some sort of “fragment parsing algorithm” that detects if something should be text or should be an HTML element. This makes code like this “just work”:

const test = (doc, elemType, text) => {
  const elem = doc.createElement(elemType);
  elem.innerHTML = text;
  console.log(elem.innerHTML);
};

test(document, "p", "2 + 2 < 5");
// prints "2 + 2 &lt; 5"

Now, because CSS wants unescaped symbols, the usual escaping rules seemingly do not apply to it:

test(document, "style", "p > h2 { color: blue; }");
// prints "p > h2 { color: blue; }"

…except, of course, when you have a virtual Document

test(new Document(), "style", "p > h2 { color: blue; }");
// prints "p &gt; h2 { color: blue; }"

Because of course! Why would that work? Clearly I am extremely silly for expecting otherwise.

I spent a lot of time spinning my wheels on this one, probably because I had skipped lunch 2 hours ago by the time I got to this point. Having a lunch and coming back, the solution was easy: just do

const output = `<html><head><style>${rules.join("\n")}</style></head>${document.body.outerHTML}</html>`;

I didn’t need to keep everything as elements in the first place! All I cared about was the output text. A bit more magic to remove all scripts and make all images load from data URLs², and bam it was working flawlessly. In the end I still had a lot of fun with this, hope you learned something too!

Footnotes

I’m probably misusing this term here. Oh well you get the point! ↩
Secret ;) ↩

More Photography!

Tue, 11 Jun 2024 12:00:00 GMT

I haven’t been active on this blog much, but I have been very active at work! One consequence of that is I’ve gone on a few work trips, and the inspiration from those trips compelled me to start taking artsy photos again, so here I am! My photography page has been updated with these trips, to Chicago and London respectively. I’m fairly proud of myself for getting back into it at all, and with a borrowed camera from my dad (thanks, Dad!) I can hopefully keep this hobby up even longer.

In other neat news, I have started to learn Japanese! I’ve memorized the basic hiragana and katakana alphabets and am starting to learn Kanji thru WaniKani. It’s pretty slow going, but that is how I learn best I think, thru routine-building. Still very early so we’ll see how this goes.

In less good news, I have contracted COVID-19 for a second time :(((( This time around was hopefully not from any family members (that would be bad!) but from some other transit/event. Still recovering, but honestly working on this blog is low-effort enough and I am bored & needed something to do so this is good :P

That’s all for now, hopefully I will have another update before 3 months have passed again!

You Won't Believe That I Found Another Chrome Rendering Bug

Sat, 23 Mar 2024 12:00:00 GMT

EDIT: As of 2024-10-15, when Chrome 130 was released, this bug is now fixed! Yahoo!!

So in one of my previous posts, I outlined how Chrome had a rendering bug where it did not properly mask anti-aliasing pixels under certain complex CSS mask conditions. And wouldn’t you know it, right as I publish an update to that post, I find yet another Chrome rendering bug at work.

This story starts a couple months ago, when I’m working on a refresh of some of the colors and layout of our app’s main page to match a new design. The designer comes up to me and says “ayo check out this weird bug I’m having”, and shows me.

On their screen, many of the colors look darker than normal. And suddenly, when they hovers over a specific UI element, all the colors snap back to normal. As soon as they stops hovering, the colors go back to being weird again. I have never seen this happen before.

I take a look at their browser state in devtools, and discover something distrubing: all the CSS is correct. Everything is exactly the same as on my machine, and yet the results are different. I’ve done nothing wrong except fail to understand what the heck is going on here.

Narrowing Down A Culprit

As for “why do the colors snap back to normal when a div is hovered?”, that part had a simple yet outlandish explanation: hovering a certain div creates a blur effect, the addition of which must’ve done something to the way the rest of the page was rendered. As for “why are the colors wrong in the first place?”, that part was harder to figure out.

I had put it on the deep backlog at the time, because “shipping the thing” was more important than “solving a bug that seems to hardly ever occur in practice”. But just recently, enough people independently had the bug that we finally had some leads.

One of my coworkers noticed this bug only occured for them when another specific coworker was signed in with their test account. Wait, What? They also noticed that, when the test account’s profile picture showed up on screen, their display backlight lit up more, which usually occurs when trying to display an HDR image.

Excited for a reproduction, I lauched the app on my own machine. The bug didn’t reproduce. Then, I recalled that the monitor I usually test on might not be HDR-compatible, so I dragged it to another monitor that was: my Macbook’s built-in screen. And lo and behold, I had a reproduction. I opened up devtools, unloaded the image thought to be the culprit, and sure enough, the bug went away.

So! Something about HDR images, and whatever div setup we have going on is causing trouble.

Why Is That??

To answer this question completely, I would have to dive into Chromium internals. Unfortunately, I am not that good at diving into Chromium internals, so instead I just found a couple bug reports that might give some insight:

HDR Was A Known Problem

This bug was the first relevant result when I searched for something along the lines of “blur interacts weirdly with HDR MacOS Chrome??”. Back in Chrome M80, they had to disable HDR support for MacOS because they “fall out of HDR mode if there are extra render passes”, like blur, “which looks bad.”

Despite the age of this bug, this is almost the exact result I was seeing! Adding a blur anywhere did cause other stuff to fall out of a (presumably bad) HDR mode. And because the fix for this bug was to just disable HDR, it could be likely the underlying issue was never addressed fully. Feeling confident I was on the right track, I kept looking…

HDR Still Is A Problem

This next bug is much closer to what I was seeing: an HDR image causing a background with opacity to render incorrectly! The reproduction isn’t exactly the same as what we have going on in our app, but hey, close enough.

Below, I’ve demonstrated the reproduction from the bug report: an HDR image with position: absolute outside a parent div with an opacity less than 1:

If you’re in most browsers, you should only see the border color of the square change when the checkbox is checked. If, however, your are:

using Chrome
on MacOS
on an XDR-enabled display

Then the above demonstration should make it so that checking the checkbox also changes the background color of the lower square. This should never happen, because the square is simply background-color: white; opacity: 0.5; over an already-white background, so it should be white always (like in most browsers). The reason the background color changes is because the checkbox has a backdrop-filter: blur property only when it is checked, and as mentioned previously, enabling this blur disables the buggy HDR.

If you don’t believe me, observe this screen recording I took of the bug. Yes, it does show up in screen capture (thank goodness), so it’s definitely not a display bug of any sort:

Video showing a recreation of the bug

Note that this bug is extremely sensitive to the types of stacking contexts that can be used to trigger it. You can have roughly the same layout with any of:

changing which div has the position: relative on it
positioning image based on left instead of right
image overlapping div

and the bug will not trigger. This is why I wasn’t able to make a minimal reproduction of what’s actually happening in our app. In our app, the image can (1) be at opacity: 1, (2) overlapping the div, and (3) all transparent colors everywhere are affected, which is even wilder. Shame I wasn’t able to show it off…

Anyways, now I am fairly certain it is a Chrome bug. However, because I do not have the chops to write a patch that fixes it at the source, I have to work around it just using CSS.

Take 1

My thinking:

If adding some blur anywhere stops whatever’s going on with the transparency, then just having a random 1x1 div in the corner of the screen should do the trick!

Sure enough, that did it! No more weird transparency rendering issues, everything looks normal again, we’re all happy yaaaaaaay :)

yeah. um. Remember the first bug? Turns out this trick disables HDR entirely, which is not a good thing if your CEO likes to view HDR videos in his app and can tell the difference.

Take 2

The problem is a combination of HDR, Weird Stacking Contexts, and Transparent Colors. The easiest one of these to fix is Transparent Colors, so if I make the offending background colors Not Transparent, everything will be fine!

Yep! Turns out it’s not too hard to hack together something with Color.js that roughly emulates what the transparent blending would otherwise do. I got it so that even in a side-by-side comparison, I could hardly tell the difference.

And that’s that!! I’ve worked around yet another Chrome bug purely by contoring the CSS into the correct shape, which is why I’m a Software Engineer :)

Conclusion

This bug was really hard to find because it only happened under very specific conditions, and the outcome seemed simply impossible. Blur on one element affecting all elements on the page? Chrome drawing transparent colors incorrectly?? Surely you must be joking, Mr. Feynman. Nope, turns out Chrome has rendering bugs, and stumbling across them is more likely than I thought.

Anyways, hope this was somewhat educational, see y’all next time!

RSS Feed Is Live

Sat, 16 Mar 2024 12:00:00 GMT

My blog now has an RSS Feed!. Again. After the one that came with Hakyll got removed in the upgrade to Astro, and I had to update to @astrojs/rss. I also:

ditched my old janky globbing setup and switched to using Astro content collections
minimized my CSS massively (by hand! wordpress theme was bloated for what I’m doing with it)
removed all custom fonts (it’s using your system serif and monospace fonts now)
fixed the caching in Github Actions so now it won’t have to regenerate all my images every time I push.

All in all, very productive blog work! but boy is it late at night now whoops.

Working Around A Chrome Bug (And Other CSS Limitations)

Fri, 15 Mar 2024 12:00:00 GMT

In my previous post¹, I covered how Chrome’s mask-image support doesn’t do any antialiasing for radial-gradient, and what that meant for making notification badge cutouts. In this post, I’ll go over a different way of providing a mask-image that does have antialiasing (in all browsers!), namely SVGs.

Recall that we left off with a mask that looked something like this:

<style>
.take3 .pfp {
  border-radius: 50%;
  mask-image:
    radial-gradient(
      ellipse 20% 20% at 80% 20%,
      transparent 100%,
      white 100%
    ),
    linear-gradient(
      to bottom,
      white 40%,
      transparent 40%
    ),
    linear-gradient(
      to left,
      transparent 20%,
      white 20%
    );
  mask-composite: subtract;
}
</style>
<div class="horiz">
  <div class="avatar take3">
    <img class="pfp" src="/randomlogo.png">
  </div>
  <div class="avatar take3">
    <div class="pfp"></div>
  </div>
</div>

That is, one that was a little chunky on the curved parts. This got me wondering: is it possible to do better?

As it turns out, yes!! Using an svg instead of a CSS geometry object works a lot better. In the work codebase, I wrote a fancy SVG generator that embeds things directly inline with url('data:image/svg+xml;utf8,${svg}'), because we already use a lot of templated CSS generated by JS, but for this demonstration I’ll just be using regular urls:

<style>
.take4 .pfp {
  border-radius: 50%;
  mask-image: url("/2024-03-15/mask1.svg");
  /* This is used because it's a lot easier to generate complex SVGs that are black and white, instead of transparent and not transparent */
  mask-mode: luminance;
}
</style>

Great!! Now, let’s just slap our notification badge back on and… voilà!

<style>
.badge {
  font-variant-numeric: tabular-nums;
  border-radius: 14px;
  background-color: var(--color-orange);
  font-size: 16px;
  line-height: 16px;
  position: absolute;
  top: 6px;
  left: 66px;
  padding: 6px;
}
</style>
<div class="horiz">
  <div class="avatar take4">
    <img class="pfp" src="/randomlogo.png">
    <div class="badge">100</div>
  </div>
  <div class="avatar take4">
    <div class="pfp"></div>
    <div class="badge">9001</div>
  </div>
</div>

100

9001

Now, for the sake of the argument, just like last time, let’s consider what happens if you want More Badges. And also, you want to have this be composable, so you can nest them. Well, then this happens:

<style>
.take5 {
  mask-image: url("/2024-03-15/mask2.svg");
  mask-mode: luminance;
}
.badge-lower {
  position: absolute;
  background-color: var(--color-green);
  width: 28px;
  height: 28px;
  bottom: 6px;
  right: 6px;
  border-radius: 50%;
}
</style>
<div class="horiz">
  <div class="avatar take4">
    <div class="take5">
      <img class="pfp" src="/randomlogo.png">
      <div class="badge">100</div>
    </div>
    <div class="badge-lower"></div>
  </div>
  <div class="avatar take4">
    <div class="take5">
      <div class="pfp"></div>
      <div class="badge">9001</div>
    </div>
    <div class="badge-lower"></div>
  </div>
</div>

100

9001

Gah!! Suddenly, all the talk about no-clip and no-repeat from the last post comes rushing back. This is no longer any sort of Chrome bug I don’t think, but rather just a way the CSS mask property is specified to behave.

Reading the mask specification more thoughroughly, we begin to see a solution:

Use no-repeat and no-clip (obviously)
Explicitly set the mask-position and mask-size so it will only cover the part of the image we want
Combine it with another mask that does cover the whole image, using mask-composite: exclude to cut away at that mask.

Doing that, we come up with the following solution:

<style>
.take6 {
  mask:
    /* We're actually going to cut away at the lower mask, and need to use an inverted image */
    /* mask-image: */ url("/2024-03-15/mask2-inverse.svg")
    /* mask-mode: */ luminance
    /* mask-composite: */ exclude
    /* These make it so we only mask the one part */
    /* mask-repeat: */ no-repeat /* mask-clip: */ no-clip
    /* position the mask and size it appropriately */
    /* mask-position: */ bottom left / /* mask-size: */ 100px 100px,
    /* And finally, the mask that covers the whole image */
    linear-gradient(white, white) no-clip luminance;
}
</style>
<div class="horiz">
  <div class="avatar take4">
    <div class="take6">
      <img class="pfp" src="/randomlogo.png">
      <div class="badge">100</div>
    </div>
    <div class="badge-lower"></div>
  </div>
  <div class="avatar take4">
    <div class="take6">
      <div class="pfp"></div>
      <div class="badge">9001</div>
    </div>
    <div class="badge-lower"></div>
  </div>
</div>

100

9001

And there you have it! Perfectly anti-aliased, composable masks for all your CSS masking needs.

I swear we’re the only application I’ve seen that actually does something like this. Everywhere else I’ve looked either doesn’t use a mask at all, or just sets a border color to fake a mask.

This is why I put in the work tho, to prove we’re better than that. CSS allows us to do these things, so do them we shall!!

Footnotes

oh no it’s been very too long since that last post. I meant to get this out sooner but have been procrastinating with other projects in the meantime… figured that, while I was doing other work for my blog, I should at least write an article :P ↩

My Blog Has More CI Now

Fri, 29 Dec 2023 12:00:00 GMT

At work, I’ve seen my coworker set up this new cool tool called Rennovate which automatically scans your repository and opens PRs for updating dependencies. You control it through an online dashboard, or, and this is the part I found really cool, through the Github UI itself via (presumably) some fun webhook hacks.

However, for this system to work properly, there needs to be some way of knowing if an opened PR is good to merge or not. I don’t really want to do the manual work of “check out an automatically opened PR” →\rightarrow→ “make sure it works locally” →\rightarrow→ “merge the PR and clean up local git” when I could have computers do that work for me! So I did a bunch of work to make computers do that work for me :]

I borrowed heavily from the canonical Astro usecase, their own documentation site, which seems to have been a pretty good template. Lots of “fun” making typescript and eslint and prettier all play nice together, but yep now it works now!! It correctly failed on the major version update of Astro, which I’ll need to do manually at some point, but that’s for another time yes.

I Think I Found A Chrome Bug

Thu, 28 Dec 2023 12:00:00 GMT

At work, I’m in between a few big projects, spending my time polishing the UI so it shines. One of those polish items has been to revamp how we do notification badges, and through that, I think I discovered a WebKit bug EDIT: Chrome bug. Upon further testing, this bug does not appear to be in Safari, despite also appearing in Firefox? idk man it’s just weird

The simplest way to do a notification badge is like this:

Nice ‘n easy! Just slap a position: absolute inside a position: relative and everything’s pretty much good-to-go. The only downside to this is that is doesn’t quite look professional enough. A neat little detail to have is a little transparent space between the profile picture and the thing going on top of it, like in Discord.

Now, you can just fake this effect by slapping on a border: 5px solid var(--background-color) onto the the thing, but if the background color ever changes you have to animate this border property with the same curves, so for separation of concerns it’s just easier¹ to have the image be truly transparent in that area.

So! Let’s try to do exactly that. A quick dive into what exists out there in general brings up the mask property, with a specially crafted radial-gradient as the mask source. Let’s see how this looks, first without the indicator:

<style>
/* irrelevant properties omitted for brevity */
.take1 .pfp {
  border-radius: 50%;
  mask-image: radial-gradient(
    ellipse 20% 20% at 85% 85%,
    transparent 100%,
    white 100%
  );
}
</style>
<div class="horiz">
  <div class="avatar take1">
    <img class="pfp" src="/randomlogo.png">
  </div>
  <div class="avatar take1">
    <div class="pfp"></div>
  </div>
</div>

Interesting!! What this is doing, effectively, is drawing a completely transparent circle at a specified location within the bounding box, using a transparent 99% and white 100% stop to effectively get a hard line instead of a gradient.

Unfortunately, this approach is not without problems.

Problem 1: It Looks Chunky

This becomes more apparent when we try to add the dot:

<style>
.indicator {
  aspect-ratio: 1;
  border-radius: 50%;
}
.take1 .indicator {
  width: 30px;
  position: absolute;
  bottom: -3.5px;
  right: -3.5px;
}
</style>
<div class="horiz">
  <div class="avatar take1">
    <img class="pfp" src="/randomlogo.png">
    <div class="indicator green"></div>
  </div>
  <div class="avatar take1">
    <div class="pfp"></div>
    <div class="indicator green"></div>
  </div>
</div>

Now, we can more clearly see that the masked-out part is not anti-aliased, unlike the circular borders of the profile picture and indicator. This is part of the bug I’m referring to in the post title.

Problem 2: It Clips

Things get worse if we try to add a second indicator (and not just from a design perspective, lol).

<style>
.take2 .pfp {
  mask-image: radial-gradient(
    ellipse 20% 20% at 85% 85%,
    transparent 100%,
    white 100%
  );
  background-color: initial;
}
.take2 img {
  border-radius: 50%;
}
.take2 .img {
  width: 100%;
  height: 100%;
  border-radius: 50%;
  background-color: black;
}
.take2 .indicator {
  width: 30px;
  position: absolute;
}
.take2 .indicator1 {
  bottom: -3.5px;
  right: -3.5px;
}
.take2 .indicator2 {
  top: -3.5px;
  left: -3.5px;
}
</style>
<div class="horiz">
  <div class="avatar take2">
    <div class="pfp">
      <img src="/randomlogo.png">
      <div class="indicator indicator2 orange"></div>
    </div>
    <div class="indicator indicator1 green"></div>
  </div>
  <div class="avatar take2">
    <div class="pfp">
      <div class="img"></div>
      <div class="indicator indicator2 orange"></div>
    </div>
    <div class="indicator indicator1 green"></div>
  </div>
</div>

Huh?? You may have many questions, such as “why not just put class="indicator2" outside the part that clips?” and “wait hold on why is is clipping like that in the first place?” The answers to which are:

React Component Hierarchy²: it’s simpler to make a component that says “I want to optionally add exactly one new indicator” on something than so say “I want to add a configurable number of indicators with variable positions”, because otherwise you have to somehow blend the masks which can get hairy³.
We’re Getting To That

It may have been wiser to bite the bullet and figure out how to reorganize the component hierarchy, but now I was curious to get to the bottom of this and figure out if I could do something about the first problem at the same time.

Poking around in developer tools, we can confirm that it’s just the mask property causing this clipping. With the mask on, affecting the bottom right corner, we suddenly get clipping in the top left. A panicked search for “mask clips things that it shouldn’t” brings us to the mask-clip property. Swell! Let’s just specify no-clip then, and:

Great!! Seems we’re done then, no? Ignoring the slight imperfection in the mask, of course. Well, if we happen to move .indicator2 to the right side, then…

<style>
.take2 .indicator3 {
  top: -3.5px;
  right: -3.5px;
}
</style>
<div class="horiz">
  <div class="avatar take2">
    <div class="pfp noclip">
      <img src="/randomlogo.png">
      <div class="indicator indicator3 orange"></div>
    </div>
    <div class="indicator indicator1 green"></div>
  </div>
  <div class="avatar take2">
    <div class="pfp noclip">
      <div class="img"></div>
      <div class="indicator indicator3 orange"></div>
    </div>
    <div class="indicator indicator1 green"></div>
  </div>
</div>

HUH. WHAT. That’s strange, to say the least. All we did was change left: -3.5px to right: -3.5px and suddenly the top is shorn off?? Playing around a bit more and setting right: 10px, we see a clue:

That indentation in where it gets cut off is suspicious.. it’s almost like…

The Mask Is Repeating

That is, when the content goes outside the bounding box, a duplicate copy of the mask is placed there. Skimming the sidebar of the MDN tab where we found mask-clip, we do indeed see there is a mask-repeat property, whose default value implies this exact behavior. Attempting to set no-repeat on the mask, however, just results in it clipping again.

Problem 3: It’s Problem 1 Again, But Different

The designer comes back and says “that indicator in the top right is actually meant to be an unread message count, and should have a pill-shaped mask.” OK, no problem, some reading of MDN has us come across the mask-composite property, and a bit of fooling around with it later we have a prototype:

<style>
.take3 .pfp {
  border-radius: 50%;
  mask-image:
    radial-gradient(
      ellipse 20% 20% at 80% 20%,
      transparent 100%,
      white 100%
    ),
    linear-gradient(
      to bottom,
      white 40%,
      transparent 40%
    ),
    linear-gradient(
      to left,
      transparent 20%,
      white 20%
    );
  mask-composite: subtract;
}
</style>
<div class="horiz">
  <div class="avatar take3">
    <img class="pfp" src="/randomlogo.png">
  </div>
  <div class="avatar take3">
    <div class="pfp"></div>
  </div>
</div>

I could not honestly tell you how this works, only that it does. But as you can see from this image (and maybe even on your browser!), there’s a huge problem: not only is the circular part of the mask not anti-aliased, but it’s not even masking the anti-aliased pixels from the border-radius. This is the clearest demonstration of the WebKit bug I was promising to expose in this post, so here you go :)

A Better(?) Approach

Fortunately, that same day I found this bug I also came up with an approach that solves all of these problems, at a small cost to generated code size. But this post is long enough as it is already, and you’re probably fatigued reading this, so I think I’ll leave that for a follow-up post.

As an aside, Markdown is really nice for things like this. I can just include HTML right in with the rest of the formatted text and it just works?? Magical.

Footnotes

You’ve read the rest of the post. You know this isn’t actually easy. ↩
Yes, we ship a React.js Electron app. Yes, the performance is much of a problem as you think it is. I will not be taking any questions about this. ↩
As if I didn’t end up writing something like that anyways, lol ↩

New Blog Framework: Astro

Sun, 22 Oct 2023 12:00:00 GMT

I was fed up with Hakyll, so I’ve switched my site to use a new blogging framework, Astro.

The reasons for this are twofold:

Hakyll Stank

I covered this in more detail in my last post, but the overall gist is that Hakyll deploys were way too slow for me. I spent a long time setting up Github Actions correctly, only for them to get no speedup in the average case because the cache entries expire in between the (very long) periods between my posts. I could just

aws s3 sync --delete _site/ s3://www.hazelduvall.dev/

every time I updated my site and just have everything locally since I’m running the preview server that automatically populates the build directory anyways, but gosh darn it I’m just unreasonably attached to CI. Main benefit: it lets me have “draft” posts by just not publishing them to git. (that’s kind of it)

Astro Is Nice

I tried it out for another website (that I made after this one) and really enjoyed it. Much easier to get set up than something like Hakyll for sure. Also more flexible than something like Zola, which I used for my Rust Stuco Website. My blog has gotten a bit complicated in places, what with the post links and the Photography page, so having full componentized control over things is nice. Plus, I’m now even more used to thinking in the component model thanks to what I do for work (tho, I’m still not that good at it).

I did need quite a few hacks to get it to work tho. Biggest thing was doing the equivalent of Astro.glob. The docs mention that you can use Vite’s import.meta.glob instead, but don’t really provide any examples. Lacking any examples on the internet, I stumbled my way around typescript to find the following solution:

// file:src/env.d.ts
// the order here is important! Astro's `import.meta.glob` typing is broken for me :(
/// <reference types="vite/client" />
/// <reference types="astro/client" />

// file:src/data/posts.ts
// A `MarkdownInstance` is the output from Astro's integration of Vite when it imports markdown files from a glob
import type { MarkdownInstance } from "astro";
// A `AstroComponentFactory` represents a component object that can be used like `<Component />` instead of `{Component}` in the body of the Astro component it's included into. Import looks a little strange but it's fine :P
import type { AstroComponentFactory } from "astro/runtime/server/index.js";

// Interface for the frontmatter in the posts
interface RawPost {}
// Interface for the components passed to the eventual post component
interface PostProps {}

export interface PostPropsWithComponent extends PostProps {
  Content: AstroComponentFactory;
}
export const getPosts = async (): Promise<PostPropsWithComponent[]> => {
  const rawPosts = import.meta.glob<boolean, string, MarkdownInstance<RawPost>>(
    "../posts/*.markdown",
  );

  const posts = await Promise.all(
    Object.values(rawPosts).map(async (producer) => {
      const post = await producer();

      const props: PostPropsWithComponent = {
        Content: post.Content,
      };

      return props;
    }),
  );

  return posts;
};

I hereby release the above code snippet under an MIT license, or something. Basically, feel free to adapt and extend as you see fit, just link to this blog post or something :P

Other nice things I got out of Astro:

Built-in code highlighting for language-tagged code blocks. (It would’ve been theoretically possible but kind of hard with Hakyll, so very nice to get that for free!)
Built-in image optimization (I haven’t used it yet since it would require further restructuring)
LaTeX\LaTeXLATEX, via remark-math and rehype-katex plugins (very easy to install!)

What’s Next?

Not sure! One thing I’d like to do is update this blog’s theming. Not too big a fan of the font anymore, nor some of the color/layout choices of the Wordpress Twenty Fifteen theme (yes, really) (still the best of the default Wordpress themes, imo). That smells suspiciously like doing even more rewrites of my blog, though, so perhaps I’ll leave that alone for now :)

I'm A Maintainer Now??

Wed, 11 Oct 2023 12:00:00 GMT

Recently, some code I hacked on for work made its way onto Hacker News, where it got a decent amount of attention. Nothing crazy, but enough that I was pleasantly surprised.

The code in question is node-webrtc, specifically, my fork of it. This is an NPM package for letting NodeJS access WebRTC APIs normally only accessible in browsers. It does this by taking the original libwebrtc source code and plugging it into Node Native Addon bindings that closely mirror the ones in the WebRTC specification.

I needed to fork this package because the original became unmaintained in 2021, and had suffered enough bitrot when it came to the Node Native Addon side of things that it no longer worked with the latest Node LTS (18). In order to use it in our codebase (for various reasons), it needed to be upgraded.

In addition to the upgrade of the APIs, I also started to upgrade the version of libwebrtc it was building against, also because our code needed features from slightly newer versions. I upgraded it from M87 to M94, going through M90 and M92 first, and am in the middle of upgrading to M98. These incremental steps proved to be the best option for getting things working (hard to fix things when everything is broken, better to only break a few things at a time), but means it’s slow progress towards an eventual goal of M118 and our patched version of libwebrtc.

This is all well and good, and once I was moved to another project I thought this would effectively be the end of the saga. I did want to keep working on it, but other tasks needed doing! It wasn’t until the Hacker News post came around that I caught the attention of node-webrtc’s previous maintainer, who offered to hand over ownership of the Github organization and NPM package.

So, yeah, that happened! I haven’t merged my fork back into upstream yet, nor added a new NPM package version. This is because I am still working on other tasks for Roam and haven’t put a lot of time into this library recently. I’ll keep plugging away at it when I can, and will be responsive to Github issues on the fork during normal business hours. Not sure what the appropriate channels are to make an announcement when I do make the changeover, perhaps just a note in the README.md is appropriate.

This feels surreal! Very interesting to have a codebase dumped on me like that, just because I put in the work for it. No pressure or anything since it doesn’t seem widely used, but still thinking about how to uphold a legacy and start building a reputation more. I would like to be able to devote more energy towards improving open source and building cool stuff people can see that has my name on it. Frontend work is cool and all, but this is more the stuff I signed up for really when I wanted to become a Software Engineer.

Remember to Munge Your Headers!

Fri, 08 Sep 2023 12:00:00 GMT

So for work, I’ve been hacking on the Selective Forwarding Unit (SFU) that we use to send media between clients. Through this, I’ve found a couple bugs, one of which had quite a long but rewarding debugging journey.

The Bug

This bug manifested in such a weird way that I didn’t actually catch it at first. Regular audio and video streams worked just fine. Screenshares would start, but then crawl to a halt after the first few frames, or after a couple seconds. I wasn’t catching this bug at first because I was sharing my static desktop and wasn’t expecting it to change; only after I started doing more intensive tests, running a clock on the screenshare for exact timing, did I notice that things were weird.

Debugging

At this point, I had done enough other debugging to have a fairly standard checklist of things to look at. As usual, chrome://webrtc-internals was my go-to.

Looking at the graphs for the screenshare upload, I was seeing plenty of packets sent, as well as a burst of packets whenever there was large activity on the screen.

So, I could rule out:

The upload transceiver not getting created
The connection not getting negotiated
The upload MediaStream not getting attached to the transceiver

Next, I looked at the download side. I saw a graph that matched almost exactly what I was seeing on the screen: a burst of packets, and then nothing.

Because I was seeing something, I could rule out the same things as with the upload. Which was concerning, because up until this point that’s all I had been checking. It seemed I needed to dig a little further.

I first fired up Wireshark to confirm that, indeed, packets were actually being sent from the SFU to the downloader. This was weird to me, since Wireshark was reporting packets being sent even though Chrome wasn’t reporting any packets

I then ran Chrome in debug logging mode and looking at a live tail of chrome_debug.log. In this, I saw a very suspicious log message being spammed a bunch as soon as the faulty screenshare started:

[6612:28688:1011/191519.091:VERBOSE1:rtp_transport.cc(196)] Failed to demux
RTP packet: PT=96 SSRC=2139439780 MID=e^A
[6612:28688:1011/191519.091:VERBOSE1:rtp_transport.cc(196)] Failed to demux
RTP packet: PT=96 SSRC=2139439780 MID=e^B
[6612:28688:1011/191519.091:VERBOSE1:rtp_transport.cc(196)] Failed to demux
RTP packet: PT=96 SSRC=2139439780 MID=e^C

Those ^ characters were actually unprintable! (Just had to reproduce them here as text). It was looking like uninitialized memory, or a counter, and the very least, something that should never have been encoded as text. The relevant line in chromium confirmed that yes, something was going very wrong, since this value should always be a string in a properly-encoded packet. Somehow, the SFU was sending malformed packets!

Where Does This Value Come From?

This MID (Media Identifier) value is supposedly to distinguish between different media sections when multiple are bundled together. In practice, the SSRC (Synchronization Source) value seems to be more than enough, but the WebRTC standard requires that MIDs be supported if multiple streams are bundled in one connection. To save on the overhead of creating connections, we do bundle the screenshare downloads with the camera downloads, and looking at the session descriptions it was clear that this header extension was being negotiated.

What Are Header Extensions?

Header Extensions are defined in RFC8285 and are basically little bits of extra data that don’t need to be in all RTP packets, and are negotiated per-stream like everything else in the SDP¹. To save on bandwidth, each of the possible header extensions represented by a URI (Universal Resource Identifier) gets mapped to an ID determined by the offering participant. An example mapping looks like this:

a=extmap:3
http://www.ietf.org/id/draft-holmer-rmcat-transport-wide-cc-extensions-01
a=extmap:4 urn:ietf:params:rtp-hdrext:sdes:mid

This means that packets will have data for header extension ID 3 is fed into the transport-wide congestion control mechanism, and data for header extension ID 4 is the MID for the packet.

Because the malformed MIDs are specified in a header extension, I decided to dump all header extensions for screenshare packets coming out of the SFU, and sure enough, the header extension ID for the MID extension was present and being set to invalid values!

The Solution

Due to the nature of the SFU, the header extension id mappings can be different for the uploading participant and the downloading participant. And, as it turned out, we were not properly stripping all header extensions from uploader packets before forwarding them to the downloaders. The correct thing to do (which we were already doing for some header extensions) is to store any header extension data out-of-band for forwarded packets, and just before sending the packet to the downloader, add back in the header extensions with the correct ids for that download stream.

And that’s it! Hope you were able to take something away from this debugging journey, I sure learned a lot about WebRTC as I went.

Footnotes

The SDP also contains negotiation for media codecs, codec extensions, what streams exist, etc. ↩

WebRTC Perfect Negotiation

Wed, 30 Aug 2023 12:00:00 GMT

As part of my job, I have been working pretty closely with WebRTC, a technology for doing real-time audio/video communications. It’s an open standard steered entirely ahem supported by Chrome, so it’s the natural choice for anything Electron-based.

It’s a peer-to-peer technology, so part of the standard includes specifications for how the peers do a dance for agreeing on who’s sending what tracks, who’s initiating the TLS encryption, what are the mutually supported codecs anyways, that sort of stuff. This gets pretty complex since it’s all being done over the network, asynchronously. Perfect Negotiation, then, is a way of setting up the callbacks for this negotiation such that everything “just works” and both sides can use higher-level APIs instead of worrying about the nitty-gritty.

The blog post linked above does an excellent job of explaining the concept in more depth and providing code to go along with it. However, it’s actually slightly incorrect in my experience!

In the section for handling remote session descriptions, there is a piece of code that goes like this:

if (description.type == "offer" && pc.signalingState != "stable") {
  if (!polite) return;
  await Promise.all([
    pc.setLocalDescription({ type: "rollback" }),
    pc.setRemoteDescription(description),
  ]);
} else {
  await pc.setRemoteDescription(description);
}
if (description.type == "offer") {
  await pc.setLocalDescription(await pc.createAnswer());
  io.send({ description: pc.localDescription });
}

The error is as follows: calling setLocalDescription({type: "rollback"}) in Chrome when pc.signalingState === "stable" will throw an error. Additionally, Chrome doesn’t seem to guarantee the promises inside Promise.all are necessarily run in order. So, the correct code would actually be something like:

if (description.type == "offer") {
  if (!polite && pc.signalingState !== "stable") return;
  await Promise.all([
    async () => {
      if (pc.signalingState !== "stable") {
        await pc.setLocalDescription({ type: "rollback" });
      }
    },
    pc.setRemoteDescription(description),
  ]);
  await pc.setLocalDescription(await pc.createAnswer());
  io.send({ description: pc.localDescription });
} else {
  await pc.setRemoteDescription(description);
}

The original blog post seems to assume that the order of arguments to Promise.all will control the order more than I have observed. setLocalDescription may be queued first, but it isn’t necessarily polled to completion first; in chrome://webrtc-internals, I see setLocalDescription queued before setRemoteDescription, but setRemoteDescriptionOnSuccess before setLocalDescriptionOnFailure.

Not sure if this information is present anywhere else online, I sure couldn’t find it when searching for “Failed to execute ‘setLocalDescription’ on ‘RTCPeerConnection’: Called in wrong signalingState: stable”. Not that it will matter, my blog has terrible SEO lol.

Whoops! Let's Get Back To Posting

Sat, 17 Jun 2023 12:00:00 GMT

I seem to have fallen behind on updating this blog! A lot of important life events have occurred, which I probably should have blogged about at the time when they were still fresh:

My final semester at college
Graduating from college
Starting my job
Some secret fourth thing

I’ll backfill these posts in the coming days (weeks?) now that I have even better CI for this site. Riiiight after I change all the links from sequentially numbered to date-and-title format, breaking all my SEO that was never there in the first place :P

Starting Work

Wed, 31 May 2023 12:00:00 GMT

Wow! It sure does feel weird to no longer be a college student any more.

After the usual craziness that is moving between apartments, I have settled in Brooklyn, NY, the city right next to the Big Apple, for at least until the lease expires.

Here, I’ll be doing remote work for a remote work company, Roam!. I talked about them earlier, but today was my first day so it’s really official.

Team seems nice, work seems interesting, it’s a pretty big codebase to learn but I’ll get right on it!

Graduation!

Sun, 14 May 2023 12:00:00 GMT

A ton of extended relatives came out to see me graduate, which was super kind of them and honestly pretty fun to hang out with.

I can’t believe I’m finally done with it! I had some good times, some bad times, made a lot of friends, took a lot of classes, and the best part is, I don’t think I have regrets really. I made the most out of what I had, which was a lot, and had fun while doing it.

So long CMU, thanks for the memories!

I Was An OS TA

Fri, 05 May 2023 12:00:00 GMT

Wow. I actually made it. Since early-ish at CMU, I had known that an “OS TA” was a laudable thing to be¹, but after The OS Semester, I wasn’t sure I’d actually be one. Sure, I’d applied the semester I was already scheduled to be a TA for 15-316, and applied again “just in case”, but I didn’t really think I’d actually get it. It was only (checks notes) December 24th that I found out for sure, through a friend refreshing the TA assignment page for their own position. What a Christmas present lol.

This final semester at CMU, I only took 15-462 Computer Graphics, so I had plenty of time to attend another class, do research for a professor, teach my stuco, and TA this class. hm. work expands to fill the available time, huh?

What Does An OS TA Do?

In a nutshell:

grade projects
hold office hours
slowly add to existing projects

All in collaboration with Professor Dave Eckhardt, who has been teaching the course for long enough to have taught some current professors at CMU.

Oh! Also! I was the only “official” TA that semester, all others were alumni or PhD students kindly contributing in their spare time.

Grading

This is The Big One™. All 5 projects, 2 homeworks, a midterm, and a final.

For those latter 3 categories of written assignment, it’s not too big a deal. There are fairly standardized rubrics/methodologies for grading, and the main feedback given is point values (which students can come into office hours to discuss if they want). What makes OS grading so infamous² for its grading is the sheer amount of extremely detailed feedback given on all the coding projects.

For those coding projects, there is a standard set of tests that the code has to pass³, and then the rest of the points are left to Architecture and Code Quality. How do you grade these? You read the code, of course! And by “you” I mean “me and Dave and a small army of masters/PhD students/alumni”.

Reading all the code written by other students (especially by masters students!) was really useful. It made me appreciate my own coding style⁴ also opened my eyes to other things you should pay attention to while coding.

I will be part of the “small army” next semester⁵, and will be more than happy to help read more code :)

Office Hours

As the only “official” TA, I did hold a lot of office hours. Dave (Eckhardt) and Dave (O’Halloran) also held their own professor office hours, and a few alumni held remote office hours on weekends.

In office hours, I was supposed to never give out solutions, only ask questions about what they had tried and hadn’t tried in an attempt to get them to think about the problem more fully. This method is difficult and frustrating on both sides but on the whole I agree that it does lead to better learning outcomes.

I stuck to a fairly regular schedule and was always surprised by when students did or didn’t come in. There were sometimes big influxes after an assignment was released, right before it was due, or even just in the middle, but just as often those times would be empty. And I would work on grading during the downtime instead (:

Adding To Existing Projects

Due to a limit on the number of hours I could work in a week⁶, and a general sense of wanting to do other things besides OS, I unfortunately did not do a whole lot of this. I tried my best at updating the VM infrastructure to the now-publicly available version instead of the older still-proprietary one CMU uses, and failed due to some very interesting reasons that I should probably write up some time.

I was able to write a small test program for P4 that ran on a hypervisor that the students wrote, but really Dave did most of the tough stuff with super funky cursed Assembly and I just had to understand why it worked and hook it up into a small shell. It is called dog and when you type “dog” it types out “cat” instead, wow crazy.

Did I Enjoy It?

I think this was a very worthwhile experience! I highly respect Dave Eckhardt so having more time to interact with him was fun. There were also some fun parts to grading and office hours, but on the whole it was a lot of work. I still would do it again though. Even if it feels like a slog during, I can come out the other side feeling like “yeah, I Did That”. Just like taking OS, I guess.

Footnotes

It’s a TA for a 4xx-level class, after all! And a prestigious one at that ↩
For being perpetually late. I did the best I could to not continue this trend, I believe a big reason is Dave taking on a huge workload and not being able to deliver until it would be exceedingly inappropriate not to, not just regularly inappropriate. ↩
Dave likes to say that “just passing all the tests will get you to a C”. I cannot say enough about the grading mechanism to confirm nor deny the veracity of this statement. ↩
Lots of documentation. Everywhere. Explain the why of every single piece of code before I write it down because I know I will forget stuff later. ↩
I needed basically the whole semester to get into a groove with grading, and Dave joked that he wants a return on his investment. ↩
20 hours. That turned out to not be enforced at all, and I could go over very easily without raising any flags it seemed. ↩

Some Cool Labs I Helped Make, Pt. 2

Tue, 17 Jan 2023 12:00:00 GMT

Read Part 1 first! That lab was already super cool, but this one is even better imo 😎

Lab 2: Authorization & Trust

This lab is all about writing an automated theorem prover for a limited set of theorems to prove. As a part of that, you get to learn the answers to fun questions like

What Does It Mean To Prove Something?

That is, how can we show something like (A→B)∧(B→C)→(A→C)(A \rightarrow B) \land (B \rightarrow C) \rightarrow (A \rightarrow C)(A→B)∧(B→C)→(A→C) is true?

What Does It Mean For Something To Be True?

Great question! Is AAA true? Why not?

The answer is, it depends on your interpretation of AAA. And I mean that word “interpretation” quite literally, that’s the vocabulary we use for “the assignment of truth values to variables.

There is an interpretation where AAA is true: {A:⊤}\{A: \top\}{A:⊤}, and an interpretation where AAA is false: {A:⊥}\{A: \bot\}{A:⊥}.

So to be more precise, when we are asking if some logical formula is true, we are really asking whether it is true in all interpretations. Here’s some symbols:

⊨P↔∀I.I⊨P\models P \leftrightarrow \forall I. I \models P⊨P↔∀I.I⊨P

So Back To The Main Question: Proofs

So now we have this idea of interpretations, and truthiness from the standard boolean connectives (∧,∨,→,¬)(\land, \lor, \rightarrow, \lnot)(∧,∨,→,¬), and need some way of showing that a formula will hold under any interpretation. Not only that, the way we do this needs to be in some format that a computer can quickly recognize and manipulate, so text is right out¹.

The solution we use is proof transformation rules in a tree structure. A single rule might go something like this:

Γ,A⊢BΓ⊢A→B,Δ(→R)\frac{\Gamma,A \vdash B}{\Gamma \vdash A \rightarrow B, \Delta}(\rightarrow R)Γ⊢A→B,ΔΓ,A⊢B(→R)

The more complex rule is at the bottom, and gets broken up into simpler rules as you go up the tree. More explanations of symbols:

⊢\vdash⊢: If everything on the left side is true, then at least one thing on the right right is true. Like a big implication but more flexible for our needs.
Γ,Δ\Gamma,\DeltaΓ,Δ: Leftover clauses on the right and left hand sides of the ⊢\vdash⊢ to signify which rule we are applying to and when we can carry things over².

To finish a proof, you need to “close out” every single branch, signified by ⋆\star⋆. Here are some rules that do this:

⋆Γ⊢⊤,Δ(⊤)⋆Γ,⊥⊢Δ(⊥)⋆Γ,P⊢P,Δ(id)\frac{\star}{\Gamma \vdash \top, \Delta}(\top) \quad\quad \frac{\star}{\Gamma, \bot \vdash \Delta}(\bot) \quad\quad \frac{\star}{\Gamma,P \vdash P,\Delta}(id)Γ⊢⊤,Δ⋆(⊤)Γ,⊥⊢Δ⋆(⊥)Γ,P⊢P,Δ⋆(id)

These types of proofs (in the right format) can be easily checked by a computer, since all it as to do is, for each proof rule, check that it was applied properly. Then, since you know all the proof rules are valid, the entire proof must be valid!

Unfortunately, I can’t write out an example, since doing simple ones of those is a homework problem early in the class, but u get the gist right? Moving on!

What Are We Trying To Prove, Exactly?

So imagine you are a smart door lock. You have some notion of “who is allowed to access this door” and “who is allowed to delegate access to this door”. These are can be encoded as affirmation logic³ principles like:

A	B
`door` says `canopen(hazel)`	`door` says forall `p`. (`hazel` says `canopen(p)`) implies `canopen(p)`

These statements have a close tie to cryptography! A “says” clause is like a signed statement, and proving the identity of the signer requires knowing about Certificate Authorities, and it all gets pretty complicated, but! There is a reason for all of it and it can be automated very naturally.

The lab is then to present a series of virtual doors with the appropriate signed statements and a proof using those statements to show that you can enter the door. The proof goals are as follows:

Given a signed statement you can enter the door, construct enough of the rest of the proof (with CAs and such) to get the “says” statement
Given a delegation of authority, open the door
Given a very broad transitive delegation policy, open the door
Exploit the door’s certificate checking code to open a door you logically shouldn’t be able to

Sounds simple right?? :)

I really enjoy this lab because it combines together a lot of disparate topics and they all still work together well, and it really shows the power of computers to do Logic, very quickly, and it feels like magic.

That’s pretty much all I can say without spoiling the rest of the lab, if this still sounds interesting to you I think you should read it and try it out! It is free! I really appreciate that this course is just,, publicly available to anyone who wants to learn it (even if the lectures really bring out the best in the subject but i digress).

That’s about it, thanks for reading!!

Footnotes

Yes, not even ChatGPT can reliably do stuff like this 😔 ↩
As an exercise, convince yourself that all the proof rules I’ve shown are true! ↩
Working with these “says” clauses is a little tricky and has a lot of subtlety to it. If you’re interested in learning more, I highly recommend reading the 15-316 course notes. They are available for free online and are really comprehensive!! ↩

Some Cool Labs I Helped Make

Mon, 16 Jan 2023 12:00:00 GMT

As a full disclaimer, the vast majority of work for these labs was done by the professor for 15-316 Software Foundations of Security and Privacy, Matt Fredrikson. As a TA, I helped a little bit with coming up with the ideas for the labs and also a little with basic file/object structure and autograders.

So, while I can’t really even claim 20% credit for these, I’m still really proud that I had any part in them that all and think they are super neat labs to work through. Also, they’re open source so you can follow along at home too!

Lab 1: Analyzing Safety

This lab is all about instrumenting code and then proving properties about the original code through the instrumented version. You may be wondering what any of these words mean so let’s go through an example.

Tinyscript

Here is some code:

x := i;
while x < (i-i)-1 do
    x := x + 1
done

You may notice that the syntax is similar to, but not exactly the same as, any language you’ve ever seen¹. This is because it is a language you have never seen. For simplicity, this course uses a fairly small language we call Tinyscript with simple and intuitive and easy-to-specify semantics. Full syntax tree and interpreter can be found in the repository.

Safety Properties

Anyways back to safety: a safety property is class² of traces that can be defined by excluding all failing traces³ with just a finite prefix. That is, as soon as you see “something unsafe”, you know the safety property has been violated.

Unfortunately, you can’t neatly divide programs into “satisfies a safety property” and “doesn’t satisfy a safety property” since that would require solving the halting problem⁴. Plus, with our arbitrary-precision variables, you can’t brute-force every input combination to see if the safety property is violated (or the program doesn’t halt in a reasonable amount of time). So, what can you do?

Dynamic Logic

There is a formula called a box modality, which we denote as [α]P[\alpha]P[α]P where α\alphaα is a program, and PPP is a boolean predicate, which is itself a predicate saying “for all runs of α\alphaα, PPP will hold in the terminating state”.

Turns out, there are rules on how to simplify this when given a specific α\alphaα and PPP! I’ll spare you most of the details (take the class or read the code to see how it’s done), but I should note one, which is the rule for loops:

[while(Q) do α done]P↔[if(Q) then α else pass endif;while(Q) do α done]P[\texttt{while}(Q)\ \texttt{do}\ \alpha\ \texttt{done}]P \leftrightarrow [\texttt{if}(Q)\ \texttt{then}\ \alpha\ \texttt{else}\ \texttt{pass}\ \texttt{endif};\texttt{while}(Q)\ \texttt{do}\ \alpha\ \texttt{done}]P[while(Q) do α done]P↔[if(Q) then α else pass endif;while(Q) do α done]P

yeah um there’s a bit of infinite recursion here. One way to handle this is by giving up and pursuing a different area of computer science, since this is again equivalent to the halting problem. Another way is to unroll the loop for “enough iterations” and call it a day. We take the latter approach.

With all these rules to turn box modalities into regular formulae, we can just plug that resulting formula into Z3 and bask in the glory of modern computation. Hooray for automated theorem provers!

Program Instrumentation

So now we have a tool to help us reason about a condition at the end of a program, but safety properties can be violated at any time in our program. How can we reconcile this? Simple: we rewrite the program so that any operation that could violate the property sets a marker that cannot be changed afterwards.

Here is how we would do that for the previously-shown program, for the safety property x > 0:

#violated := 0;
x := i;
if !(x > 0) then #violated := 1 else pass endif;
while x < (i-i)-1 do
    x := x + 1;
    if !(x > 0) then #violated := 1 else pass endif
done

We’ve instrumented every assignment to x with a check afterwards to make sure that it’s value doesn’t dip below zero.

By the magic of “it’s really easy to recur over syntax trees”, doing this instrumentation automatically is fairly simple if you’re clever. With this method, proving the safety property becomes equivalent to proving [f(α)](#violated=0)[f(\alpha)](\texttt{\#violated} = 0)[f(α)](#violated=0) (where fff is the instrumentation function), which we have the tools for!

All the problems in the lab (bounded execution, no unused variables, taint analysis w/o implicit flows) are meant to be solved using this strategy. If you have some time and are curious about this, definitely check it out! It’s free online and the writeup there is much better than what I could explain here.

Footnotes

Unless, of course, you’ve taken this course before ↩
like a set, “but better” ↩
Assignments to variables at each step in the program ↩
Consider: a program that does something unsafe only after (safely) determining if a passed-in program terminates. ↩

Photography Page Is Up!

Wed, 04 Jan 2023 12:00:00 GMT

Hello 2023 I have some big news!! It took a bit longer than I thought but I finally finished the section of my website dedicated to photography.

Photography

^^^ click that above link to go to it. I worked pretty hard on these photos during the semester and learned a ton and I would be very honored if you would take a look at them :)

Oh but do be warned, there is approximately 0 optimization on these so each page is like 100MB or more, make sure you have a good internet connection before trying to load. This is also the part of the website I was complaining was hard in my previous post, turns out component-like things like “extra metadata per photo displayed in a common format” is harder than anticipated.

That’s it for now, hopefully I have another post soon bye!!

I Don't Actually Like Hakyll That Much

Thu, 29 Dec 2022 12:00:00 GMT

Yeah, you heard me right. The static site generator I tried out on a whim so long ago and struggled against for a while isn’t serving me that well. Let’s go over why I’m fed up with it and how we can do better.

The Good

So despite the complaining I’m going to do later in this blog post, Hakyll is honestly really nice. You probably get the best markdown engine anywhere, Pandoc, and for those that really want to spice their text markup experience they have the flexibility to do so with custom plugins¹ and whatnot. It’s also really good at being flexible in how you structure your site, once you write all the rules necessary for it.

The Bad: Haskell

I have not invested nearly enough time into having a good Haskell setup. This could be partly due to the fact that “good Haskell setups” are hard to come by and require deep knowledge of build systems, mostly just found at large companies and even then it’s likely in constant breakage². I had this on Stack, and have now moved to Nix in an attempt to get something that would be able to cache build artifacts for CI because it takes a really long time to build everything from scratch even on my not-puny machine.

Neither of these are perfect solutions. I do however now have a language server and code formatter set up so the dev experience isn’t as horrible as it was before and overall the language is kinda nice actually, it’s just that

The Ugly: Hakyll

As explained in some of my previous posts, Hakyll incremental rebuilds work via a pretty nifty dependency system. Unfortunately, tapping into that like I tried to do is stupid hard so I wouldn’t recommend it.

Full rebuilds are probably recommended anyways, because you’ll need to do one every time you rebuild your site executable. Why would you need to rebuild your site executable you ask? Well, if any of the following happen:

Adding a new page not already covered by a glob
Adding/renaming “derived metadata” fields
- That is, metadata not directly encoded into the file
Adding a new template layer to a page/changing the name of a template
- Template layers are effectively how you do “components”, so these may change a lot depending on your style

Also, actually doing components with Hakyll is a PITA. “Make an item with no route and then use a partially applied template to load those items from a listField” statements dreamed up by the utterly deranged³. After having used Astro and Svelte for another project and getting past the initial Javascript apprehension I actually really like their model a lot more. When making a page you can just drill down on the components + loops put in-line with the markup, rather than bouncing between your content folder, template folder, and hakyll rules, having to rebuild every time you edit the latter.

Will I Rewrite This Blog?

Maybe! I’ll definitely be using Astro if I do. Astro still supports markdown and is fairly flexible in how it can be structured so I hope that a conversion wouldn’t be too hard. I’d also like to not spend that much time on it and once this project is over I don’t foresee Hakyll taking this much maintenance and frustration for a while again, so maybe later.

Footnotes

I’m using one for LaTeX\LaTeXLATEX embedded as SVGs which in hindsight isn’t the best idea of accessibility but carrying on ↩
Just like C/C++! ↩
This is actually what I’m doing for another page on this website that I’ll make a post about soon. ↩

Fall 2022 Semester Is Over

Thu, 15 Dec 2022 12:00:00 GMT

So. That’s it huh. Final final project turned in, gearing up for going back to see family over break, I made it through.

In this post I’ll review each of the classes I took this semester and maybe some general thoughts on what spring will be like. Only ordering is lexicographic class number.

10-414: Deep Learning Algorithms And Implementation

Absolute banger of the class. The two profs, Tianqui (TQ) Chen and Zico Kotler were/are doing cutting-edge research, and even I recognized their names from big papers. So already u know they’re gonna be dropping knowledge bombs.

This was the second time this class was run, but it hardly showed. You wouldn’t be able to tell just from the lectures (really good, both of them), just some rough edges in certain places in the labs but nothing that couldn’t be overcome.

This class also ran as as open online version with everything needed to complete the course public, which I think is super cool and really shows their commitment to open source.

18-349: Embedded Systems

“Mini-OS” is how I liked to refer to this class. Lab 4 especially (building a small kernel for the embedded hardware) was wayyyyy more tricky than it needed to be; debugging tools for that board were awful (gdb with hardly any breakpoint support, no trace recording, no printf since that’s where the bugs come from). However, it was really fun to do all the labs in Rust, an unusual arrangement we got instructor permission for.

The content was pretty interesting, lots of introductions to topics you have to care about when putting computers in systems that will interact with the real world. Shame some of the labs were so crazy it spoiled my opinion a little, otherwise I really enjoyed this class. Never had Prof. Gregory Kesden before but I kinda get what the hype around him is, Prof. Mark Budnik was really nice too.

21-373: Algebraic Structures

Just a chill math class with Prof. Richard Statman. Turned out to be only homework-based, which I did not know going in but really liked!! Learned the theoretical basis for the Fundamental Theorem of Algebra, starting allllll the way from Groups. Sets me up well for understanding more theoretical stuff in the future.

62-142: Digital Photography I

So all the above classes are like exactly the type of tech classes I like to take, and I really enjoyed them. Yet, this humanities course with Prof. Dylan Vitone ended up being my favorite! Likely because it was very low-stress, and us students were just encouraged to have fun in it. The lectures were still very useful, learned a lot more about just how wide the world of photography really is, expanded my horizons for what pictures can be and look like.

I should post some of my assignments here; fairly proud of them and they deserve a place to show them off, even if it means breaking my rule of “no photos of me on the internet”.

Other Activities

TA for 15-316: Software Foundations of Security and Privacy

I really enjoyed this class with Prof. Matt Fredrikson last year, and was really excited to TA for it this time. All the content was just as cool as I remembered, and Prof. Fredrikson just as nice + caring.

There were a couple big changes this year, that I played a little bit of a role in bringing about as part of course staff: there are two new labs, written from scratch (mostly by Prof. Fredrikson’s herculean efforts) and both of them are freely available online.

The first goes into one approach to formal verification (instrument a program, run a SAT solver on the instrumented program)concretely applied to a toy language, embedded into a Python DSL.

The second shows how to build a simple theorem prover and we have an entire Python DSL for proof trees. While the goal of this lab was to just apply simple tactics to manually prove certain statements, one student ended up writing a general theorem prover for our language which I think is really cool. I plan on doing a blog post about this lab in more detail since I think it deserves more attention.

CMU AAC Officer

Not a whole lot to talk about yet. I’m running our fall zine (which I tried to participate in but then couldn’t complete…) which should be published before spring starts, and Secret Santa which has yet to start. Haven’t drawn nearly as much as I would’ve liked this semester…

Looking Forward To Spring

As it turns out, I only needed one more class to graduate¹! As such, next semester will be extremely light. I plan on auditing interesting classes anyways, but having the extra flexibility/less stress will be very nice.

A couple classes fill this requirement, so I’ll need to decide between 15-462 Computer Graphics, 15-330 Computer Security, or 15-455 Undergraduate Complexity Theory. Kinda leaning toward the last one tbh, only because it’s something I can only get at CMU. Heard mixed things about 15-330 but overall Computer Security is something I’m already interested in so it can’t be that bad. Graphics less so but likely still interesting. Thanks to a generous drop/tuition adjustment deadline, I’ll have 2 weeks to choose once the semester starts, we’ll see how it goes!

Footnotes

I could’ve taken that class this semester and graduated early but I didn’t count units right and thought I needed another full semester anyways so I didn’t whoops ↩

First Job Secured!

Mon, 12 Dec 2022 12:00:00 GMT

I have some big news: just signed my first job offer!!!!!!!!!!

Congrats! Where Are You Going?

Starting this summer, I’ll be working for Roam, a very small startup, doing Software Engineering of sorts.

What Does Roam Do?

Best way I can think of describing it is “a tool that makes a remote office that works”. Basically, actual virtual presence without any weird VR stuff.

I also wasn’t entirely convinced until I tried out their demo, and then I was hooked, really something you have to experience to believe.

How Did You Find Out About Them?

Professional Network. Super lucky to have already met one of their engineers and their CEO beforehand, and again lucky enough for my skills in “trying out just about anything computer” being a good fit for what they were looking for.

And seriously, I’m super excited to be working here, lots of great engineers to learn from and never-ending stream of interesting projects ahead. Looking forward to the summer!

What Makes Rust Amazing?

Thu, 27 Oct 2022 12:00:00 GMT

I’ve been thinking about this for a while, and I think it’s finally time to put these thoughts into a blog post. Will update this later to fill out missing sections but right now I just gotta put these thoughts down while I have some time.

Overview

References and Ownership
- Lifetimes
- Real Move Semantics
Good Defaults and Good Tooling
Culture of Safety
Algebraic Datatypes
Traits Instead of Objects

Note I have organized these in decreasing importance, at least in my view.

References and Ownership

This is the biggest thing ever. Like the other stuff I mention later on is really important too, but without this you don’t have Rust, you have something like Zig or D which while still very good, simply just can’t provide the same guarantees that Rust does in terms of memory safety.

A simple explanation of references in Rust: there are two types:

&T, an immutable a.k.a. shared reference
&mut T, a mutable a.k.a. exclusive reference

Coming from C++, the analogues are const T& and T&, and coming from pretty much any other language the analogues are… kind of nothing. If you have a reference to an object in Python, Java, Javascript, Go, etc., it’s a mutable one, with kind of um no guarantees around really really classic problems like thread safety and mutable aliasing.

A classic example of mutable aliasing, a.k.a reference invalidation, is the following:

#include <iostream>
#include <vector>

int main(int argc, char** argv) {
    // Create a vector
    std::vector<int> x = {1, 2, 3, 4};
    // Get a reference to the second element in the vector
    const auto& y = x[1];
    std::cout << y << std::endl;
    // Mutate the vector, invalidating `y`
    x.push_back(5);
    std::cout << x[4] << std::endl;
    // ***Insta-UB***
    std::cout << y << std::endl;
}

The funny thing, the absolutely hilarious thing is that, at the time of this writing, no modern C++ compiler (gcc, clang, MSVC) catches this in their standard compilation pass, even with all warnings enabled. The equivalent Rust code:

fn main() {
    let mut x = vec![1, 2, 3, 4];
    let y = &x[1];
    println!("{}", y);
    x.push(5);
    println!("{}", x[4]);
    println!("{}", y);
}

shows a very readable error messages explaining exactly how You Messed Up:

error[E0502]: cannot borrow `x` as mutable because it is also borrowed as immutable
 --> src/main.rs:5:5
  |
3 |     let y = &x[1];
  |              - immutable borrow occurs here
4 |     println!("{}", y);
5 |     x.push(5);
  |     ^^^^^^^^^ mutable borrow occurs here
6 |     println!("{}", x[4]);
7 |     println!("{}", y);
  |                    - immutable borrow later used here

For more information about this error, try `rustc --explain E0502`.
error: could not compile `playground` due to previous error

This is huge and prevents whole classes of memory errors simply by enforcing mutation to only be done by one “section” of code at a time.

Lifetimes

How this gets enforced is with lifetimes. A lifetime can be thought of as “how long a reference is good for”. In the above example, we got an error because we had a shared reference’s lifetime overlap with that of an exclusive reference.

Lifetimes are also useful because the maximum amount of time a lifetime can be valid for is the lifespan of an object, from when it is created (think malloc, but you can also have stack-only objects) to when it is dropped (think free or popping the stack frame).

Real Move Semantics

Ownership is the other mechanism that makes this possible. I distinguish Rust’s ownership and move semantics from C++‘s icky thing where a """moved""" object is still valid but """empty""" somehow because this is the Real Deal; once you move a Value (data in memory) out of a Variable (label in the program), that Variable is no longer valid. To even do a move in the first place, the compiler must guarantee that there are no outstanding references to that variable.

This mechanism allows the compiler to automatically insert calls to drop values when it detects a variable with ownership cannot possibly be used again. Again, this entire system is just a really neat way for the compiler to statically guarantee exclusive mutation of objects, which is much more complicated but just as important as doing bounds checks on all accesses.

TODO: explain this better

Good Defaults and Good Tooling

TODO

Culture of Safety

TODO

Algebraic Datatypes

TODO

Traits Instead of Objects

TODO

So yeah sorry it’s unfinished and a bit rambly, I promise I’ll get around to the rest of it eventually, at some point, later idk

Identity, Pt. 2

Thu, 20 Oct 2022 12:00:00 GMT

So maybe you’ve read my other post about this from way back, or maybe not, and in either case I think it deserves an update. TL;DR I’ve used an alternate name online for so long that it feels like another persona.

This is still going on huh

Yeah I don’t know if I was or wasn’t expecting it at the time but it’s been almost 2 years of this equilibrium, and I think at this point it’s safe to say it will probably continue indefinitely.

Do u have any idea why

I have two main hypotheses for why this is happening:

Internet Escapism
- factors mentioned in the previous post, i.e. it’s freeing to be less on-edge about how I’m presenting myself at all times
I’m Built Different
- this means whatever you think it means

Are you going to connect these identities anytime soon?

Highly unlikely! In any case, through my involvement in AAC, it’s kinda already happened (because I was careless)! It’s still tricky enough to connect the two, and I’d like to keep it that way for now.

So what’s the “update”?

The main thing that’s changed is I’m much more invested in this alternate identity. I have a separate website, more projects under the other GitHub account, more social media profiles with that handle, more art, etc. Some people I became friends with online refer to me by that name, in public, because that’s just who they know me as.

Another more recent change, and the thing which set off this post, is that I’m really starting to explore what this means with other people, now that there are enough people who know both identities and therefore know my conundrum. If you’re one of those people, and you’re reading this blog post, hi!

This is all still new for me, but in the meantime, I’ll keep doing my same ‘ol same ‘ol, being the best person I can be. Hope you’re doing so too, see you next time!

TA-ing

Wed, 19 Oct 2022 12:00:00 GMT

This semester, I’m a TA for 15-316, Software Foundations of Security and Privacy taught by Matt Fredrikson, who’s just an absolutely phenomenal professor.

While I have taught the Rust StuCo at CMU for a 2nd semester now, there are a couple of new duties involved in TA-ing that I’d like to give thoughts on.

Creating Homeworks

I am not very good at this (the Rust StuCo has had only around 2 “successful” homeworks in its run so far). Fortunately, Prof. Fredrikson is very good at this, and has a bunch of material to draw from already having taught this course a couple times before. He does most of the writeups, and the other TAs help come up with most of the questions, but I did help create a lot of the infrastructure for a completely-new Lab 1 this semester, so that was fun, and also a lot of time since I was learning things as I went along.

Grading

This part was less strenuous than I thought. We divide up the workload pretty evenly, and I’ve found some good gaps in my schedule where I can just plow through all 60 students once I get into a groove. As a student taking this course, I thought my answers were very clever and unique, and now I realize everyone here is like that, which is good since that means the vast majority of them are correct.

Holding Office Hours

This part is the most interesting part of the job in my view. I get to interact with students directly and try to get them to understand new subjects, something I already had a bit of experience doing from tutoring at CMU. The first couple weeks were extremely hectic, but now it seems people are coming less, I think because there’s less work? I’m not entirely sure, here’s hoping people come in more for this next lab…

In conclusion, TA-ing is a very enjoyable experience, especially since I’m getting paid pretty OK for how much I’m putting in hourly. The professor is nice, the students are smart, and the content is just good stuff that I can get excited about.

Also I’m back baby, post hiatus over! Maybe! I’m still more active on other platforms but I thought I might as well put something out here while I’m on Fall Break since otherwise I just won’t do it!

A Strong Defense of Copyright

Mon, 04 Jul 2022 12:00:00 GMT

First this I want to note is this is not just my writing: this is a discussion my friend and I have a discussion about how modern copyright has been affected by various forms of piracy. I thought it was really interesting and got permission to post it here under a pseudonym, Corporate Shill (CS). I’ve distinguished between their words (in green, the color of Money) and mine (in an innocent blue).

CS:

Question from watching too much youtube: I’ve noticed that VPN ads have shifted from security¹ to talking about how British Netflix has different shows.

I assume the regional differences are due to license agreements with the rights holders. How long until streaming services are compelled to take better steps to confirm their client location, or until “choose your server country” VPNs are targeted as a means of subverting restrictions or something?

Hazel:

That’s already happened, there was this evil blog post about how Amazon was partnering with a company to do exactly that (block VPN users)

https://twitter.com/geocomply/status/1522249780675551234

GeoComply (@GeoComply) Millions of U.S. adults use VPNs to illegally access streaming services in other countries. But you can easily block content pirates with GeoGuard VPN and proxy detection via @aws @cloudfront.

The original tweet and blog post was deleted, but there’s a backup on archive.org: https://web.archive.org/web/20220505221032/https://aws.amazon.com/blogs/media/blocking-illegal-viewers-from-streaming-services/

CS:

I love all the replies saying if you block this piracy they’ll do more explicit piracy.

Hazel:

I mean they (the users) are literally still paying for the content! And I get that licensing is a mess but ya to me this technology seems like revenue-losing only.

CS:

Well they’re not paying for the content, they’re paying for different content, deciding what they think they should be paying for, and gaining access by force.

Hazel:

“Different content” well sorry if I can’t control where I live…

CS:

The streaming service probably does better overall from a larger effective library, but I can’t imagine rights holders are happy to be paid for one thing while providing another.

You can choose what to buy and not buy, or instead choose something not for sale (on alternative platforms, individual sale, rental, etc). There are arguments about pirating things not available for legitimate sale and whether the region based licensing makes sense, but given the situation I don’t think blocking misuse is improper.

Hazel:

“Paid for one thing while providing another” I can’t pretend to understand how finances work but I can’t imagine how this would put licensers in a bad spot at all; like they aren’t actively providing anything.

CS:

A better way to go about this could be to bind location to account and limit how often it can change, idk why they don’t.

To respond to “not actively providing anything”, they (studios) have an up front cost to produce a piece of media, which in exchange they get copyright to the product. Maybe they want to sell to Netflix in the UK, but can get a good deal in the US with a particular TV channel. Netflix pays less than a worldwide license, (while the content is still available worldwide through VPNs), so the rights holder gets less money, even assuming the US market is still fully available.

Hazel:

Still not sure why the rightsholder didn’t go for the thing that would make them the most money in the first place?

CS:

The actual financial relationships are weird, it’s like “can this company continue to fund this studio under the assumption that they can produce products that on average over their lifetime can produce a profit?”, lots of lag time and middlemen involved.

In the scenario I described, someone is willing to pay a premium rate, but only in a certain region; that’s the US TV channel operating in a single country that wants an exclusive show to compete with regional and global channels and services. The buyer (Netflix) doesn’t want a worldwide license because of language and broadcast regulations and transmission licenses and budget and marketing scale or any number of possibilities, but the seller still wants to access other regions that don’t interfere.

idk how the separate netflixes and all work, I would assume they’re all a single interest, but as long as there are regional players in the market, I can imagine regional contracts making sense.

Hazel:

I now get why they would sell to a single streaming region (actually clears up a lot of my confusion, thanks), but still don’t see how region spoofing eats into these profits?

CS:

In the case where there is a legitimate alternative delivery, you are circumventing that.

Hazel:

Unless the legitimate alternate delivery is another pay-per-view streaming service, hasn’t the money has already been made? Like advertisers already paid for ad slots based on average ratings and all.

CS:

In my example the TV channel gets less than they paid for and Netflix gets more.

The rights holder gets the same for completed sales but loses trust in future deals.

Also think about it: where do the average ratings come from? :P

Hazel:

No clue, probably the cable box keeps track?

CS:

I mean, they come from past statistics. By changing present statistics, you change future averages.

It’s another point of lag, but the effect still happens eventually.

Hazel:

Ah! That is true.

CS:

The next season comes around, and the US TV channel can’t trust the studio’s exclusivity contract not to be circumvented by VPN users and made worthless, advertisers can’t expect the old amount of impressions, the parent company of the studio (if one exists) can’t expect licenses of future products to be as valuable.

I don’t really like artificial exclusivity of media in general, but going around it seems to just concentrate success in the streaming services that have the weakest detection.

Hazel:

Just like capitalism intended 😤

That was a good explanation of why artificial exclusivity can be helpful for studio longevity, but on moral grounds it feels weird ya

CS:

Circumvention doesn’t feel like a functional way of influencing the change you actually want, just a short term financial imbalance until streaming services change or stop getting sold.

Yeah if we’re just going moral as a consumer, there’s also a stance that you can steal from any big company doing well 😜. Most of my Disney lately has been basically pirated, thinking about it but I’ve also watched a lot less since that habit started, and I probably would watch new stuff in theaters if the situation was right.

“What are your crimes? comment below with your full name and address”

Hazel:

Hello I torrent a lot

of Linux installers 😎

CS:

:ooooo

Hazel:

also question: do you mind if I put our discussion text on my blog? appropriately attributed of course

CS:

As an “Anonymized Corporate Shill”; feel free, but I have no research or numbers or anything behind this, it’s just a logical sequence that I think would exist and explain what we see.

Hazel:

I do like logical sequences and also the existing stuff has no rigor so it’s thematic :P

CS:

heck yeah

Also fun piracy regionality subject I see frequently: arcade games

Hazel:

Round 1’s “This machine should not be distributed outside Japan” lmao

CS:

Lots of stuff is not even sold, individuals may not have operators in their local area, and versions/games may go out of date and no longer have legitimate purchase options for that exact product. It’s a breeding ground for illegal piracy.

Oh yeah they’re also full of licensed content lmao I still never figured out how that works

Hazel:

Maybe small enough market that it doesn’t have an impact on the bigger licensing picture? And again the “not legally available” just creates new (illegal resale) markets.

CS:

Like, Konami networking is by special contract with R1² (and D&B³ to a lesser extent), and on a technical level may work as an extension of Japan or its own region.

There’s also the question that I think often gets ignored about loss of arcade machine sales. Maybe Pop’n Music 4 isn’t run because of newer versions existing, or Museca being shut down entirely, but there are other experiences like them.

If you weren’t playing Chuni⁴ on data, at home, for free, whenever you wanted, you might be making choices between steam offerings and local arcade options, or even the new home game versions from Konami.

The standard for an “old” game in the community tends to be between 6 months and a year, with game data being released when fairly similar games are still being sold and operated.

Hazel:

WOW that’s fast turnover. Does this mean resale eats more or less into profits compared to a longer release cycle?

CS:

TODO: reword the following: This common commitment to “n-1”, i.e. releasing full unlock updates for versions at least 1 behind current may work for those who want the latest features now, but honestly getting game updates with full 3rd party server support and bonus features on a 6 month delay is insane.

Wacca is still live for a bit to my knowledge, but the community is already deep into making a version compiling all versions, custom chart maker, etc.

This is really confusing because we again have this studio/provider/consumer structure like with the video content but also there’s custom physical hardware, there are live service elements, the consumer payment is pay per view and revshared between publisher and operator. Konami actually charges the arcade 15¢ per credit or something, I think Sega sells it up front but it’s the same model.

Hazel:

Oh wow! So resales really would directly impact their bottom line huh unless there’s some DRM forcing this payment to Konami somehow.

CS:

I think there is also a monthly fee for the live service and updates.

Yeah so online features are a big thing, some games refuse to boot without a server connection, others have limited features and severely limited song selection as well as no access to unlocks and player save data. Operating legitimately during the life of the live service is poor if not impossible. At end of life, I think offline patches/kits are common, but these often are not the sum of everything that has been, taking away licensed pop music and whatever else they decide.

For example, Chunithm went through a cabinet change to become Chunithm NEW, with the old cabinets getting stuck on an offline patch of the latest version pre-NEW, Paradise Lost. For R1 Japan, these cabinets have very low value, especially if they’re getting the new ones, so they shipped them to America where Chunithm was not previously operated by them. The song library is severely gutted, and especially with the lack of save data, no enthusiast plays it.

This could create a pressure to acquire the new cabinets where there is demand, or push players to the games that are available, but emulation is an option. You can actually skip the cabinet part altogether and use a third party controller to play on data.

Hazel:

Ah yeah that’s the type of DRM I figured.

Emulation just works by ripping the data as it goes over the network and parsing it in a separate machine?

CS:

Off the drive mainly, but third party servers are made too. The game program is modified to add back in songs whose licenses lapsed, subvert online checks, etc.

There are a lot of capital costs, secondary markets, tertiary markets, and copyright that make everything complex and slow moving. However, I would assert that emulation and modification has set up a condition that would make a legitimate entry to the US market less successful. Like, maybe people will be moral and stop the frequent dumps if access were expanded, and additional interest will have been generated, but it just seems like too much uncertainty for the publisher to deal with.

Hazel:

Apple shows you can push the jailbreak game pretty far, seems like this is still lax at the moment.

CS:

Oh boy you have no idea: the encryption, the VPN routers, the multiple types of security dongles, it’s clear that they have responded to activity multiple times lol

Hazel:

Ah! Nevermind that is pretty advanced.

Ya it’s a bit unfortunate that there is enough hype to make it profitable, but now they’re competing against free which is just impossible.

CS:

Yeah, and especially if someone bought a controller instead of going to a public cab if you got the new 400chunicontroller(probablyexpensivetoshiptoo),andthenNEWcameout5minutesfromyou,wouldyoureallyditchyourhomedeviceandpay400 chuni controller (probably expensive to ship too), and then NEW came out 5 minutes from you, would you _really_ ditch your home device and pay 400chunicontroller(probablyexpensivetoshiptoo),andthenNEWcameout5minutesfromyou,wouldyoureallyditchyourhomedeviceandpay1/play?

Hazel:

No way!

I’m really starting to see why studio execs are so freaked out about piracy, since it fundamentally changes the competitive landscape in a way that doesn’t favor how they currently do business.

CS:

Also! There are simulators, free clones of games that don’t even require an arcade machine or any hacking.

I don’t know much about Clone Hero, Stepmania, and all the JP rhythm ones but they seem more if not completely legal and then you can have community features where people transcribe the stuff that comes out in the original game.

Hazel:

Oh really? I would’ve though the music licensing is still weird for that, similar to Osu! Transcription also seems like a whole separate minefield wow.

CS:

https://youtu.be/xXN9adKkmQE hey look DrumMania licensed song and level data from the latest version available in DTXMania

I’d actually argue in favor of the chart videos generally, I feel like they’re not a replacement and encourage exploration, interest, habit forming the dancerush charts are taken from direct video output on legitimate cabs.

Hazel:

Sure, those are a very different experience than playing directly

CS:

But when it’s something playable on a sim that gives feedback and score… and then from a rights perspective too, is the level design data copyrightable?

Hazel:

I would think so yes

CS:

I would think so too, but apparently like board game rules aren’t. Rules text may be, “mechanics” might be the word I heard.

Hazel:

Oh wow that’s interesting! Wonder what the distinction is/where the line is drawn.

CS:

You too can clone monopoly with unique art assets!

But yeah, you get this weird place where all the parts are mostly copied, but by different people with different scales and distribution methods. DTX⁵ is especially funny to me because the charts are usually very dependant on the song, driven by the composition of the music. There is some room for originality, but it’s totally possible that someone could make a chart for an anime song that’s exactly the same as what Konami makes when they get to licensing it.

Hazel:

brb about to make a game whose rules depend on the order of every single word in the rulebook so i can copyright the “mechanics” too

CS:

So is Konami the one that has the copyright there? Is it different if we assume it was copied/inspired rather than derived from the original source?

https://youtu.be/dwcb0YasK68 side note I enjoyed this one a good bit; note the distribution method of the video being unlicensed music working over YouTube’s special agreements :P

Hazel:

I think this would fall under the cleanroom exceptions as understood by the UNIX clone of MINIX and Google v Oracle⁶. Not entirely sure how that works but iirc deriving the same thing from the same source idea is fine and not copyright-infringing.

Also YouTube music LMAO

CS:

I remember Twitch and Facebook Live released music things at similar times

Hazel:

side side note every music platform is kinda inconsistent in their metadata tagging/normalization

Wait FB has music stuff?? I had no idea, wouldn’t be surprised tho their app is huge

CS:

Twitch had a whole thing where they made deals with tons of indie publishers and added an integrated player and tried to create an environment that could supply music with rights acquired like. It seems like a lot of effort and mone, but people hated it because you couldn’t broadcast whatever pop music they already wanted to listen to, also DMCA strikes.

Meanwhile I think Facebook managed a YouTube-style deal with all the major labels for their live platform, idk if there’s revshare or flat cuts or pay per play or view or what but it’s like YouTube’s thing to me, sounds impossible, probably nearly is, completely changes copyright in practice in a single space, but the space is really big it kinda feels like there’s reform that was needed to match a culture, but a private company got it done first, and thus lowered the pressure people talk about adoption and economies of scale with youtube competitors, but their rights agreements seem like a big part of what let’s you just kinda upload whatever and have it usually be fine.

Hazel:

Oh yeah that Twitch thing!! dogdogHS still gets hit by mutes for playing from Spotify playlists lol

(also big agree on the impossibility of YouTube licensing agreement, probably only works because they’re so big)

CS:

just on the vods tho?

Hazel:

CS:

Classic

And another example, it’s not like it cuts music sales, people don’t go to his stream for free Katy Perry. But also, he definitely doesn’t have the rights, is aiming to entertain thousands for commercial gain, and presumably if he’s playing it it enhances the show and thus makes him money

Hazel:

it does enhance the show and he is making bank

Going back to YouTube, there are still DMCA strikes, but yeah most of the time it’s just “hey there was music playing in this so I can’t make money off it anymore also we’ll link to the original song” and the original video stays up at least.

Still not a great automated system if someone really does have rights or gets hit with bogus claims or is using it under fair use.

CS:

Yeah, compared to mass piracy using your platform and delayed takedown notices you might pass to users and compelled moderation it’s just done like it’s still annoying and abused often, but you don’t have to be careful in order to actually do things perfectly, and you don’t get usually hit with stuff later, and the ads/link offer a middle ground solution with no custom (expensive?) negotiation needed for every individual case.

This has trailed off a lot lol but yeah, arts economics and rights are weird.

Hazel:

lol yup may be my fault a little for all the random questions :P

CS:

Oh I trailed down plenty of my own rabbit holes too

I’m sure though even with other structures besides Capitalism, valuing art is a hard thing to do

Hazel:

Oh totally, I have no idea what would be a better alternative. Just the logical conclusions of all this are wild

CS:

I feel like when you see those government arts funds, they tend to have a specific cultural goal in mind, and competition for their money. This couldmove art out of the true culture and put value on presentation rather than substance.

Hazel:

what even is “true culture” anyways? anime feet THIS IS A JOKE

CS:

Man of culture screenshot 🤣

The market system does this too I guess, but it’s more random because the people with the money aren’t all together :P

Hazel:

“People with the money aren’t all together” I have some bad news for you buddy

CS:

I’m pretty sure there’s at least two groups, one of them is Disney, I’ll get back to you on the other.

Excited to buy my internet plan that comes bundled with cable tv, Disney+, ESPN+, and whatever the third thing that’s still Disney is. I love choice.

CS:

IT’S HULU?! c’mon man…

Hazel:

Hulu is also Disney??? that’s wild

CS:

I think it’s a comarketing thing, not a firm corporate structure relationship

Hazel:

“I can’t believe there are so many streaming services we have to pay for” they’re all just Disney in different hats

CS:

I do not have the rights to host this audio luckily I’m not please buy terraria and the terraria soundtrack

Hazel:

Does TikTok? Does Discord? (Does my website linking to Discord?)

CS:

TikTok seems to viral for anything to be stopped; Discord seems to big to be convinced to check

Hazel:

And my website is too small to matter :P

Ok that was it, sorry for the fizzling-out conclusion, thanks for reading all the way through hope you got something from it!

Footnotes

Good debunk video by Tom Scott: https://www.youtube.com/watch?v=WVDQEoe6ZWY ↩
Round 1, an arcade franchise in the US ↩
Dave and Busters, a more well-known arcade franchise in the US ↩
Short for Chunithm, an arcade rythm game made by SEGA with open-source controllers ↩
A rhythm game played with a lifelike electric drumset controller, where the maps are like actually playing the drums to songs ↩
Don’t quote me on this I am not that good at computing legal history ↩

I Easily Get Lost In Cities

Fri, 17 Jun 2022 12:00:00 GMT

Whoops, forgot to actually push to the site, in case you couldn’t’ve guessed I have been busy with my internship :P. Multiple post dump time!

I don’t know if you’ve heard, but New York City is a big city. There are many skyscrapers in Manhattan (like the central island thing containing Central Park), all bunched up right against each other. The streets are long and gridlike, and sometimes the sun sets exactly in-line with them so you get to fully see all the dust in the air.

At least where I’m staying, the streets are relatively clean, only a slight amount of litter left out for the pidgeons. There have also only been a few instances of crazy cars (in bike lanes, taking lefts into walk signs) and for the most part because traffic moves so slowly anyways it’s pretty safe to be a pedestrian in that regard.

Ok but like how do you get lost when the streets are all numbered?

Great question! I’m not entirely sure myself! I have a hypothesis:

I usually navigate in relation to (non-sun) landmarks, and the buildings all occlude each other so that doesn’t work out.

Especially in Central Park, where there aren’t any marked roads you can see, it was exceedingly easy for me to get lost. Not just once, but twice, I went out for out-and-back runs there that ended up being out-and-out so I was double the distance away… Learned my lesson there and now I run with Google Maps :P

However, this is just my go-to navigation mechanism; surely attempting to actually see what corner I’m on would help me get where I want to go?

I keep forgetting to orient myself at intersections, and have easily lost orientation when taking multiple turns.

Especially with east-west streets, I’ve gone very long blocks in the wrong direction before realizing it (yes I know I could’ve just looked for the shorter north-south directions, but remember i am dum)

GPS/location services is very spotty.

… which doesn’t help with any of the above. I’ve heard it’s because GPS signals reflect off buildings and the supposed Wi-Fi/Bluetooth MAC address-based location services never really panned out like it promised to.

This Also Happened In Seattle

Oh yeah I should mention! The team I’m interning for had their calibration this week in Seattle, so I was invited along for expensed business flights and hotel :) Seattle is a neat city too (much smaller), had the best sushi of my life so far there.

The place near my hotel had street grids at 45 degree offsets from each other, navigating which was a bit funky. Overall, I did a lot better than NYC (only got lost once!), possibly because there were easier-to-use landmarks, and also possibly because I’ve gotten more in the habit of navigating grids.

I did not go up the Space Needle unfortunately, but I did go on the monorail and light rail which were cool. Seeing it and hearing good things about Redmond too is slightly making me consider west coast more… nah, east coast for lyfe 😤

I'm In New York!

Tue, 31 May 2022 12:00:00 GMT

How Did I Get Here

As I’ve mentioned previously, I’m interning for Meta (formerly Facebook) this summer. There was a location selection process, and I ended up choosing + getting assigned to New York City (east coast baby!). Unlike last summer, which was entirely remote, this internship is in-person so I’m getting flown out to, having housing provided for, and paid to live in The Big Apple. (getting very spoiled i know)

How Was The Flight

Fairly standard, exiting LGA was pretty quick (although I’ve heard not great things about getting in). Taking a Lyft from the airport wasn’t bad either.

Things I’m Looking Forward Too

I’ve never lived in as big of or as famous of a city as New York (understandably, since few cities besides it can claim such a status). While I’m here for a full 12 weeks, I’d like to be able to visit a good number of museums, eat at some famous restaurants and some local restaurants, see a Broadway show, and overall just enjoy all the other activities going on.

The place I’m staying is right next to Central Park (2-3 blocks iirc) which is pretty epic ngl, I’ll likely be spending a lot of time there. The “long-stay hotel” also has some good amenities I hope to enjoy.

Things I’m Concerned About

It looks like I will need to commute to work, so taking the subway regularly will be a first for me.

Tech pickup also has a tight window, I think I can make it in time, they should be fairly lenient regardless.

Overall, it’s pretty exciting to be in Da Big City, looking forward to my first week!

QOTD: 15

Mon, 30 May 2022 12:00:00 GMT

Wow, I haven’t seen one of those before!

TSA agent at DCA, commenting on my fightstick decorated with anime stickers

am i bringing these back actually? probably not lol

What I've Been Up To Pre-Internship

Tue, 24 May 2022 12:00:00 GMT

It’s been very nice to relax after that semester, I needed it.

Due to large amounts of relaxing, hanging out with family, petting/walking my remaining dog (R.I.P. Calvin, you are currently missed a lot and will continue to be missed because you were the bestest and fluffiest and cutest dog), I haven’t been doing a whole lot of computering. However, I do have a carnal need for stimulation, and have satisfied my need by casually working on the following projects:

UEFI Variable Writer for Windows, Written In Rust

I had worked on this a bit last summer, but it didn’t fully work for reasons I wasn’t entirely clear on. Turns out I had everything correct (linking with windows-rs => acquiring SYSTEM token for a thread => calling the SetFirmwareEnvironmentVariableA function) except for not NULL-terminating the string I wrote (Rust doesn’t do that for its strings, C needs it). So after fixing that, it works!

“But Hazel,” I hear you ask, “why did you want such an obscure tool in the first place? Playing around with firmware variables is scary!” Well, my dear reader, I wanted such a tool for my Arch Linux dual-boot, because booting from Windows into Arch required mashing a key on boot which I often forgot to do and then I would have to reboot all over again and it was slightly annoying so I did what any good programmer would do and wrote a tool to automate it.

The bootloader I use, systemd-boot (which imo may just be the best bootloader for UEFI btw), checks for a “LoaderEntryDefault” UEFI variable and boots that configuration if it exists. This is very handy and is how the included bootctl tool changes the default boot without actually modifying the config file living in the EFI system partition (very different from GRUB).

My goal, then, was to be able to write to that same variable, in the same format, from Windows, so I could easily do something like windows-bootctl set-default arch.conf and on the next reboot my computer would boot into Arch without me having to spam any keys!

So about the code: i uh put it under my other identity which has also been very active on Twitter (and I posted lots of prototypes there so). If you’re very curious, I’ve been informed it’s not impossible (quite feasible even) to link the two with public online info, just please keep it a secret if u do find out.

OpenAFS For Windows

I’ve been on this since last year and I think the only reason I didn’t make a post about it at the time is because I didn’t finish. I’m still not done, but have made enough progress that I think I should share.

Some motivation for why I wanted to do this: CMU uses AFS as a shared file system, useful for a lot of CS classes because your files will be available from any machine you SSH into. Currently, I can download those files with WinSCP pretty easily, but it’s slightly annoying to have to open the app and click “login” every time. Builds of OpenAFS for Windows already exist, both official and unofficial (unofficial preferred since it backported a few security patches), but I have a compulsion to stay on the bleeding edge so I thought I’d give it a shot myself.

Oh boy was I in for a surprise. This is hands-down the gnarliest codebase I’ve ever encountered. A succinct way to describe it is that the last big upgrade to the code seems to be to support Windows 2000. There are still files that manually load function pointers from external DLLs (via certain syscalls) without relying on the OS. There were many <stdint.h> headers missing, #defined names that conflicted with those in internal WDK headers, and an ungodly number of compiler warnings.

In the end, after 67 files changed, 4159 insertions(+), 2034 deletions(-) (most of that in Visual Studio .vcxproj and related files tbh (which I needed to add since their driver build system was not longer supported past Windows 7)), I was able to get all their files to build! However, that’s not the end of the story: all of the executables complain about a “incorrect side-by-side configuration” and the drivers don’t want to install since they’re not signed.

I’m focusing on the driver issues first, since being able to develop drivers on Windows seems more applicable outside of this one project, but uh that’s sort of taking a while since Visual Studio isn’t cooperating (and the smallest possible WinDbg setup is very complex).

Doing Art

I’ve been voted in as Tech Chair for AAC for the second year in a row! (I was the only one who ran for that position (but also there were no votes of no confidence so that was nice)). Somewhat related to that, I’ve been trying to do a lot more art, studying from Micheal Hampton’s book Figure Drawing: Design and Invention. I can see the improvement in my sketches just after doing a few exercises; being able to “see” what the shapes are helps a lot with recreating them in an image.

I’ll be contributing to the club’s upcoming collaborative banner (past banners here, i was a part of both :)), and probably participating in our internal ArtFight in August. Drawing is just fun and I’m glad I can be part of this club even if I’m not at the same level :).

Looking forward

This next week I’ll be doing more of the same, still mostly relaxing, and ramping up prep for my internship at Meta, working with the Community App Health Engineering team in New York City. I am not sure what the team does yet and also not sure where I’ll be living but I’ve heard other people are in the same spot so here’s hoping it’ll work out.

The OS Semester

Wed, 11 May 2022 12:00:00 GMT

Wowza, that was quite the semester. A list of the classes I took, also the order I’ll cover them in:

10-405 Machine Learning With Large Datasets
17-200 Ethics And Policy Issues In Computing
21-301 Combinatorics
15-410 Operating System Design And Implementation

10-405

Taught by Prof. Virginia Smith, this class was an introduction into how you’d actually construct machine learning systems to perform well at scale. “At scale” usually meant “on multiple computers”, which is indeed a challenge due to things like communication overhead. We started small, covering things like distributed Linear Regression and distributed PCA, and ended with state-of-the-art approaches to Neural Architecture search and Federated Learning.

Overall, I liked this class a lot. The lectures were very good and the homeworks chill. One common complaint about homeworks that I share is that they were sometimes too structured; there was a lot of “fill in this function signature that we’ve made for you already”. The later homeworks did a better job of this, and I can understand why they did this in the first place (makes grading easier and eases the learning curve for frameworks like Apache Spark and Tensorflow).

This class was within its unit count of 9hrs/week.

17-200

Co-taught by Prof. Michael Skirpan and Prof. Sarah Fox, this class was a very practical overview of contemporary policy issues in computing and how to apply ethical thinking to those. I really liked this compared to the previous ethics class I took, which covered ethical frameworks in a more abstract term; getting to learn about surveillance as “instrumentarian power”, unionization movements in big tech, and dark patterns in UX felt much more real. This class humbled me with the realization of just how much power is wielded by broken systems, but also gave some concrete paths for improvement as future tech workers.

This class fulfilled my writing requirement, and overall the weekly writing assignments and writing midterm/final weren’t all that bad. I only wish I had more time to work on the assignments (no fault of the class, just busy with all other classes too) and was better at the writing process (mental block on drafting + reviewing…).

This class was within its unit count of 9hrs/week.

21-301

Taught by Prof. Irina Gheorghiciuc, this class covered some novel ways of, you guessed it, counting things. It started off simple like “what a combination is” and then quickly progressed to wild things like using functions and their derivatives (whose coefficients of their Taylor polynomial are the things we want to count) to do counting, then expanding on that to cool theorems like Hands and Decks, Lagrangian Inversion, and Inclusion/Exclusion. Very cool to know that u(x)=xϕ(u(x))u(x) = x\phi(u(x))u(x)=xϕ(u(x)) has a unique solution,

[xn]{u(x)}=1n[xn−1]{(ϕ(x))n}[x^n]\left\{u(x)\right\}=\frac{1}{n}[x^{n-1}]\left\{\left(\phi(x)\right)^n\right\}[xn]{u(x)}=n1[xn−1]{(ϕ(x))n}

Absolutely magic how this is proved :)

I can see why all my math major friends meme about Irina’s lectures (very unique and humorous style) and her tests (some problems are quite hard). I really enjoyed this class; while I probably won’t get to use much directly later on, it’s nice to keep my mathematical thinking sharp.

This class was within its unit count of 9hrs/week.

15-410

Taught by Prof. Dave Eckhardt, this class covered everything you need to know about writing a simple UNIX-inspired OS kernel, and then some. x86 cdecl stack layout and calling convention, x86 memory model/atomic operations, x86 interrupts (traps, exceptions, faults, hardware), x86 segmentation, x86 virtual memory mapping, how to structure a large software project, the thread/process abstraction model (and how to implement it), concurrency problems like deadlock, syscalls (and how to implement them), and more stuff that we didn’t get around to implementing like filesystems, transactions, IPC/RPC, paravirtualization, memory consistency, and the experimental CHERI architecture. In case you couldn’t tell from the previous sentence, we wrote our OS for the x86 architecture and got very familiar with its idiosyncrasies. In the end, we had a lightly-featured (and working!!) kernel with a simple threading library able to run some useful multithreaded programs with keyboard input and console output.

The class moved at a breakneck pace to cover all of these topics (which were all covered in a lot of detail!); there was hardly a point in lecture where I wasn’t moving my pencil. There was also hardly a day past the first week where I wasn’t working on the OS project; a final sloccount of the kernel showed 5000+ lines of mixed C and Assembly (however it counts that (and there was probably more documentation than that too)).

This class was not within its unit count of 15hrs/week :) More like 25 I think? Didn’t keep too close track, just estimating from the times I tended to program every day.

15-410: It’s A Partner Class

For those familiar with 15-410, you may recall that it’s mainly a partner class; 10 of the 14 weeks are spent working on projects with another person.

This did not turn out well.

I didn’t realize that people generally find OS partners before class starts, so everyone I already knew in the class had already found someone by the time I started looking. In a historic twist of fate, I talked to the person who always sat next to me in the front row, who turned out also to not have a partner, and we quickly made a group.

Keep in mind, I had never even seen this person before this class. I had no idea how this was going to go, but given the questions they asked in class (and the fact they showed up to class at all) I was fairly optimistic.

They turned out to have many issues. Mental health, physical health, and family issues, all in a negative feedback cycle. Naturally, this precluded getting work done, often in week-long stretches during schedules where missing a single day could be a bad setback.

I ended up designing and implementing every single syscall, and every submodule to support all those syscalls. They contributed some good code and helped walk through design decisions (especially in the P2 and P4 projects before and after the big P3 kernel), but overall I’m still a bit salty about the work distribution even though nothing could really be done.

The course staff knew about these issues and we were awarded generous (and very necessary) late days. I’m (moderately) proud of what we turned in, but holy cow I don’t want to crunch that hard ever again.

Random Other Stuff

OS sapped all my time and I couldn’t make many late-night or weekend Frisbee practices this semester :(
My knee is fully better by now though, been working out and going for runs just fine
I’ve been voted in as Tech Chair for AAC for the second year in a row! Wow thank you!
Speaking of, I’ll try to do more sketching practice this summer. I completed Figuary again this year and have been using Hampton’s book a lot and I can see the results (from just the parts I’ve read tho)
StuCo thoughts coming soon! (when I finish “grading” the finals)

Wow, What A Year

Fri, 31 Dec 2021 12:00:00 GMT

2021 was a year mainly full of one thing: COVID. From stunting last spring semester to be entirely online (first time living alone!), to lessening in the summer and fall, and now variants coming back with a vengance, the overall pace of the year felt slowed down by the virus.

It’s hard to believe this is the same year I learned OpenGL and C++ to make a simple demo, or ran a 6-minute mile; those events feel like so long ago (yes they were but still). Other big events were renting my first apartment, interning at facebook, and getting sick with Lyme’s Disease.

Besides all the COVID variants, there was another huge event in 2021 for the history books: the Jan 6 insurrection against Congress ratifying the presidential election. Trump’s handling and encouragement of that mess definetly deserved him getting kicked off all major social media platforms.

With Biden as president, “normal” has been somewhat achieved: a Congress that hardly passes any bills of import, and a constant dissatisfaction with the executive branch for not making that happen. It depresses me how ineffective the national government is, and all the state governments banning abortion and mask mandates don’t help either. Count myself lucky to not be too affected (yet).

Anyways enough with the bad news from last year, what am I looking forward to in 2022? Some main things:

I’m teaching a StuCo on Rust alongside Cooper Pierce next semester! Very excited for this, we both enjoy Rust alot and started this almost as a meme but now it’s a real thing that 39 (!!) other students have signed up for. Hype 🦀
I’m taking 15-410 Operating Systems next semester! This is “the big one”, all of the 15-4xx systems classes at CMU are known to be very difficult but very rewarding in term of what you learn. Due to the expected increased time commitment, I’ll need to stop tutoring though, which is a bit of a bummer since I enjoyed it this semester. But! OS awaits, and I can’t wait!
Getting back to playing Ultimate Frisbee! It took all winter, but my knee is finally healed after I busted it doing a heavy lift right after I was cured of Lyme, having not exercised in a month. I’ve been doing other lifts and track workouts since, and it’s held up fine so I can’t wait to actually play. Hoping that practices aren’t scheduled at the same time as class again…
I’ll be interning at Facebook, now Meta again this summer! I got a return offer and took it, love being able to skip interview season and relax! Again, I’m still apprehensive about working for them given some of the relevations that came out this year, but on the whole I expect it to be fine.
I have no idea what next fall semester will be like!

That’s it for me this year, hope your 2021 was bearable and I’ll see you in 2022!

Made it to Finals Week!

Sat, 04 Dec 2021 12:00:00 GMT

As I sit here, the day after a hectic final week of classes and 2 days before finals week proper, I start to reflect on just how much has happened this Fall 2021 semester. I accepted Facebook’s return internship offer, signed a lease for next year’s apartment, saw my family for Thanksgiving, achieved a 2-plate bench, and besides that…

Hm. I guess I have been focusing on school a bit too much. The past semesters have been like this as well; I hardly have time to hang out with friends, let alone work on cool projects.

I’m probably also being impacted by burnout right now. Because no assignments could be due the Monday or Tuesday after Thanksgiving, all assignments were due on Wednesday, Thursday, and Friday this week. Thanks to not working over break (which I needed), I had to grind hard this week (which I didn’t need). In the end, I made it through, turned everything in while maintaining my usual quality, but at the cost of just not wanting to work on anything for a while.

There are still finals on Monday and Tuesday though, so I’m spending today to just recover and will spend Sunday doing whatever review I need to do. These aren’t finals I can just cruise through, so I am still a bit stressed about, and more stress is the exact opposite of what I need right now.

Despite all of this, I actually do have something I’d like to show! My Deep Learning Research Project, which I’ve back-dated to the day it was due. There’s more details in the post itself, go check it out!

Deep Learning Research Project

Fri, 03 Dec 2021 12:00:00 GMT

This semester, my group and I at CMU did some simple research for 10-417, Intermediate Deep Learning. By “group” I mean fellow undergraduates in the class. You can view the paper here.

Our project inspired by the arXiv dataset on Kaggle, specifically one of the tasks, “Title prediction by abstract”. That seemed pretty cool, and a good opportunity to learn about modern NLP models, so we dove right in.

Looking back on our progress, I probably should’ve spent less time reading and debugging frameworks that didn’t work, and more time training models. I read a lot about Language Models, models that predict how likely a given piece of text is to be non-gibberish, only to come to the conclusion they aren’t very good at summarization anyways. Still, trying to understand them to the level that I attempted to implement them was very helpful to understand the basis behind a lot of other techniques; I got to see how everything “fit together” in a sense.

One of the main takeaways from the project was that this task is a lot harder than it seemed! Many titles are pretty creative, not having an obvious relation with the abstract or using non-standard lingustic constructs. Other titles were able to be predicted a lot more easily, and on the whole there was a wide range of prediction accuracy. Seeing predicted titles was humorous at times, I wonder if I should set up a Twitter bot or something…

TL;DR paper here, didn’t get good results but learned a lot.

Wait, How'd I Miss Semptember?

Mon, 04 Oct 2021 12:00:00 GMT

Wow, what a month. I’m back at college, and was very much immediately thrust into all sorts of activities as school started. The rare occasion of 1) remembering the blog and 2) having time to update it has finally come to pass, so here we are: A List Of All Memorable Things That Have Happened

I Messed Up My Knee(s)

The first gym workout back. The warm-up set of squats. I went too low and injured my left patellar tendon (the one on the front of the knee). Not knowing the severity at the time, I rested it off for a week. Then went for a run and messed it up again. Then waited another week to see a nurse who said “ya you shouldn’t be doing anything rn kiddo”.

As an added bonus, my right knee now has similar symptoms because any time I needed to do a squat-like motion in daily life, I just did a single-leg squat with it. So now both knees are out of comission for a while. This sucks a lot because…

Frisbee Is Back

And I made the B team! I was hoping to make the A team this year, but injury and also the fact that there are many people better than I means I’ll be on the more relaxing team again. This isn’t a terrible thing, I just wish I wasn’t injured so I could put more time into it.

I’m a Club Officer Now??

I forget if I’ve mentioned this before, but for Artist Alley Club, I am officially the “Tech Chair” in charge of technical matters like organizing the Google Drive and setting up bots in the Discord. Very chill, running the activities is fun. Oh yeah here’s that art I promised earlier:

This is the piece I did for AAC's Summer 21 Zine

Looking For Research

If you are a professor I emailed while looking for research and you found this blog, hi! Hope it isn’t too messy :)

So basically, I have a return offer for Facebook next summer, but I would prefer to do research at CMU if at all possible. I’m doing a research-ish project for 10-417 this semester, and should be able to make time this Spring to do research if things don’t work out for the summer.

I Am Still Bad At Fighting Games

I bought Guilty Gear Strive for PC, and boy am I just getting my butt handed to me by other players. There is a Fighting Game Community (FGC) at CMU, and when I showed up to a local I was by far the worst player there. I really want to like these games, but man is it fustrating to lose over and over and not really know what to do about it besides “play more and maybe the instincts will come”.

Using a New Shell

I’m trying out Nu Shell on Windows as my default shell, and so far it’s going pretty well. The only hiccup I’ve encountered so far is that Python virtual environments don’t work too well in it, since they rely on Batch files to set up environment variables and Nu isn’t compatible with that. Other than that, it has good tab complete and simple Git integration, worth seeing where it goes. I’m just happy to have another shell rn lol

Ok phew that was all the stuff I wanted to talk about. Sorry it’s a little too self-centered, but I guess that’s what a blog is supposed to be anyways. Until next time!

OpenSSH and Kerberos

Wed, 25 Aug 2021 12:00:00 GMT

Recently, with the extra free time provided by being off the internship & not allowed to exercise, I resumed my quest to sign into unix.andrew.cmu.edu without needing to enter my password every time.

I had actually accomplished this in the past on Linux. I installed MIT Kerberos with a package manager, copied the /etc/krb5.conf file from the remote server, signed in with kinit, enabled using Kerberos with GSSAPIAuthentication yes and GSSAPIDelegateCredentials yes for the server profile, and bam it worked.

However, now I use Windows full-time, not even touching MSYS2, and want it to work there as well. The steps look roughly the same, but it’s a bit more involved.

Trying to work with Windows-native Kerberos

TL;DR it don’t work see the next section

Windows actually has Kerberos support built-in! And, with a fairly recent patch to OpenSSH that comes with Windows, enabling GSSAPI for SSH will use that Kerberos (or other authentication using the Security Support Provider Interface).

Configuring this Kerberos to log into CMU

Unlike MIT Kerberos, Windows Kerberos isn’t configured using a config file, but rather with registry keys that can be managed using Group Policy. wow so enterprisey

I found this tutorial to be the best resource on how to configure this for my computer, using the ksetup command (built-in if you have the Kerberos support installed).

Logging into the CMU Kerberos domain

Finally, the moment of truth: I had everything set up with ksetup, and tried to log into CMU:

C:> runas /user:hduvall@ANDREW.CMU.EDU cmd
1787: The security database on the server does not have a computer account for this workstation

As it turns out, there’s some extra steps in the tutorial above that are needed to actually allow non-domain computers to log in like this. Seeing as I don’t have access to the Kerberos Domain Controller, I had to give up on this route.

Modifying Windows OpenSSH to use MIT Kerberos

My next step was to get the SSH ported for Windows to use non-built-in Kerberos. Before you ask, yes I tried PuTTY as well but that didn’t work.

Installing MIT Kerberos manually

In order to actually use non-built-in Kerberos, I had to have a non-built-in Kerberos built. I downloaded MIT Kerberos 1.19.2 from their release page, and it was really simple to compile and install according to the instructions in the source.

Installing Windows OpenSSH, linking against MIT Kerberos

The Windows OpenSSH is hosted on GitHub and has clear instructions for how to install as well… so long as you have Visual Studio 2015. I have Visual Studio 2019 so I had to modify some PowerShell scripts by hand to get it to detect my build system correctly.

I also had to manually roll back to pull request that shimmed in SSPI for GSSAPI, making it link against the GSSAPI provided by my installation of MIT Kerberos instead. Editing the code wasn’t too hard, but I had to edit all the .vcxproj build files to get it to find headers/link correctly.

All in all it was a lot of messy, manual work that really only benefits me, so I didn’t bother making a pull request or even pushing to GitHub.

Conclusion

So It Was Actually Lyme Disease

Thu, 19 Aug 2021 12:00:00 GMT

So as I mentioned two posts ago, I had a pacemaker put in. Initially, the doctors thought it was “Athletic Heart Syndrome” but that didn’t really make any sense since I don’t work out that much. They put in a lab for Lyme Disease which came back positive.

This makes a lot more sense, since I do regularly go for walks and runs in a wooded area with known deer. I guess a tick bit me and I never noticed.

The strangest part is that I didn’t get any of the other usual symptoms of Lyme, just Lyme Carditis.

The other symptoms also have a much higher prevalence than Carditis, so the doctors figured I was very unlucky to have it strike like this but very lucky to have it be caught in time.

I’m taking IV antibiotics for the next weeks, right up until August 27th, switching to oral antibiotics provided my condition improves. I have a semi-permanent IV in my arm called a PICC line to allow me to take the antibiotics at home, and after that I’ll be fully cleared to go back to exercise!

I can’t wait to finally be better and really hope this works.

Xtreme Hakyll Upgrades

Mon, 16 Aug 2021 12:00:00 GMT

This is sort of a tradition how: I get more free time, decide to work on my blog, and instead of actually writing posts I work on the backend and don’t actually post about it until I’m done, at which point I’ve started to run out of free time.

Fortunately, this time I have a lot more free time, and the upgrades are actually complete! Really! For real this time! (ok not really I still want to implement syntax highlighting eventually but that’s for later ig)

The thing I’ve done: taken Hakyll’s incremental compilation abilities and extendend them to work better with my “numbered post list with links between them” format.

Why Hakyll’s Dependency Tracking is Currently Not Good Enough

With standard Hakyll, if all posts link to each other (see each other’s titles), then updating one post will update all the other posts via a link like:

post 1 --(depends on)-> post 2 --(depends on)-> ... --(depends on)-> post n

However, I’m only viewing a post’s title, not anything about its body, so a post should only be re-compiled when:

It’s source file changes
It depends on a post whose source file has changed

because source file changes are the only way for metadata to potentially change. Side note: depending on the metadata directly is a bit harder, with what I know now it may be possible but what I have works currently and is a bit better for numbered post lists in general.

How I Changed It

I added a new PostMetadataDependency type to the Dependency enum, as well as a numberedPostListMap to the DependencyFacts struct. Every time you make a numbered post list and use fields in its context to view titles/urls of other pages, it makes a PostMetadataDependency for that page through the magic of the Compiler monad.

If this sounds similar to how Hakyll already works, that’s because it is. Here’s the key difference: You can have a PostMetadataDependency on Nothing. Why would you want this? Say you are adding a new post to the end of the current list of posts. This will be the new last post. The previous last post, which didn’t have a link to any later post, now needs a link to the new last post as a later post. How else can you model “please re-compile if something that doesn’t exist yet happens to exist in the future”?

The implementation is not elegant. I don’t know Haskell very well, and Hakyll is quite a complicated library with many layers to dig through. In addition to Dependencies.hs, I also had to modify Rules.hs (for a marker of whether something was a numbered post list), Rules/Internal.hs (for combining all those markers into a map of all numbered post lists), and Runtime.hs (for getting that map from the result of the Rules monad and passing that to Dependencies.hs). Oh and also Binary serialization/deserialization for all the new datastructures I created. All in all it took a lot more effort than I expected.

Was It Worth It?

Sort of. It wasn’t a huge deal in the first place anyways, as a rebuild of all posts is very fast. But gosh darn it, if Hakyll advertises incremental compilation, then ima use it, even if I have to implement it myself.

I may also write a more thought-out blog post going into exactly how Hakyll’s Compiler and other systems work, now that I’ve read and re-read so much of that code. Will have to be soon because I’m liable to forget it :P

TL;DR my blog compiles v fast because i made it do that with much haskell effort

So I Have A Pacemaker Now

Sun, 15 Aug 2021 12:00:00 GMT

Recently, I was admitted to the hospital because I felt my heart start to beat wrong, and it turns out I was right: a combination of skipped beats and lowering heart rate led to low blood pressure and led to me almost fainting while still lying down (in the hospital bed)! The doctors gave me something to temporarily raise my heart rate so I didn’t black out on the spot, and within the hour I was scheduled for surgery.

Fortunately, the surgery was a huge success and I was discharged the next day (!!!). I can’t be thankful enough towards the entire cardiac team for making it go as smoothly as it did. I’m still recovering & the doctor’s orders are to not exercise for the next few weeks and stay in the area until the follow-up appointment. While I’m peeved at the change in schedule (not being able to go back to college as soon), I’m just grateful to still be here, with my family, even having the possibility of full recovery.

Another fortunate thing is while this happened during my Facebook internship, it was the last week and I already had all my tasks done. My manager was very understanding and let me take as much sick time as I needed. I’m thankful to everyone on my team there as well, for getting me through the internship and making it a good experience even with that hiccup at the end.

TL;DR I’m fine and won’t die, thanks everyone!

Working at Facebook

Mon, 21 Jun 2021 12:00:00 GMT

I think I’ve mentioned this in a previous blog, but ya I have summer internship at Facebook. A big part of it was probably my school, as there are a lot of other CMU rising juniors and seniors and graduate students here as well (all at the same start date for convenience), but I like to think part of it was also a good interview performance (what can I say i know some graph algorithms).

They had a whole team selection process, and probably because I put “comfort using Rust(lang) = high” on the sheet and had systems just below AI in terms of project preference, I was put on an infrastructure team working on a large Rust code base.

The Code

Ima wax poetic about how cool Rust is for a sec because this is my blog and I make the rules here. So: Rust is super cool holy cow.

rustfmt makes sure no matter who writes the code, it all becomes readable at the end because everything from capitalization to spacing is opinionated.
Builder patterns are encouraged by the language and safe to use and work pretty darn well as a result. If you don’t know what a builder pattern is, it’s basically something like

let foo = FooBuilder::new()
            .set_bar(5)
            .set_kaz(17)
            .build();

The type system makes everything safe and even though you “fight” the compiler at times it’s really all about making you code better and I love the very descriptive and colorful error messages

Seriously if you have a project you’re considering doing in C or C++ due to performance reasons just use Rust you’ll thank me later.

Ok so back to Facebook: because the team is using Rust, and they didn’t quite expect an intern to have known Rust before the internship when they were drafting up the project plan, I breezed through the first few tasks and am now working on pretty real stuff that hopefully will be useful to other engineers there. Can’t divulge too much about the project specifics unfortunately but it’s fairly widely used internal app, I’m just adding more user-facing metrics to it.

At this point, I’m sort of in a Rust + SQL part of the project, and may even transfer over to Python + SQL once I need to use another certain internal framework. I haven’t actually used too much SQL before, but SQL Builder has been very nice to use in Rust, and the Presto Docs are surprisingly well-writen and informative. Sub-queries, GROUP BY, and FILTER(WHERE ...) are magic, query executors doin some crazy stuff to get those to be efficient with so much data.

Internal documentation alternates between “We have something from 2016 on it but it’s horribly out of date, just read the code (oh wait you don’t even know what the thing you’re looking for is called? sux to be u)” and “Everything is explained perfectly and has examples and test cases and is formatted well” which is hm so I’m glad I have a good manager. Which brings me to…

The People

With few execptions, everyone I’ve met at Facebook has been very nice and willing to help. My team and especially my manager + peers have been extremely helpful at answering my n00b questions at any hour of the workday and making me feel welcome and appreciated. Another great thing, what I also liked about CMU and TJHSST and what I should expect from most good tech companies, is that everyone is super smart. I like being surrounded by smart and hard-working people because that in energizing and I like both the environment and how it makes me a better person.

In some respects, I may have been lucky, because hearing from a few other interns (during optional organized “just get on a call and hang out with other interns in an unmonitored breakout room” that of course I’m going to go to) it seems not all the teams are this quality or they’re having trouble adapting to work and not getting enough support or they don’t really like the work they were assigned or whatever. I can get jazzed about lots of computer-related stuff and like to be sociable so that helps, but I’m still very glad I got the team I did.

TL;DR Facebook is a neat place with neat projects, still don’t know how much I agree with their ethics even with them trying to turn themselves around.

disasm on AWS Lambda!

Sun, 23 May 2021 12:00:00 GMT

2023 Update: This Site Is No Longer Online

rest of post preseved for posterity :)

I’d like to formally announce that my fork of ret2jazzy’s disasm.pro is back up and being hosted on AWS Lambda!

This was a quick project to re-familiarize myself with cloud computing paradigms and technologies before my internship starts (tomorrow, I guess). Thanks to needing a custom fork of the keystone disassembler, this whole project took a bit longer than expected.

The solution I ended up going with was:

Putting the entire project in a Docker container
- This overcame my previous distaste for Docker quite well, because it worked very well and solved the messy dependency problem nicely.
Using serverless-wsgi to glue the AWS Lambda interface to the existing WSGI outline.
- Using the raw Serverless Framework did not work very well at all, disappointingly.
Putting the Lambda function behind AWS API Gateway at its two POST endpoints, /assemble and /disassemble.
Using CloudFront to redirect any requests not to those endpoints towards an S3 bucket containing all the static files.
- The container and API gateway actually also contain support for static files on their own, but this makes things faster for sure.

So all in all, my goals for this project were accomplished: I stood up my tool again and also learned a lot about AWS. Hopefully I won’t have to do this again for a while, debugging everything was awful!

The Housing Search Is Brutal But At Least It's Almost Over

Wed, 19 May 2021 12:00:00 GMT

For the past couple of days, I have been searching for off-campus housing.

As for why, I have been spoiled by my current apartment and years of living in my own bedroom, enough to never want to repeat the experience of my Freshman year when I lived with an annoying roommate in a tiny flat. This led to my 4 main criteria: 1 bedroom (not a studio), close to campus, nice kitchen, hardwood floors.

Yes, I’m that picky. Spoilers: no I did not find something matching all of those again like I lived in this semester.

The Cause

When signing up for university housing, I didn’t get a single-bed room. They were flat out (wow I wonder why) and some other early riser decided to register themselves as my roommate without so much as contacting me first. I don’t blame them though; I blame the university for such a high-pressure system. You technically have all semester to find roommate pairs, but everyone I knew either had a roommate group already planned or was living on their own, and the one new potential roommate I found flaked on me in the last week before room selection. So, some other poor chap, also without a roommate and with only 5 minutes to register his room before his timeslot is over, sees a room with an empty spot on a low floor and just goes ahead because our sleep schedules aren’t that far off.

Now, I could deal with their sleep schedule. Maybe. I’m also picky about that. But what prompted the whole off-campus housing search was what I could glean about their personality from their online profile. Basically, they’re the kind of person to start a non-profit for their College application and ditch it after they graduate High School. I despise that whole concept, having gotten roped in to one of those once, and I don’t note it anywhere because I don’t believe it accomplished anything of value. Heck, I even list my 1kbwc game engine project and not that because at least the silly Python-and-pure-JS app shows something about my capabilities.

Is that a bit too judgemental of someone else? Probably yes. I’m sure it could’ve worked out, especially if I tried hard to rid myself of this animus. But after what happened Freshman year I don’t want to take chances. So anyways, back to the housing search:

The Result

Before this process, I had no idea how to conduct a housing search. This was always the realm of my parents when I was but a wee lad (duh), and I was content with University housing until the above. So I was not prepared for the process of “scheduling a tour only for ‘oh sorry someone else bought it already’” to happen like 5 times in a row. Talking to my parents, apparently this is normal for hot housing areas. People want property so you gotta play the game to get it. The landlords control the supply and everyone else is fighting over the demand.

Another thing I was not prepared for was “sorry we don’t sell to undergrads.” This came up more often than you’d expect, because I seemingly do not look or sound like an undergraduate to most people, so when I say I go to CMU they peg me as a graduate at first glance. I’m not entirely sure why some places have a ban on undergradutes, some hypotheses are that they make too much noise late at night or don’t do a good enough job with cleaning/maintenance. Neither of those apply to me I don’t think but some were pretty hard-line about this, which sucked because that landlord controlled a lot of property.

The Conclusion

In the end, I found a place that matched 3 of my 4 criteria, only difference being that it’s carpeted (I can live with that, so long as that’s not in the kitchen). I was pleasantly surprised when I inquired about it two days in a row and the unit was still available, so I sent in the application and fee and am now hoping for the best. The rent is slightly more expensive than CMU rent, but the latter was for half of a bedroom while this is for a full bedroom so I think that’s fair. Fortunately the rent will be completely covered by what I’ll make over the summer, so lucky to have no issues there.

In any case, I’m glad it’s all over, I can feel the difference in not having the weight of “where will I live next semster” on my shoulders. Everything tends to work out :)

S21 Semester Actually Over!

Mon, 17 May 2021 12:00:00 GMT

Can’t believe the day has finally come!

All my finals and final projects are turned in
I’m moved out of my apartment and am back in my parents’ house with my family.
Just chilling before starting my remote internship on the 24th

It’s very nice to not have any more deadlines, although it’s sort of been like that ever since last Monday (with the exception of my 15-210 final on that day and my Japanese final paper due Friday), so I’m already in that Summer groove. I will definitely hang out with friends more, or at least try to, now that we are all relatively free of deadline shackles. Here’s to a good Summer ‘21!

Spring '21 Semester Almost Over??

Wed, 28 Apr 2021 12:00:00 GMT

Wow that was fast! I can’t believe that in just another week, all my classes will have finished their lecture series and, with the exception of one class, all my final projects will be turned in and done with. After that, I’ll have an entire week to just relax before I pack up and move home for my remote internship.

It’s a little early to do a semester post-mortem, but there are some things I would like to comment on now before my probable series that comes out truly after-the-fact:

Despite a less crazy school and work schedule, I still didn’t feel like I had enough time to do all the projects I wanted. With the exception of Februrary, I did very little art or personal programming projects. I’m not sure if this was not having time (what it felt like) or just not making time.
I did well in all my classes. Perfect GPA maintained (sunglasses emoji).
Could’ve done better, but I maintained a good workout schedule.
Could’ve done much better, but I was able to see friends often, especially with Frisbee practice becoming a regular thing.

Oh yeah and it was my birthday the other day too (: Family came, and it was nice to see them again for a little bit, and I’ll see them again soon.

Grad School Thoughts and a Blog Post That Resonated With Me

Sat, 10 Apr 2021 12:00:00 GMT

The scene: someone in the CMU Discord, #scs channel, posted a link to a fairly cool article about a misuse of algorithmic complexity in social sciences. The writing was engaging, and I had time to kill, so I started reading other linked posts, and eventually got around to reading their latest post, “The 5 Year Update on Skipping Grad School (and Whether I’d Recommend It)”.

If you know me, or have seen my early posts (or really even the existence of the AI and CV tags), you’ll know really like AI and CV and potentially even be comfortable starting to do research in those areas. In fact that’s part of what I did for my internship at Aerospace, and if things go flawlessly, at Facebook as well. And yet, I don’t find myself looking forward to doing a PhD or Grad School or really even doing undergraduate research.

I think that’s part of why Alex Irpan’s post and especially its prequel from when he was an undergrad struck me. Here’s an eloquent guy, who’s done some pretty cool Reinforcement Learning research with robots at Google for crying out loud, talking about a similar feeling he was having as an undergrad. He put into words some of the reasons why, even though I think research is super cool to get done and is very nice to read, it can just get too overwhelming. Their writing is much more engaging than I can replicate here, but if you too are at least considering Grad School then I think both articles (and the ones he links to) are worth a read.

After reading these, my current plan of “just try to go into industry with a little undergrad research experience from a summer maybe and hope you can do cool industry research as well” hasn’t changed, and if anything seems even more feasible. Even if I don’t end up doing research, AI-adjacent SWE would be fun as well, just a different set of problems to overcome.

“Wow a Saturday post??? ur wildin” ya u better know it I just wanted to share this article lol

My Mile Time

Mon, 05 Apr 2021 12:00:00 GMT

I ran my first sub-6 mile today, a 5:59! I had no idea I could do this!

A bit of background: the captains of the Mr. Yuk Ultimate Frisbee team (which I am a part of, and is still running practices despite all games being cancelled) posted a challenge in the group Slack. If anyone could beat their 5:50 mile time, the speedster would get a special prize. While I wasn’t expecting to beat their time, let alone come anywhere close, I thought it might be fun to try because I haven’t run a timed mile since Sophomore year of high school, when I ran a 6:43.

One reason I may have improved: Since last summer, I have been running every week. At first it was just simple Frisbee track practice drills to keep my stamina from last year, then it was training with my brother and his actual track coach to improve speed and sprinting form, and now for the entire spring semester I have been working on distance runs, slowly pushing from 4 to 5 to now almost 6 miles as my last run.

And this is all on top of other weight training I do. I am probably at my current peak fitness, even counting Junior year of high school when I was doing wrestling practice 6 days a week or Freshman Spring of college when I could (just barely) play 6 frisbee games in 2 days.

This feels good. So good, in fact, that I remembered my blog and also that I forgot to post last weekend. Oh well, I didn’t have much to write about anyways except for more cool CV projects.

Happy Pi Day!

Sun, 14 Mar 2021 12:00:00 GMT

At 4:32PM today, it was 3/14 15:92, which is approximately π\piπ. Very neat.

Don’t have too much to talk about so have a grab-bag of things which I think are pretty cool:

My Friend’s Blog

Just getting started, but has a really good first post! https://williamoconnell.me/blog/post/magic-pirate/

Someone Else’s Blog

jez.io has some cool posts too. I forget how exactly I came across this blog, but it’s pretty good too.

A CV Project I’ve Done

This was for class, doing some (non real-time) AR composition:

The Kung-Fu Panda trailer is composited onto a textbook cover

Yes it is very shaky, no I'm not entirely sure why (probably because the homography computation was unstable, wasn't sure how to fix before the deadline).

The Closest Thing We Have to Magic Nowadays

Sun, 28 Feb 2021 12:00:00 GMT

Alright, so this is a bit silly, but it’s one of those things I think about often enough that I’m writing about here lol.

Background

I’m fascinated by magic systems that seem to have some sort of underlying ruleset to them. I’ve been heavily influenced by magitech-like series such as A Certain Scientific Railgun, Ender’s Game, and The SCP wiki (especially works by djkaktus ), as well as more fantasy-esque works like Quiet & Antagonism (this one is very obscure but I love it so much it’s definitely worth a read) and Bone .

OK so maybe not all of these are all quite rule-based as I would like for a good example, but the point is that when I think of “magic,” I don’t immediately default to thinking about western wizard-style magic like DnD, LotR, Harry Potter, or magic from insert generic isekai fantasy here, or maybe a shonen fantasy’s magic system.

Narratively, a rule system sets up some extra hurdles the protagonists must work around, showing their perseverance, or some neat things they can exploit in unexpected ways, showing their cleverness. A good rule system allows for endless possibilities, even outside of what’s shown in the story. I like this because it gets me thinking, “how would I go about figuring that out?” which is very engaging.

What this has to do with real life

There are plenty of memes about it, but just think: electronics are basically magic. I’m taking an electronics class right now (18-310, Fundamentals of Semiconductor Devices), and a lot of the stuff we are learning just blows my mind how scientists were able to figure out in the first place. And yet, here we are: we have a model for exactly how little invisible """electrons""" move around inside materials carrying some intrinsic, quantized """charge""" and if you set things up just right we can, to borrow from the overly simple phrasing in the meme, “make sand think.” But this isn’t magic, it’s real! For everything we learn in this course, we can set up an experiment to test it! People already have!

To me, that’s the most awesome part; this world does have a “magic system” and I’m learning it right now. Sure it isn’t, “if I think hard enough, I can shoot a fireball out of my fingers,” but that’d be too easy and those are the lazy systems I hate anyways. You need to know a lot of math and physics and chemistry to do anything here, but that’s OK. All I can say is that I’m thankful to be around when so much has been figured out already, and we have the tools to figure so much more out too.

Aside

I would like to see more math used in magic systems. Math is so important in real science, after all. This has motivated me to start designing my own world/magic system, the both of which are still woefully incomplete. I hope to one day write a story using it, but who knows if that day will ever come lol.

Cooking at College!

Sun, 21 Feb 2021 12:00:00 GMT

As a college student with ample amounts of free time, I have taken up cooking as a semi-hobby to pass the time and remind me of home. I say “semi-hobby” because I don’t have a full passion for it, but I do still like to cook and practice it regularly

Some things I’ve cooked:

In chronological order:

Chicken Tetrazzini

Spaghetti with Meat Sauce

Cream of Chicken Vegetable Noodle Soup

Taco Materials (Tacos not shown)

Lasagne

In order of best to worst (at least to my taste buds):

Lasagne
Chicken Tetrazzini
Cream of Chicken Vegetable Noodle Soup
Spaghetti with Meat Sauce
Taco Materials

This rating is mainly because it didn’t store in the fridge well; when it was fresh I would’ve rated it higher than the soup.

Do not that this isn’t an extensive list, as I have cooked other stuff but didn’t take pictures. Cheesy Scrambled Eggs, Turkey Sandwiches, and of course the classic Peanut Butter and Banana Bagel have been staples of my diet as well.

History of these dishes

Everything except the soup is from a recipe my Dad has taught me. Of course he’s not the sole inventor of any of them but I do believe each of these recipes has been tweaked significantly by him, over the course of many years cooking for our family, to be the dishes I know and love. I’m currently in the process of copying his cookbook one recipe at a time by phoning home to ask and then making them here. I like the food I grew up with, and unless I decide to start cooking more Korean-style stuff due to the Korean grocery nearby, I think it will stay that way for a bit longer.

Future plans

Cheese Gnocchi Soup
- From a “Dead-simple cookbook” my grandparents got me, where each recipe assumes you have access to full kitchen supplies and many different spices and wines and meats and no, I do not have those. I didn’t skimp while buying stuff for my apartment kitchen, but I didn’t go all-out either. I do not have a mixing machine or a pressure cooker or wine or 5 different spices geez.
- So, this recipe will be modified but I think the base will still taste pretty good.
Chocolate Chip Cookies
- This is where my Dad’s experimentation truly paid off. You think you’ve had the best chocolate chip cookies? You have not had these cookies yet. I know it’s cliche but really you have to try them. If you’re on campus, shoot me a message and I’ll bring you some once I’ve made them.

The Senate Trial

Sat, 13 Feb 2021 12:00:00 GMT

So, it happened. After being impeached for incitement of insurrection, Donald Trump was acquitted by the Senate by a vote of 57-43, 7 Republicans joining the 50 Democrats in a total 10 shy of the 2/3 super-majority needed to convict.

What follows is a summary of everything I personally have heard about the case:

Main Arguments For Convicting

There is little ambiguity that Trump directed his crowd towards the Capitol Jan. 6 with violence-imbued language.
- Near the end of his speech: “And if you don’t fight like hell, you’re not going to have a country anymore. … So let’s walk down Pennsylvania Avenue.”
He did not immediately call the crowd off once it was clear they were turning violent and breaking into the building.
Later statements did not help either and could be interpreted as praising the rioters.
Trump’s language and messaging from the November election until the riot consistently repeated the falsehood that we won the election and that the entire system was rigged against him in a massive conspiracy. These claims are false, but believing in them can be dangerous especially when you think the very fate of your country is on the line.
- Trump must’ve been aware of this, and besides, ignorance of a massive angry mob does not excuse you for continuously inciting a massive angry mob.
All of this could be interpreted as Trump wanting a coup attempt to overturn the election that he knew he lost. It should be unforgivable for anyone, let alone a sitting president, to do this.
Conviction would prevent Trump from holding office again, a good thing given what happened last time he was removed from office.

Main Arguments Against Convicting

I’ll order these from most to least convincing, at least for me:

Conviction/impeachment is primarily a tool to remove a president from office, an argument that Mitch McConnell used despite also denouncing Trump’s actions for many of the reasons above.
- Counterargument: conviction/impeachment can also be a tool to keep a former president from holding office again, which is pertinent in this case. If it was only for removal, then a lame-duck president could do absolutely anything they wanted their last few days in office because it’d be unlikely the House and Senate could move fast enough to impeach and convict.
Trump was simply exercising his first amendment rights! Democrats say “fight” all the time in their speeches too!
- Counterargument: First amendment rights are not absolute; see any number of hate-speech crimes (where similarly, plausible threats of violence occur) or perhaps leaking confidential information. Both are """just people saying things :)""" but clearly illegal. There is not much different about this from the former, and in fact the statements are higher (implicit threats against the highest legislative body in the nation, not just a single person).
The riot wasn’t actually that bad.
- Counterargument: they literally broke into the Capitol. That alone makes the United States look like an utter fool on the world stage. The fact that a sitting president was seen to incite it makes it even worse.
Antifa did it
- No, i think it was the far right actually. Too many of the people convicted have ties to far right groups for that to hold water (use https://github.com/iamadamdev/bypass-paywalls-chrome btw, great extension).
- Honestly if you actually believe this I don’t think I can convince you. Just know that buying into a conspiracy that requires you to constantly doubt every piece of information you come across isn’t always a good one. Occam’s Razor please.

I think from reading this you know what side of the fence I sit on. I believe that convicting Donald Trump would’ve been the correct thing to do and I highly respect the 7 Republican senators as well as the entire body of Democratic ones who voted to convict. Trump’s stain on America’s reputation will not be forgotten any time soon, especially by the world at large. I worry for what this means for America’s future.

First Week at College! Or, How to Install MPL (MaPLe)

Sun, 07 Feb 2021 12:00:00 GMT

All in all, not too bad of a week at all! Classes haven’t ramped up yet, so all the assignments this week were easy introductory ones. Computer Vision (16-385) should shape up to be a very good class based on the lectures and the fact that programming homework is 70% of the grade (I love programming!!).

For being alone in an apartment during quarantine, I’ve been doing pretty well. I’ve taken lots of walks around campus and Schenley Park as well as participate in Figuary (figure drawing month).

However, one day I got really bored with all my free time, and decided to do something programming-related instead. My Parallel Computing (15-210) class uses MaPLe, a home-grown fork of the MLton implementation of Standard ML (SML) designed with simple parallel/concurrent programming primitives. We learned SML syntax and functional programming concepts in 15-150, and I already have some experience in parallel from TJ, so using it shouldn’t be too hard.

CMU provides the MPL compiler on their remote servers, available by SSH, but I prefer to work locally. So, like ya do, I compiled MPL myself using their instructions. It was surprisingly easy so I thought I’d share how to do it with the rest of my class. Here is the full guide I wrote up and shared on the class Diderot:

Learn you a local MPL install for great good!

Hello everyone! Today I will be showing you how to install MPL (MaPLe, the Parallel SML compiler thingy) locally, on your own machine, because you might be like me and not prefer to do things on the Andrew machines.

First, like with any good recipe, I will tell you my life story. A long time ago, I did not know how to compile things. Then, I started using CentOS (for various reasons) which never has up-to-date packages. Wanting to use up-to-date packages, I had to learn how to compile things. Along the way, I got pretty good at compiling things, even if they were hard to set up.

Fortunately for you, MPL falls on the easier side of programs you have to compile yourself! And without further ado, let’s get right into that:

Installing Dependencies

I will be providing two sets of instructions: one for those running Debian 10 (or derivatives like Ubuntu or Linux Mint), and those running MacOS Big Sur. I have tested both sets on respective fresh and squeaky clean installations, so if they don’t work for you then too bad good luck fixing your environment. If you are running another Linux Distribution (or even worse, a BSD *shudders*), then I trust you known what you are doing well enough to adapt these instructions to your own liking. If you are on Windows, just get Debian for WSL (so much better than using a virtual machine)¹.

Installing Packaged Dependencies

According to its README.md, MPL requires the following packages to be installed for it to be built (N.B. make sure your package manager is working and up-to-date before running these!).

Debian:

sudo apt install build-essential libgmp-dev git wget

macOS: First, install XCode from the app store and homebrew from https://brew.sh/ if you haven’t already. Then, run

xcode-select --install
brew install coreutils git gcc gmp mlton

(Note: binutils, make, and bash should be present by default on macOS with the XCode CLI tools installed. If they are not and compiling MPL fails, run brew install binutils make bash and try to compile MPL again)

Installing MLton (Linux only)

Homebrew is nice enough to have a pre-compiled MLton in their package repository, but unfortunately Debian is not so nice. So, we will simply have to download the package and install it ourselves²:

wget https://github.com/MLton/mlton/releases/download/on-20201002-release/mlton-20201002-1.amd64-linux.tgz
tar xzf mlton-20201002-1.amd64-linux.tgz

Next, skip ahead to the “Installing MPL” section and complete the first step, replacing all instances of mpl with mlton. Then, change directory back to where you ran the wget command and run:

cd mlton-20201002-1.amd64-linux
make PREFIX=$HOME/.local/mlton install

You should have mlton in your path now; try running mlton with no arguments to test. If it prints out a version number, you should be set. If not, then idk try to figure it out.

Compiling MPL

For both systems: change directory to a suitable build folder, and run the following:

git clone https://github.com/MPLlang/mpl.git
cd mpl

Then, on Linux:

make all

MacOS is a bit tougher, but not too much. MPL was written with Linux in mind, and as such, uses a few Linux-isms in its code. So, I made a patch that gets rid of those in a way that doesn’t affect the functionality of MPL too much (I don’t think, at least). The following commands download and apply that patch, as well as tell the Makefile to use XCode tools:

curl -sSL https://gist.githubusercontent.com/hazelduvall/0ed62747ee765ed663eeca59a141c820/raw/d65448c91fac52e59afd58982cad0663488842ac/0001-MacOS-fixes.patch -o 0001-MacOS-fixes.patch
git apply 0001-MacOS-fixes.patch
make AR=ar RANLIB=ranlib all

Installing MPL

Now, if you’re like me, you like to just type mpl to run something instead of /huge/path/to/a/build/folder/mpl/build/bin/mpl. I like to put built binaries in a ~/.local directory, and then add that directory to my path. Doing that for mpl:

cd $HOME
mkdir -p .local/mpl
# Run this line in your shell, and also add it to your ~/.bashrc or ~/.zshrc or similar:
export PATH=$PATH:$HOME/.local/mpl/bin

Now, navigate to the directory you cloned mpl to earlier, and run:

make PREFIX=$HOME/.local/mpl install

And that’s it! You’re done, congratulations! As a test, go to the examples folder of mpl and try to run make nqueens to see if the compiler works correctly.

DISCLAIMER: DESPITE MY BEST EFFORTS IN TRYING ALL THESE STEPS MYSELF, THEY ARE NOT GUARANTEED TO WORK FOR YOU.

So please, use your head and a search engine before typing a comment about how you got a scary error message and want someone to fix it.

ALSO READ THE WHOLE THING PLEASE

Footnotes

As much as I would also like to provide instructions for native Windows, being a Windows power-user myself, the preferred method for emulating a Unix environment, MSYS2/MinGW64 doesn’t let MLton use fork() for some reason, and I staunchly refuse to recommend Cygwin, so yeah just use WSL. ↩
The astute reader might have noticed that we are not using the latest version of MLton. That is because the Github release for it with glibc 2.23 is broken (bad ld.so in the mlton-compile executable, “only” requires a minor change to the released bin/mlton script that is more trouble than it’s worth to put here), and the other version is for glibc 2.31 which is too new for Debian. ↩

Last Post of Winter Break!

Sun, 31 Jan 2021 12:00:00 GMT

Wow I can’t believe winter break is over! And yet at the same time, I can’t believe it was so long.

I’ve moved into my new university apartment faily close to campus and am writing this post from my trusty laptop in the now-designated computer room. Due to the de-densifying policy CMU has adopted, because this was a 3 person apartment that 2 people inhabited last semester, I am the only one in it this time around.

It’s pretty nice, having the space for 3 people all to myself. I don’t feel cramped at all (unlike the last place, smaller than just this room and I had a roommate!).

COVID is still putting a damper on meeting people, though. There’s a mandatory 14-day “quarantine” where all classes will be online and extra-dorm activities will be limited to exercise and grocery runs so long as you stay on the CMU campus. Not even seeing anyone in-person outside of people on your floor, and even then it has to be outdoors + with masks!

All told though, I’m very lucky given the circumstances. No sickness, good setup (thanks Mom and Dad for helping me move in; you guys were awesome and I couldn’t have done it without you) and 100% prepared for classes.

See y’all in a couple weeks, by which time I may or may not be overwhelmed with classes! We’ll see.

My First "Real" Open Source Contribution

Sun, 24 Jan 2021 12:00:00 GMT

Near the beginning of winter break, I had an idea to contribute an implementation to ONNX, Microsoft’s Open Source neural network accelerator. I had just been dropped from the Aerospace internship (due to accepting a position for Facebook’s summer software engineering internship (more on that later I swear)), and AI ideas were still fresh in my mind, so it was a good opportunity to keep my programming skills sharp.

I’m proud to announce that I’ve done it! You can check out my work here. The full README.md is a bit too wordy to include (and honestly is redundant to list again here), but I’ll include the description section so you have an idea of what I was working on.

Description

Fully Convolutional Networks (FCNs) are a neural network model for real-time class-wise image segmentation. As the name implies, every weight layer in the network is convolutional. The final layer has the same height/width as the input image, making FCNs a useful tool for doing dense pixel-wise predictions without a significant amount of postprocessing. Being fully convolutional also provides great flexibility in the resolutions this model can handle.

The specific model I contributed can detect 20 different classes, corresponding to the COCO 2012 class analouges of the PASCAL VOC classes. The models were sourced from PyTorch models pre-trained on the COCO train2017 dataset using this class subset.

Oh, also, something I should mention: today is a double post! Go check out the previous post for more code stuff I did recently.

Live2D Demo Application Expanded

Sat, 23 Jan 2021 12:00:00 GMT

The slow burn project I mentioned earlier has concluded for now at least. Not for lack of progress, but actually because I completed a major milestone in actually getting it to work!

What does it do?

Currently, the application:

Renders a sample Live2D model at 60fps
Takes an image from the webcam and detects a face in it at 4fps
Composites the two images together

This required the intersection of many different graphics technologies which I’m not sure were all made to go together:

SDL creates the window and initializes an OpenGL context (together with GLEW to help with the dynamic loading that OpenGL requires)
Live2D Cubism Native Framework loads the proprietary model format and renders it (when in the OpenGL context).
OpenCV handles opening the webcam and detecting a face in it
A Custom GUI Library built on top of SDL spawns the two threads to do Live2D rendering and OpenCV maths at the same time, as well as controlling the layout of the scene.
And finally, raw OpenGL shaders that I had to find a good tutorial for to render the OpenCV data

In the end, I got a pretty good result:

I pointed the webcam at the screen to make this

I’ve put the project on hold for now, partly because I’ve accomplished a lot of what I wanted to do (get all these libraries working with one another at all), but mostly because School is coming up soon! I leave the 29th and start February 1st which is very close. Also I went skiing in the meantime which was very fun and put a damper in doing many things code-related during it. Anyways, see you next time!

Old Thoughts About Learning

Wed, 13 Jan 2021 12:00:00 GMT

Recently, I had the idea to look through some old AP Literature writing I did. Not to brag too much but uh, it’s pretty good?? I’m including an essay that I’m really proud of here:

Creative Writing: Nonfiction (Reworked)

30 May 2019

As I transition from high school to college, I’ve done some thinking about how I best learn. After all, so I’m told, college is where students are challenged to pace themselves, the bulk of the work being done outside the classroom. This in and of itself doesn’t bother me; I can time-manage and slog through mountains of homework with the best of ‘em. No, what bothered me is this: if I don’t know my own strengths, then I can’t play to them to save time or shore up my own weaknesses.

After some experimentation, I could conclusively point out things that are not my strengths. I’m impatient; I’m set in my ways; I can’t focus on future plans; I have limited people skills; I underestimate the value of others; I can’t think fast, or strategically; I can’t visualize shapes clearly; I don’t have an ear for music. I was told I think too poorly of myself, that I can improve on any of these things with time, that professionals have dedicated years of their life towards learning something I can’t possibly hope to understand in one week. I had a hard time swallowing this. It seemed like shifting the blame, which is pretty shifty if you ask me.

As an example, the other day, I tried to play the piano. It’s just notes on a page, those are easy to read, I thought. Individually, yes, after a good ten seconds of counting up or down from middle C. Then came the unanticipated challenge of getting my fingers to press the right notes without looking. Then came stringing all that together at a reasonable pace. I pressed through for a bit longer, learning to play a few bars, but then I got discouraged by lack of immediate progress. It was just another frustrating riff on other unfamiliar subjects I’d tried before, like art, Tetris, and basketball.

So, what was I comfortable with, able to do quickly and without a second thought? Programming comes to mind. What’s so special that makes me suited for it? Broad information synthesis? A keen eye for details? A knack for language of any sort? Raw experience curated by years of specialization? If it’s the last of the bunch, I feel cheated. I want to get good at a lot of things, and yet the only path towards that goal is time? What a ripoff. Time is precious. It doesn’t seem physically possible, let alone desirable, to spend every waking hour working towards one goal after another.

As I thought these things, I confided in a friend. He told me this: your perspective is totally off. First, you need to stop comparing yourself to others; it’s unhealthy. Then, you need to slap yourself in the face, tell yourself you are good enough, and stick to whatever constructive practice regimen is necessary. You will get better. You might not ever reach your goal, but you will get better, which is what matters. I was shook, for lack of a better word. Logically speaking, it makes sense that good artists, musicians, public speakers, and sports players spend years perfecting their craft, years which I don’t have under my belt. And yet some notion possessed me to think I could get the same results, only better and more quickly, simply because I could understand what they were doing, not ever really how.

Being relieved of that thought train was sobering. It did nothing to lessen the frustration I felt when I couldn’t draw a precise curved line, couldn’t beat my friend at Tetris, or couldn’t put that darn large orange ball in the darn red hoop. Instead of submitting, though, I now know what to do: keep going. I’m still at a loss for how to play to my strengths. But by some miracle of hard work I made it through high school, and gosh darn it, I’ll work my butt off to keep improving through college and life as well. In my opinion, that’s what being human is all about.

In the almost 2 years since I wrote this, I’ve kept the same perspective. And wouldn’t you know it, I’ve gotten a lot better at things. My art is leagues above where is was before (and I still have leagues to go). I didn’t practice Tetris or basketball, but I did practice some fighting games, and now I can beat everyone I know (not always easily, and none specialize in fighting games either and I still get destroyed online). I’ve also been working out (weight training + some running) and have seen steady improvement there too.

Having this mindset and being able to see the results has been very gratifying. It’s such a privilege to know that yes, if I put in the work, I really can do anything. I’m also fortunate to be in an environment where it’s even possible to put in so much work. Thanks to the support of my family and friends, I have achieved all my goals in school and mini-goals in hobbies. I look forward to doing even more!

Live2D License Close Reading

Fri, 08 Jan 2021 12:00:00 GMT

One project that’s been on slow burn throughout this winter break has been an attempt to get a Live2D Cubism avatar animated using OpenCV facial detection routines.

Becuase Live2D provides pretty much zero useful documentation in English, I have resorted to “copying” their Windows OpenGL Demo to try to make something useful. By “copying,” I mean that I had their code up on one screen, while I took note of the structure and names and typed similarly structured and named code on another screen. Like the good citizen I am, I had the thought, “wait, could it be possible that this isn’t legal?”

Turns out it’s legal

…although not for the reasons you may think. Follow along with the Live2D Open Software License Agreement if you woul like. Keep in mind that I Am Not A Lawyer so anything I say should be taken with a grain of salt.

Restrictive things

Most of the license seems to be dedicated to preserving the license itself. Many clauses are about how, under no circumstance, can anyone come into possession of licensed code or use it without first agreeing to the license (also no re-licensing under a different license because duh). Once that’s done, the restrictions are fairly minimal: just don’t commit crimes, and also if you publish your source code that uses this open source code, Live2D can use your code too. Live2D also retains half the copyright to any modifications you make to their source code. Also you need to keep all copyright notices intact.

Things you can do

You can redistribute the code (if people agree to the license), you can publish code that is based off the license, you can make modifications (to the code, not the licensing). You also retain full ownership over the parts of Derivative Works that aren’t Live2D code. All in all, it seems to be a fairly permissive license.

Where does my copying fall under this?

I was worried that my copying might be not allowed at first glance according to section 5.1, titled No Modifications. It basically says the same thing as the title. Under certain interpretations, my “copying” might be construed as a modification (because the same result could’ve been achieved by first copying, then modifying, or really the modifications were happening on a copy in my brain, idk). The Software protected by this license includes all copies and modified copies, so the code would still be protected by the license. If it had included language like “Only Live2D is allowed to distributed the Software” then I would’ve been out of luck. Fortunately, that is not the case and I am still allowed to distribute this regardless of whether or not courts would think what I have done counts as a copy.

Code

Still very unfinished at the moment, but here it is. Pay no mind to a previous version written in Rust; it current has no Live2D components, just an OpenCV demo with a broken Windows API hook (that I plan on fixing later, then porting the Live2D logic to, can’t be a copy/modification if it’s in an entirely different language).

Also sorry for not posting recently; Winter Break has been very fun and relaxing but putting stuff on my blog has not been part of my routine. I think leaving it like this, with monthly-ish posts might be good enough. No pressure on myself, though. Maybe I’ll post about an AI thing again one day (after my PR gets accepted probably, be excited ;)).

Finally, More Time

Mon, 21 Dec 2020 12:00:00 GMT

What am I going to do with myself?

Finally, the Winter semester has ended, and I finally have time to do non-school activities!

I have a couple things planned:

Work on contributing the FCN model for https://github.com/onnx/models. Currently, I am installing ONNX and ONNXRuntime which has proved a bit harder than pip install onnx onnxruntime for various reasons, unfortunately.
Do some art! Daily practice is the plan. I found some warmup routines to do, which includes a lot more gesture drawings than I was comfortable with but I think it has already made me a bit better
- Part of this will be to make some personal art, and part will also be to make some art for AAC’s Secret Santa art exchange.
Walk my dogs
Practice Skullgirls?
Other AI reading/programming stuff as it comes up

Wait I don’t see any Aerospace on that list…

Yeah, and not just because I forgot to mention it. As a consequence for accepting another internship for the upcoming summer, I am no longer allowed to work for Aerospace during the school year. This kinda sucks, because I was having a great time there getting to do cool AI research. I hope the next internship (which I will elaborate on more in a later post) will be as good.

I mostly accepted the other offer because I wanted to explore more. The obscene amounts of money had little to do with it (although that is a nice perk). I am under a confidentiality agreement on the exact amount, but it definitely exceeded by expectation for intern pay, even for a California-based company.

Other notable updates:

Might get my first B in college due to getting consistent B’s on 21-355 exams. Although maybe not? Depends on whether the exams are out of 115 or 100 points.
I have been compiling pandoc on the machine hosting this blog for almost 2 full days now. It is still not done and probably won’t be for another 2 days at this rate. Fun!
Friends have organized a modded Minecraft server known as JEFFREY, which is pretty fun even though I am not very good at Minecraft.

Free time => More Hakyll Upgrades

Wed, 25 Nov 2020 12:00:00 GMT

Thanksgiving break, woo! I finally had enough time on my hands to get around to something that had been bothering me with this blog for a while: Hakyll doesn’t do dependency tracking right.

Wait, What?

Yes you heard me right. Haskell is a very complicated language, and Hakyll takes full advantage of all those complications to make a surprisingly robust system. But the robustness makes it a bit inflexible, which is why I ended up having to write a lot of extra code to get these numeric page ids.

But that’s not what was broken.

The thing wrong with Hakyll

Hakyll’s main branch treats the metadata of a page changing and the actual page content changing as the same event. This is troublesome because my pages all depend on the ones directly before and after it for metadata like title and date. This is separate from the actual content of the page before or after, so if the metadata stays the same, I do not expect the pages to rebuild.

“Not so fast,” says Hakyll. “This page you just edited is out of date, so I must propogate that out-of-dateness to its neighbors because I cannot tell if just the body changed or the metadata did as well.” Ok, that’s all fair and good, but this new out-of-dateness ends up propogating to those page’s neighbors, then their neighbors, so if any page gets updated, all pages get rebuilt.

I would rather this didn’t happen

So after understanding all that, I forked the Hakyll engine and added that distinction with an extra MetadataDependency type in addition to the existing IdentifierDependency and PatternDependency ones that already existed. I then tweaked the dependency resolver to skip over metadata dependencies when doing the dependency update chain things, and only look 1 neighbor level deep when doing the out-of-date marking for metadata dependencies.

Did that work?

After many long hours, yes! It did! But there’s a new slight problem: adding a page (like this blog post) to the end of the list does not update the old most recent page to have a “next page” link that points to it. This is because the old page did not have a metadata dependency on the non-existent page.

It’s possible to fix this by checking if there are any new metadata dependencies in the check for out-of-date items, but I am exhausted from writing so much Haskell already so I think I’ll leave that for another day.

Anyways, Happy Turkey Day tomorrow!

Online Identity

Fri, 20 Nov 2020 12:00:00 GMT

[insert joke about japanese “three faces” or native american “two wolves” idk]

So do I have two faces?

I would say yes, actually. “Hazel Duvall” is one of them, and ”█████████” is the other.

Wait like multiple identity disorder??

No, just my online presence :P

It’s been like this for a while now, me having multiple separate online “faces”. There’s a professional one (including this blog, however unprofessional it may seem), and one much less tied to my real-life presence. However, if you know the other one, you will probably still be able to find this one. Maybe. I haven’t tried too hard to doxx myself really, but apparently some people at my school have been able to put 2-and-2 together when they knew both.

Are you going to tell what the other “name” is?

I have redacted the name because I am still not comfortable tying the two together. It’s not that I am ashamed of either; rather, I’m proud of who I am in both regards. To me, there’s something fun about not having people know who else I am. It allows me to feel in full control of my identity, which is liberating. There are things I can do and say without fear of how it would impact how people see the other me, the physical one who needs to get hired and network and boring stuff like that.

It sounds like you are very attached to this other identity

Yup. In fact, I have purposely avoided using the term “true self” or “real self” because I’m not sure which one that is. Sometimes I’ll feel like Hazel, and other times I’ll feel like █████████, and all the time I’m still myself. The identities are similar enough that usually it doesn’t matter. I have the same core values, just different expressions of interest, different speech patterns, and different responsibilities. Both can be my true self at different moments and that’s ok.

Sounds like you’re a Furry

I don’t know where you got that impression :3

But I'm Still Here

Mon, 26 Oct 2020 12:00:00 GMT

Hi. It’s been a while.

This week has started off less busy than usual, but it’s still almost to the point of unmanageability. Not yet though. I’m still trucking.

Artober has been a bust. I ended up completing 10 drawings, and severely lack motivation to finish more. It wasn’t the prompts, I liked the prompts, it was the falling behind and eventually dropping out of the habit.

Math and Computer Science homework has been going very well, feeling good making very clever solutions to these problems. But I sort of want to do other stuff besides just Math and Computer Science homework, ya know?

What Would Be Really Fun (right now at least)

When I’m bored but don’t really have time to kill, I’ve been browsing The Mizuumi Wiki, a wiki for niche/indie fighting games. I found it by looking up resources for Skullgirls (last mentioned here), but I clicked on the sidebar for other games and found a lot of fun, weird, and interesting things.

So basically, I think it might be fun to contribute to that a bit. You know, just find a fighting game that hasn’t been touched a lot, figure out its secrets, meditatively record footage of different moves, simple stuff. Not that I have time for it right now.

Other Games

My friend also shared some other cool fighting games with me: Vernal Edge and Decline’s Drops. Both have Smash Bros.-like combat while actually being a standard 2D platformer, which I think is a really interesting concept and could actually work. Both of these demos are free, and I definitely plan on trying these out when I have the time (which I do not currently).

Anyways, that’s all for now. I wish I could talk about what I’m doing on homework more but that’s a little of an academic integrity violation to post solutions and stuff (which is kind of the only reason why I would talk about it). Maybe I can talk about the CTF club I’m in next time?

I'm Tired

Fri, 16 Oct 2020 12:00:00 GMT

Title says it. I’m beat. Exhausted. Feelin’ it.

I know exactly why I feel like this: I am overworked. My week is roughly 65 (!!) hours, which leaves no time to take a break ever. My classes are already tough, and then I have tutoring and Aerospace on top of that.

Nothing is hard by itself, it’s just a lot. All of it. The problem is I really would enjoy doing any of it. I love all the topics I’m taking this semester, I like tutoring, Aerospace is fun to do AI research, it’s all really good stuff. And yet, I literally do not have time to do it all.

Mistakes were made

The bigest mistake was made back last spring, when I signed up for all this stuff.

I was just barely keeping my head above water for a few weeks now, and I finally had a major slip on Wednesday when I forgot to turn in a written assignement for 15-213. Fortunately, I will be able to drop the two lowest homeworks (both of which were created by this mistake), but I really dread this happening again. It totally could, if not in this class, then some other class.

I’m asking myself to do the improbable here, trying to stay on top of all these responsibilities at once, but I still beat myself up over things like this. Thursday, when I was still blisfully unaware of the slip-up, I had a little bit of time to do art and relax, for like 1 hour, and I was happy. For once in a very long week I actually felt good about myself. Then came this morning when I realized my mistake.

The stress is getting to me

What right do I have to complain about stress? I have a loving family, good grades, steady access to food, friends that would listen to me if I told them about this, paying jobs that would easily hire me full-time if they could, so why am I so down?

And yet I know this is the typical thought process of someone who is depressed, but I refuse to call myself that because “other people must have it worse and I’m just a big baby.”

I can’t let this affect my work

I’m scared of falling behind again. I care too much about all of the things I’m doing to let any go. I’m trying to catch what little time I have and it just runs through my fingers.

Tonight I am going to go to bed early. I don’t want to think about this anymore. Writing this post didn’t help and made me only more frustrated and potentially even more exhausted from stress later. I can only do so much and I’ve done enough for now.

Artober Challenges

Sat, 10 Oct 2020 12:00:00 GMT

TL;DR - So among other things, I am participating in a very strange Artober challenge titled “Cringetober” and I have not posted for a while mainly due to an overwhelming amount of homework and other responsibilities.

Origin of Artober

Before talking about the art I’m doing, a little history is due. As far as I understand it, in the years before 2020, there was an art challenge known as Inktober that started off small but quickly rose in popularity. The basic premise was that, every day, there was a new prompt for doing a simple inked piece of art. You could make your piece really simple, or really complex, and tag your piece with #inktober and everyone would love it.

Later, however, the guy behind the original challenge started using his copyright to take down artbooks that people had compiled (of their own art!) that used “Inktober” in the title. This caused backlash in the art community, being seen as unfair, so this year people started to boycott Inktober and instead do an assortment of other daily art challenges for the month of October, collectively referred to as Artober.

The Challenge I am doing

There are a multitude of Artober challenges, and the one I picked was titled “Cringetober.” The prompts are all things that reference strange internet culture, and I liked it because a) I like strange things and b) the expectations to produce good art based off it is not very high.

Combining this with my friends’ suggestion of doing daily 1KBWC cards, I have produced quite the set of cursed images. (I think I last mentioned 1KBWC here? not sure though, I should do a full write-up if I haven’t already). Again, I don’t think they count as high-quality art, but I like them for what they are (very strange) and I like doing them. It’s a nice break to just not care about responsibilities for a little bit and make in-jokes for my friends to enjoy.

Geez dude you haven’t posted for a while

Ya ik. I keep meaning to but keep getting distracted. I only remembered to do something now because I have a free hour, exhausted other entertainment options, and really did not feel like doing more homework. Free hours are rare but it’s good to get thoughts down every once in a while. I’ll maybe update this post with the official Cringetober prompt list if I can. Until next time then!

So I'm a Tutor Now, Apparently

Sat, 19 Sep 2020 12:00:00 GMT

So much for “daily” thoughts amirite?

As one can infer from the title, I have started a tutoring job in addition to continuing my summer internship part-time as well as attending remote classes. So yes I have been kinda way too busy to type these blog posts until this weekend, when homework has died down a little.

Who even hired me as a tutor anyways?

CMU’s Student Academic Success Center did, after I applied for it and took a half-semester long training course last spring. Even after the bit of training I feel very inexperienced and have been doing a lot of learning on-the-job.

How many tutees are ya tutling?

I have no idea if those are actual words. Six, split equally across Saturday and Tuesday. This slots into my schedule fairly nicely.

What do you tutle?

I will use this word even if it doesn’t exist so help me. A class know inside CMU as 15-151, “Concepts,” or “Prof. Mackey enlightens you while downing a 2-liter bottle of Diet Coke.” I would call it “Introduction to proofs and computer science theory maths” but that’s not as formal as the real course title (which I do not remember because it is so boring and besides, the other titles are much better anyways). 15-151 is a proof-heavy class which is really challenging if you’ve never done math proofs before. I was in that boat my Freshman year, but I quickly learned how to paddle and ended up with a pretty good understanding of the material (I think). The students coming to me also have problems in a similar vein, ranging from not knowing strategies on how to write proofs, not knowing enough definitions, not easily grasping how to apply the definitons, and overall low confidence in their own ability after having been beat down by the first homework.

This is not to disparage any of them. Anyone coming into CMU is already leagues above the average in terms of problem-solving and general mathematical ability. All of them really do know how to solve these problems, given enough time. Writing and thinking in the language of proofs is proving difficult.

How do you go about tutoring?

For the introductory sessions that happened this week, my main goal was to get to know the tutees and understand where each of them was struggling by watching them work through practice problems. I don’t have a rigorous classification system yet, but this exercise was helpful for me and I hope for the tutees as well to get immediate feedback on some of their problems.

I anticipate that the Tuesday sessions will be fairly productive because homework is assigned and due every Wednesay, so the three tutees I have then will bring plenty of questions to work on. I am allowed to help them with their homework directly (benefit of going to the official Student Academic Success Center!), only by leading them with directed questions and not flat-out giving them the answer, though.

I wanna be tutored by you!

sorry i’m too busy as it is :)

Instead of these posts being daily, I’ll still try to make them weekly. The tag will still be “daily thoughts” because I want to keep all of these related posts in one category.

Windows Development Adventures

Mon, 07 Sep 2020 12:00:00 GMT

So one reason I’ve not updated this blog in a while has been school. Yeah I’ve been pretty busy with homework already & classes & everything, but not really busy enough to not just take 10 minutes out of my day to update.

What really triggered me not posting was a project I’ve been working on. I was trying to keep it a secret until “the big reveal” or whatever, but it’s just been taking sooo longgg that I had to make a post about it before I reveal a “final” (i.e. working, haven’t even gotten there yet) product.

In case you haven’t guessed from the title of the post, I am doing some pretty heavy development for an application targeting the Windows operating system. I’m still keeping it a secret (for absolutely no reason other than I enjoy keeping secrets), but the gist is that it’s a large open source application whose Windows branch has been unmaintained since 2015.

Things have changed quite a bit since 2015. It’s no longer reasonable to use Visual Studio 2013 to build your application. Windows DDK has turned into WDK and merged with the Windows SDK. The whole concept of drivers has changed (slightly), and the whole build process has changed dramatically. I am somewhat shocked that the extremely messy recursive NTMakefile setup still works, was more shocked when most of the non-OS-specific parts of the application built correctly with minimal additions (mostly to resolve merge conflicts that I created when I merged the latest branch in with the most recent Windows branch). Like, if you haven’t seen the code, you don’t know how amazing that is: it’s a lot of hot garbage anachronisms and crazy design patterns and hacks upon hacks that it’s a miracle it compiles, and I dearly hope it runs too (although that may be hoping beyond hope).

I was brought back to earth when I realized the Windows-specific parts of the application required some serious mantinence. There were uses of outdated APIs galore, a really funky custom DLL loading system (that they really just should’ve used .lib files for, like c’mon that’s literally what those are made for), badly written NTMakefiles, and to top it all off, at least 3 drivers written for a build system last supported in 2013. It’s so old that the tools to update from it literally do not exist anymore. Unfortunately I only found that out after reading this page which promises they do exist (no they don’t stop lying microsoft).

Anyways, as a status update, I’ve converted one driver by hand to build with Visual Studio 2019 .vcxproj files and I hope to get both setting up a computer to debug the driver and compiling the driver using MSBuild to work. Wish me luck so I may be free of this project I’ve already spent more than a week on and is showing me very dark depths of Windows.

Last Day Before Fall 2020 Semester

Sun, 30 Aug 2020 12:00:00 GMT

“I feel sad, so I will resolve a 60-file merge conflict in OpenAFS. I see no possible way this could fail to cheer me up.”

me, to me

(yeah i don’t really know what I was thinking. I am feeling a little anxious about tomorrow but hope that will go away when I start to work on school more, maybe)

As to not be Silent on Current Events

Fri, 28 Aug 2020 12:00:00 GMT

Going to out and say it: I don’t like the current state of America. Trump is not a good president, no matter now the Republican National Convention tried to spin it. As if it wasn’t already hard enough to trust America as a nation outside America (see here for just a small example) or a minority inside America (see the civil rights movement) Trump has exacerbated the problem greatly by near-constantly lying to make himself look good and anyone who disagrees with him look bad.

Of course, Trump isn’t the only thing I’d like to talk about here. I don’t know how many more political articles I’ll write, but at least for now I’d like to lay out my current ideology as well as stances on today’s hot topics.

I apologize in advance for not citing enough sources; I’ll add links later if I remember to.

Overarching Ideology

All humans deserve to live a comfortable life. With the amount of technology we have, there should be no physical reason why we can’t achieve this goal.
A government’s primary purpose is to support all the people under it’s domain so that (1) is fulfilled. Democratic, federal, tiered governments seem to be the best at this so far.
Market-based economic system assumptions are close enough to true in enough scenarios that they should stay in use.
A government’s secondary purpose is to control for the excesses of market-based economics, especially with respect to the environment and (1).
The scientific method of conducting experiments and collecting evidence and using good statistics is the best way to understand truth.

Black Lives Matter

I whole-heartedly support their goal: ensuring that African-American and other minorities can finally be treated as true equals to the white majority in this country. Decades of economic and social prejudice stand in the way. I’m not sure what a solution is, and also not entirely sure what I can do besides make sure I am aware of the struggle and provide support as I see it in my everyday life. It feels like I should be more proactive but again, I don’t think I’m in a position yet to do more.

Protests and Police Brutality

Over the past month or more, there have been many protests for a specific policy target of BLM: the disproportionate use of force against minorities. George Floyd, arrested for supposedly using a $20 counterfit bill, died after a police officer very harshly put him in a hold with a knee on his kneck. Protests over his death caused more death at the hands of police officers, and the situation keeps spiraling. The most recent death publicized was that of Jacob Blake, shot 7 times in the back in front of his children for “resisting arrest” while walking back to his car. These deaths sicken me to know that there are people with so little regard for the life of others, often racistly.

A rallying cry has been “defund the police.” The protest have seen a massive backlash by the police, with many videos showing excessive use of teargas, pepper spray, rubber bullets, and other “nonlethal” weapons that nontheless cause extreme discomfort. Reporters have been teargasses and even arrested while following police instructions (one of the more extreme videos I saw was an entire news team being slowly arrested while the camera was still rolling on live television). These uses of force also disturbed me.

I’m not sure if I advocate for fully defunding the police. I do think it is a very reasonable suggestion to cut back on police budgets so they can’t afford Ferraris or stockpile military-grade equipment. I also think treating the causes of crime via community support effots is also an excellent plan. This is where my certainty stops, because I don’t have enough information of the efficacy of community support programs or how much they cost or what one would even look like. I am certain that the theory around policing needs to change from “we are the only thing stopping criminals from taking over” to “how can we make everyone in our community feel safe and respected?”

Ok that was enough for one post. In the next one like this, I’d like to tackle QAnon or maybe free speech. See ya then.

A Paper I Disagree With

Tue, 25 Aug 2020 12:00:00 GMT

A summary: the paper in question talks about the theory of programming, what causes programs to become “complex” and hard to maintain, and proposes solutions towards that end. I disagree with the paper’s premises about what constitues programs that are hard to reason about.

The paper: Out of the Tar Pit by Ben Moseley and Peter Marks. For the optimal blog-reading experience, I recommend reading through section 7 before continuing below. It’s a bit of a stretch to call it a paper even though it has the same structure, it reads more like a blog post.

What I took away from it was the following argument: current languages and programming methodologies create more “complexity” than is necessary to actually solve problems, and a certain method of “Functional Relational Programming” can alleviate much of that complexity.

The authors are clearly biased in their argument. Just read this section on page 12 (emphasis preserved):

Power corrupts What we mean by this is that, in the absence of language enforced guarantees (i.e. restrictions on the power of the language) mistakes (and abuses) will happen. This is the reason that garbage collection is good — the power of manual memory management is removed. Exactly the same principle applies to state — another kind of power. In this case it means that we need to be very wary of any language that even permits state, regardless of how much it discourages its use (obvious examples are ML and Scheme). The bottom line is that the more powerful a language (i.e. the more that is possible within the language), the harder it is to understand systems constructed in it.

This is a blatant attack on the most common systems language known to be responsible for large classes of bugs: (you already know what language I’m talking about) C and C++.

Now, I do agree that the manual memory management done by C and C++ is horrible in terms of safety guarantees. However, I very much disagree with the assertion that “state should be avoided at all costs.”

The authors make a later claim about how allowing stateful code leads to complicated informal reasoning. On the contrary, I postulate that for many algorithms, a state-based approach makes the most sense. An example given in the CMU 2023 discord was the BFS algorithm. A traditional approach is as follows: use a stateful queue of (node, depth) items and a set of explored nodes. The queue and set woudl be initialized with the starting node. While there are nodes in the queue, pop a node, add it to the explored set, and push all unexplored neighbors back on the queue. This makes sense to me and many other programmers used to thinking imperatively. Heck, I’d argue that, with explanation of the motivation behind the algorithm, it would make sense to non-programmers as well.

Now consider the functional approach. You either have an unreadable mess like

… or you end up re-implementing a mutable state with immutable arguments and a tail-recursive function.

Now, this is just one example. I concede that there are many other algorithms better implemented in a functional style. That doesn’t necessarily mean we should restrict the language we program in to be nigh-unusable as to guarantee safety. Performance is important too, although nowadays less micro-optimizations (those are better taken care of by a smart compiler or library authors) and more algorithm design. If you can’t write a smart algorithm due to restrictions by the language, then the benefits of safety are less clear (to me).

In any case, I don’t think the authors are necessarily concerned with performance at all, and consider it a secondary goal at best, as evidenced by the following quotes (page 27):

It is worth noting that because typically the informal requirements will not mention concurrency, that too is normally of an accidental nature. In an ideal world we can assume that finite (stateless) computations take zero time and as such it is immaterial to a user whether they happen in sequence or in parallel.

(page 36):

It is worth noting in particular the risks of “designing for performance”. The dangers of “premature optimisation” are as real as ever — there can be no comparison between the diculty of improving the performance of a slow system designed for simplicity and that of removing complexity from a complex system which was designed to be fast (and quite possibly isn’t even that because of myriad ineciencies hiding within its complexity).

I agree with the second more than the first, but my point still is that the types of languages the authors advocate for (Functional Programming, Logic Programming) and the types of language constructs they wish to minimize (if statements, loops, stateful behavior) by themselves will not lead to better code. It might make code that’s easier for mathematicians to reason about and prove “correctness” of, whatever that means, but it will not make the code simpler by any margin.

Another key point I’d like to make before I wrap this up: Functional and Logic programming do not capture they way that computers operate in the same way that Imperative programming does. This might be a niche view, but I believe it is necessary to understand how computers work if one wishes to program them. That view seems to have persisted in computer science education thus far, where imperative languages are taught to beginners, probably as a remnant of when programming was more of an engineering discipline.

Having been exposed to Functional Programming, I think that knowledge of its techniques are sorely needed in computer science education. Writing safe, provable, understandable code is a worthy goal. I remain unconvinced that the “Functional Relational Approach” is the best way to go about this.

In case you haven’t had enough of Functional Programming shills spouting techical jargon at you, just read the Wikipedia article on Monads. I mean seriously, who thought this was a a good explanation for people who don’t understand it already?

Cool Videogame: Skullgirls

Sat, 22 Aug 2020 12:00:00 GMT

Here’s a videogame I’ve been playing a lot recently: Skullgirls (warning: some of the characters have… questionable clothing, so only click the link if you are not at work).

It’s the first fighting game I’ve played, and as it turns out it’s actually very beginner friendly. The tutorial is set up to bring you from knowing nothing about the game into being fairly ok. Some of the harder tutorials require a lot of grinding (looking you at 3-4 and 4-7) but are so worth it for knowing how to do the timing for more compilcated combos.

As of now, I’ve completed the entire beginner tutorial and two of the character-specific combo tutorials for Filia and Big Band. I am still nowhere near the level needed to compete online, however. I’ve also watched a lot of professional gameplay and studied other game techniques.

The most other effort I’ve put into a game like this have been for (checks Steam library) Kerbal Space Program, Factorio, and Super Smash Brothers Ultimate. This sort of reflects my changing tastes through the years, moving from building games to fighting games, but I still do like those old games and no longer play them kinda only because I’m a little worn out from them.

One of my friends also plays Skullgirls, although he’s not quite as into it as I am. I think it’s good to have a hobby that’s purely for fun to enjoy with friends, even if most of my time spent on it is improving my own skills in training mode instead of playing online. I hope to find more people at CMU that also enjoy fighting games and potentially get good with people at Guilty Gear Strive once that releases.

Anyways I didn’t write about Hakyll this time, that’s because I worked on another programming project today. I think I’ll post about that tomorrow. See you then

Upcoming College Semester Plans

Fri, 21 Aug 2020 12:00:00 GMT

So, as you may know from the about page, I am starting my sophomore year at Carnegie Mellon University this fall. As you may have guesses from the context of this post and the state of the world in general, this semester will be entirely online.

It’s not all bad. I get to stay with my very nice family (who were very cooperative during my work from home for Aerospace) and eat good meals and be in a house nicer than a college dorm and have a car and dogs. However, I do lose the in-person experience of hanging out with fellow college students & teachers.

The scales there are definetly tiped towards appreciating my current situation, but I do sincerely miss spontaneous interactions with friends. Don’t tell anyone but there were two girls I was trying to get to know better and it might be awkward to just pick it up again with them after all this time but idk, I’ve never done stuff like this before.

Anyways enough virgin-ranting and more talking about real Plans, the title of this post.

I’m taking 5 classes (~50hours/week), working part-time (~5hours/week) as a university tutor, and also working part-time at Aerospace (~5hours/week).

Funky schedule? ~~Leave your thoughts down in the comments, make sure to like and subscribe!~~

Oh yeah and I also plan to keep working out & doing any other hobbies that keep me sane. So my work’s cut out for me, that’s for sure. I’ll definitely have more to say about that once it all gets going.

Haskell Packages are Pretty Ok and Other Ramblings

Thu, 20 Aug 2020 12:00:00 GMT

After writing the code for my previous blog post, today I put it into a separate Haskell package so my code doesn’t smell too bad anymore.

I also fixed a small CSS error that caused the font size on mobile to be waay too big. Forgot the . in 1.8rem *facepalm*.

Oh yeah I also installed a LaTeX package so now I can do cool stuff like

f(x)=1−x2f(x) = \sqrt{1-x^2}f(x)=1−x2

while writing these posts. Which will hopefully be useful in the future as I talk about math at school more although I don’t know how much I’ll be able to post then as I have both a heavy course load and two small part-time jobs that add up to a fairly substantial part-time job.

Also you might be thinking, “Hazel you have only posted about this darn blog for the last n days, isn’t it time to write about something else??” To which I say, yes, you are absolutely correct, it’s just that I usually write these posts right after working on the code for the site generator so that’s usually what I’m thinking about at the time lol. We’ll see what happens after I stop work on the backend and actually get to a “post grind” lol

Hacking on Hakyll for better Pages

Wed, 19 Aug 2020 12:00:00 GMT

In case you are one of the few people who know about this blog, you might have noticed that I changed from titles of posts as the HTML filenames to just the “post ID” as the filename. This may not seem like a big change, but oh boy did Hakyll require some funky upgrades to get it to work.

Basically, I re-implemented Hakyll.Web.Paginate, but with logic customized for single pages instead of pages that are actually groups of other pages. I also added title metadata extraction instead of just page number + url extraction to the context creator for extra fanciness.

The backend is messy, but in the end it turns out very nice and “declarative” whatever that means:

    pages <- buildPagesWith "posts/*" sortChronological

    pageRules pages $ \index _ -> do
      route $ pageIndexRoute index
      compile $ do
        let fullPostCtx = pageContext pages index `mappend` postCtxWithTags

        pandocCompiler
          >>= loadAndApplyTemplate "templates/post.html"    fullPostCtx
          >>= loadAndApplyTemplate "templates/default.html" defaultContext
          >>= relativizeUrls

Not to be content with that small victory, I also added pages to the archive with Hakyll’s original Paginate thingy, also adding title support on top of that. Again it was really messy but also really worth it.

Anyways I am really getting my value out of Hakyll for both reasons: I’m having a very snappy blog (static sites >>> Wordpress any day) and I’m also learning Haskell fairly quickly which is neat. Tomorrow I will probably reorganize my code into better-named modules, but for now I’m happy.

Haskell Takes Way Too Long To Compile

Tue, 18 Aug 2020 12:00:00 GMT

Wow that was rough.

I left a stack build command running overnight on the machine I was planning to host this static site from, and apparently the Haskell compiler just needs exorbitant amounts of memory to operate properly. I mean to be fair running two compilers at once with only 2GB memory available is bound to cause some problems sure but not “your system is literally dying and can no longer handle network interrupts” levels.

I was able to fix the problem by compiling the Cabal and pandoc packages individually, because both on their own were using the full 2GB and really did not like being run at the same time. It was running overnight (15 hours) and still didn’t finish.

In the end I still spent 4 hours compiling stuff. Was that worth it? remains to be seen. but now there’s a blog so i guess so

Blog Theme Upgrades!

Mon, 17 Aug 2020 12:00:00 GMT

Wow the blog sure looks snazzy now! I ripped the twenty-fifteen theme from Wordpress and applied it to my new Hakyll blog and tweaked it a little and now it looks pretty nice if I do say so myself. I’m sure I’ll find more ways to tweak it, but for today I am done.

Also I’ve been doing art more so maybe I’ll have a piece to post here one day? Who knows. anyways I need to find a place to host this soon so ppl can actually see the improvements and content instead of just me. Although I am running this blog just for me anyways. idk we’ll see.

Blog is now on Hakyll

Sat, 15 Aug 2020 12:00:00 GMT

Hi! This site looks a little different how. Just for fun, I’d thought I’d move away from Wordpress (MySQL is very slow and PHP is not a great language and also Wordpress itself has a bunch of vulnerabilities) and towards a static site.

As you may have guessed from the title of this post, I’m using a system called Hakyll to make it. Hakyll is written in Haskell which is one of the worst language’s I’ve had the pleasure of working with. It was a challenge to get the tags section customizable, among many other issues I had with learning Haskell in general.

Also, I’m still doing art! I have improved a lot but still not enough to the point where I’d feel comfortable cross-posting to something else here. Maybe one day I’ll muster up the courage to put some of my figure drawings here.

Oh yeah also Coronavirus, crazy right? Don’t worry I’ll have a post about that soon.

I'm actually working now

Thu, 06 Aug 2020 12:00:00 GMT

I was lucky enough that my summer internship let me work remote. I am double lucky that they’re going to let me work part-time during the school year. Great company, great managers + co-workers, and I get to do AI research. So far, this is pretty much everything I wanted in a job!

P-values are wack

Sun, 02 Aug 2020 12:00:00 GMT

Like look at this: https://en.wikipedia.org/wiki/P-value#Calculation

Literally just “plug your data into a formula and get out a p” how crazy is that? Really I can only make sense of https://en.wikipedia.org/wiki/Fisher%27s_exact_test but even that’s a little tough to understand. Maybe I just need to study statistics more…

I'm back baby

Sat, 01 Aug 2020 12:00:00 GMT

Wow, a lot sure has happened since I went to college and stopped updating this blog! Going to college meant I couldn’t leave my old laptop plugged in and constantly hosting on a free domain like before (sad), but now I re-discovered that my high school has free hosting (yay)! So I exported my blog here, where it will stay until I get kicked off and have to pay for hosting somewhere else. Thank goodness Wordpress is a stable piece of software.

Hopefully in the coming few days I remember to post more frequently again. I have a couple cool projects to show since last year ;) Stay fresh, anyone reading this!

QOTD: 14

Wed, 14 Aug 2019 12:00:00 GMT

“Guess this is the last time I’ll see you for a while, huh?”

-My friend, tonight, at our last get-together before college

(I’m so sad but literally nearly everyone I know my age is going through the same thing, so at least I’m not too alone in that)

Going to College

Wed, 14 Aug 2019 12:00:00 GMT

Hey so sorry I haven’t posted lately. There’s been a lot on my mind but as usual I’ve been too lazy and out of habit to write any of it down here. So here’s a few snippets of recent thoughts before I head into the main topic:

I got my custom allocator to work, and submitted the resulting program to CodeGolf.stackexchange.com for a sweet-ish +60 rep
I did a little PicoCTF as warmup for (hopefully) meeting the Plaid Parliament of Pwning @ orientation week
8chan in the news over being shut down, I think it’s sort of scary that free speech can be suppressed on the internet but at the same time find it totally justifiable to kick them off. Main thought: “No one is entitled to let you use their property (servers) to express hateful speech that would be prosecuted even in a public space.”
I’m walking a neighbor’s dog while they’re on vacation, a golden retriever that is on the down slope of aging and very needy with wanting to sniff.

So now that that’s out of the way, let’s move on to what I think about going to college: I just want to get it over with. The last thing I wanted with this whole process was some sort of limbo where I feel like half there and half here. I would have preferred to do it in one day, like snap and I go from being here and comfortable to being there and not exactly comfortable but getting used to the place. I know that sounds silly but one can hope. Instead, the feeling of “last time I’ll do this for a while” just keeps lingering and lingering and it’s hard to stand it, so I just avoid thinking about it like a dumbbutt. I still need to pack a few things, most notably all the stuff I love on my desk. And even when that’s away it’ll still be a few days until I’m actually fully unpacking it.

Dang nabbit I just want to be there already. I want to think about making friends and walking to classes and having lunch and living in my own instead of packing and photocopying forms and yeckh this is a total #whitepersonproblen because a whole load of people would kill to go to this college and have to opportunities I have, and instead I’m complaining about it like a looser. I just want to be there already. That’s too much to ask.

Programming a Custom Memory Allocator

Tue, 06 Aug 2019 12:00:00 GMT

Uh, it’s super hard and I failed in the 2 hours I gave myself. The basic idea was this: keep track of the current block of memory, its size, and the index. If the index ever goes out of bounds, allocate a new block twice the size of the current one and copy everything over to that.

I first encountered a double-free error. Just wanting to get something to work, I commented out the code that freed the old buffer knowing full well that would cause a memory leak. No dice, now we have a use-after-free. No problem, silly me, I forgot to change the pointers within the data (lots of linked lists), so I’ll just calculate the offset and add it to everything. Still nothing and now we’re back to the double-free. I honestly have no idea what I did wrong but maybe it’ll be clearer in the morning. If I can’t figure it out soon, I’ll just go to keeping track of all the malloc-ed memory in another art and free that iteratively (I was getting a stack overflow error when trying to free it recursively, which was why I build the allocator in the first place).

Visualizing Perspective

Sun, 04 Aug 2019 12:00:00 GMT

So here’s an excercise for you reader: picture a cube with a face flat towards you. Got it? Ok, now rotate it 30 degrees about the up-down axis (could be y or z depending on which you’re more used to). Can you still clearly see the cube? Ok, now rotate that already rotated cube 15 degrees around the left-right axis. Can you still picture what the cube looks like?

See, here’s the thing: I can’t. Even after one rotation I have trouble focusing on exactly what the cube looks like in my mind’s eye. Two rotations and it’s really game over. Logically, I can understand there has to be a small kite, a medium kite, and a larger kite (from geometry, a kite is defined as a quadrilateral with two pairs of adjacent equal length sides), and I know where those shapes are, but for the life of me I can’t visualize what they look like put together. It’s like I hit a mental roadblock. It’s annoying.

I’ve been thinking about that for a while, how my intuition about 3D space doesn’t translate well into drawing it. Because I’ve been having trouble with it, I challenged myself to do an art piece with a cool perspective effect. It’s hit-or-miss so far, but I’ll keep at it.

Pokémon

Wed, 31 Jul 2019 12:00:00 GMT

Me and Pokémon, like any child of my generation, or the previous, go waaaay back. My first real video game that I remember is Pokémon Pearl (which I will forever consider to be the purest, most natural form of Pokémon games). I will never forget the exhilaration of besting the elite four and champion for the first time with the team that I had grown so close to while training. One of my favorite Saturday morning cartoons was Pokémon, and our family even bought the Pokémon DVD movies for us to watch in the car on road trips.

Pokémon has awesome monster with cool attacks, so it appeals to little boys. Pokémon has a good message and doesn’t show explicit violence, so it appeals to parents. Anyone can adore the cutest Pokémon. As a franchise, Pokémon has consistently pulled off the collectible RPG battler so, so well. Any deviations from the formula are met with resistance (Pokémon GO), although some are improvements (the move to 3D).

I always admire how Nintendo and Game freak are able to keep the same core game so fresh by adding new worlds and creatures and people all the time. There have been a few complaints about the direction the games are heading in (too linear, designs are too cutesy and have deviated from the “realism” of the original entry) but overall I believe Pokémon will stay good. It already is a global force of pop culture, and I hope it will stay that way for generations to come.

Staying up late (again)

Tue, 30 Jul 2019 12:00:00 GMT

So one reason I haven’t been posting as often is that I’ve kinda fallen out of the routine of it. Why? Because my daily routine has changed.

Recently, I’ve been streaming art for longer, and starting later, pushing logoff time to 10:30PM to even 11:30PM. This is not good. The reason for going longer is obvious: the 1 hour limit I was struggling to hit earlier is now a breeze because I feel comfortable doing art for longer now. The reason for starting later is less obvious: my sister and I now watch anime together every night, cutting in to the 9PM start I aim for. As a result, I feel less inclined than before to stay up later thinking and writing “deep” thoughts on my phone like I used to just a few weeks ago.

I love summer in all the freedom of time it provides, but I really need to get my sleep schedule back of track.

screw you too, respondus

Sat, 27 Jul 2019 12:00:00 GMT

Beating Respondus LockDown Browser

Sat, 27 Jul 2019 12:00:00 GMT

It was actually easier than I thought. Now, it wasn’t the easiest, but I will explain what I tried and what worked.

So Respondus LockDown is the browser my school forces us to use to take “secure” online exams. I hate it. I despise it strongly. Everything it stands for, the institutionalized test-taking mentality, the enforcement of unrealistic conditions, the distrust of students most of all, it ticks me off to my core. So I did something about it. I “hacked” their stupid browser that fullscreens itself, disables all other applications, limits the input options, and watches your every move with your webcam.

There were two ways of going about it: the first and most flexible option is to run LockDown in a VM. The other option would be to MITM myself to have an <iframe> of google constantly beside the test. The second has the disadvantage of more easily being caught (screenshots and keylogging, duh), and being more work for me. Also I didn’t want to ever install that piece of malware on my actual computer. So I went with the first option.

I spent many hours researching VM detection techniques and their associated bypasses, accumulating in finding and understanding a github repo that generated scripts to hide the fact that a VM is a VM. The one I used is located here: antivmdetection. I first tested LockDown in a untouched Windows VM, and sure enough it knew. So I ran the scripts, and voila it worked. What the scripts did was change all the VirtualBox device driver names, like the BIOS model, the video card name, the hard drive serial number, etc. to the exact same ones as my host computer. It changes those through VirtualBox settings and through registry editing. There’s also a neat trick with the cpuid instruction that’s explained much better here.

Other approaches that I tried in the VM but couldn’t get to work were using a sandbox (LockDown didn’t like the sandbox management program running) and disassembling and editing the LockDown binary (I was just bad at that tho lol).

IMPORTANT! please note that I did this only as an experiment to see if I could because I was mad, I do not intend to use this to cheat and I will take all my exams in the correct fashion, with LockDown installed on my main computer.

Being rushed

Tue, 23 Jul 2019 12:00:00 GMT

I don’t really like it. Other people sometimes do better under pressure, but not me. I like to do things at my own pace, way in advance, so I’m fully ready when the time comes.

Which is odd when paired with another of my tendencies: to be terrible and managing long-term deadlines. If it’s not in the next week and/or I don’t care about it, it won’t get done until I’m prodded. I consciously try to counteract that because it’s a bad habit.

With deadlines I do care about, however, I get stuff done right away, early like mentioned earlier. Like in Thunderbird, I do now have a calendar to keep track of the stuff I don’t care about, and I try to check it daily. We’ll see how that works out in college.

Oh also I got terrible course scheduling for college, I did it way to late so all the good time slots were filled, shoot. I was bad with managing that deadline, I really shouldn’t have been…

QOTD: 13

Mon, 22 Jul 2019 12:00:00 GMT

“I fear for the future of your Twitter/DeviantArt account.”

-My friend after I told him my secret to getting motivated at art. (hint: it’s not spite)

Haven't shaved

Mon, 22 Jul 2019 12:00:00 GMT

…my stubble for like 3 days now. Man do I look scraggly. It’s also itchy. I don’t know how people with beards do it.

Honestly, shaving every morning isn’t even that hard. It takes 2 min max and gives good results for the effort put in. Maybe not the best results, because a 5’oclock shadow is a real thing (hair grows back!) but good enough.

I don’t know where I’m going with this, other than having a full, professional-looking beard is probably even more work to maintain than a clean shave. I wouldn’t know. I don’t plan on findings out either, definitely shavings tomorrow morning.

Past 10PM

Sun, 21 Jul 2019 12:00:00 GMT

After I’m supposed to go to sleep, but am still awake, I turn into a different person. I’m more outgoing. I care lesss about long-term repercussions. I slur my words more. I’m crazy and conscious of it yet I love it and can’t stop. No, I’m not implying that I drink alcohol (hate the stuff) or do any other sort of drug at night. No, the night is my drug.

So anyways that’s how I just confessed I was horny while writing this post. Blehck.

Edgy Jokes

Sat, 20 Jul 2019 12:00:00 GMT

They’re good when they’re done right, but absolutely awful when done wrong. My friends and I got together and the group tendency leaned toward the edgier side while playing the Jackbox party game, so we got to see the whole spectrum.

A good edgy joke makes you think. It’s not edgy for edginesses sake (although it can be made with that in mind), it’s edgy in a way that makes you question your preconceived notions about the world. Take the winner of the TeeKO game. It was a simply draw, smiling potted flower with the caption, “Big Brother is watching.” By incorporating a pop culture reference about government and now corporate surveillance of the masses with the kind of cute, childlike imagery used to masquerade or deflect questions about the massively trove of data being amassed on every citizen, the shirt design spoke perfectly to our woke teenage brains.

A bad edgy joke told later that night was during Fibbage, when someone made a crude reference to slavery. It was not well put together, didn’t twist the viewers’ expectations in any meaningful way (can hardly be called a joke if it doesn’t do that), and could even be interpreted as complicit with the abhorrent practice.

Finding the right balance between joking about something and being deadly serious about the issue is the essence of an influential edgy joke. At least that’s my opinion. This whole post is kind of a joke in and of itself, but I Don’t Care.

(Anyways update in the Visual Studio thing: after 3 more hours, got it to reliably compile! Yay now I can start actually developing.)

Visual Studio was never meant to be used with CMake

Thu, 18 Jul 2019 12:00:00 GMT

Today I was locked in bloody struggle with CMake + Visual Studio, trying to get a small-ish open source project to compile. After 3 hours of non-stop effort, I can say that I have been defeated for now but am not giving up hope. I got past all the stupid cmake errors and the stupid dependency errors but am stuck on a stupid linking error (seriously? Why is it trying to link to a binary that I never told it to? and whose name is only slightly different from the one it should link to?)

It could be worse. At least this software was meant to be built on Windows with MSVC. And has a binary release in case I give up. But I won’t. Probably.

Is Metacognition Smart?

Wed, 17 Jul 2019 12:00:00 GMT

Metacognition (n): Thinking about how you think.

First answer: maybe. Why? The weak argument of: I seem to be smart enough, I practice metacognition, ergo. Unless I conduct a metacognition study, I’ll have to be satisfied with a sample size of n=1.

Second answer: maybe not. Why? Because, can higher level thought processes really change the deeper ones that control how we think? My opinion is, once you get going down a certain way, it’s really hard to get off that train of thought. My (yes, limited, thanks for noticing adult me) experiences show that only other experiences can actually change how your brain works and make it smarter. It’s very hard to imagine any new way of thinking until you absorb an idea or emotion externally. (Well then how do people come up with new ideas? Slowly and with lots of synthesis from old ideas).

Or, maybe, I still don’t understand enough about how I think for metacognition to be useful for me yet. Come to think of it, thinking about math or computer science or English were all just another skill that I got better at by repeated, structured practice. An actual philosophy/psychology course is in order, then? I’ll think about it.

Anyways sorry for the rambly post, I kept thinking of new things half-coherently in the middle of writing. It feels good to write down these (mostly stupid) ideas sometimes, even if no one else reads them.

QOTD: 12

Tue, 16 Jul 2019 12:00:00 GMT

“OMIGOSH me too!”

-Cute girl at the mall (We met in the bookstore near the graphic novel section, sharing an interest. I rode the high of mutual flirtation for a solid 2 hours. Still thinking about it. What is wrong with me)

Operating System Shopping (2/2)

Tue, 16 Jul 2019 12:00:00 GMT

Previously: Operating System Shopping (1/2)

So, where I left of was, there are a lot of bad things about all of today’s popular operating system kernels. Let’s just roll through the list:

Windows: pile of legacy garbage.
XNU: unusable except in MacOS, which is a walled garden nearly impossible to install on non-Apple hardware.
Linux: wacky monolithic kernel, very little inter-distro compatibility, little focus on usability for users.
BSD: like Linux but with even less hardware support.

Really, what I would like to see is a viable Microkernel that can be used to run a functional desktop, but that’s currently beyond today’s development. Turns out all the inter-process communication they require is a big hit to performance, even if it’s also a win for security, so people just haven’t developed them as much. A look at the current contenders:

GNU Hurd: promising, development stalled but might get back up soon, but it’s GNU.
MINIX: probably has no hardware support for anything made this decade.
Plan 9: I’m not trying to run a distributed OS
Haiku: might actually be the most functional of this bunch, but still missing a lot of tools[citation needed]
Redox: super promising (I adore it’s “everything is a URL” scheme), but isn’t even self-hosting yet.

Ok so it’s easy to point out the bad. Now what about the good?

Windows: can actually run legacy applications with ABI compatibility, unlike most -NIXes.
MacOS: UNIX + commercial success and support = greatness
Linux: very performant, perfect for servers, you are very hard pressed to not use it as your development OS if you’re doing research.
BSD: very stable, OpenBSD is probably the most secure OS around.

In the end, I’m probably going to stick with my Windows setup and get all my Linux development fix through MSYS2 until I realllly need a Linux dual-boot. I’ve allocated 150GB on my 1TB HD for another OS install and am leaving it blank for now. I’m also seeing if I can get Redox to build on my older Linux machine in the meantime. I’d like to try kernel development, but at the same time the deeper I look into it the more I realize just how out of my depth I am. Nothing to do but keep pressing forward and learning more, though.

Operating System Shopping (1/2)

Mon, 15 Jul 2019 12:00:00 GMT

Recently I’ve been looking at operating systems to use on my Laptop. It has an Intel processor and NVIDIA graphics coprocessor, just like the majority of gaming laptops out there.

The operating system it came installed with, and the one the most people are familiar with, is Windows. To be honest, Windows is a mess of an operating system. The kernel is full of hacks and workarounds to accommodate for legacy code, the whole registry concept is ginormous and unnavigable, and good luck trying to cleanly uninstall everything. History shows that Microsoft ensured their OS’s dominance through guerrila corporate tactics of allowing MS-DOS to be licensed for multiple hardware vendors, gaining followers in the business realm through those vendors and some killer apps, then forcing those same vendors to ship Windows when it released, furthering the cycle. Simply by virtue of being the most prominent OS early in the business realm, (not the first OS created, mind you), it became the self-perpetuating standard.

Another operating system I’ve had a lot of experience with is Linux. My favorite distro used to be Lubuntu (Ubuntu shipped with LXDE), but while I was a Sysadmin that chnged to Arch Linux. Arch Linux is the bleeding-est edgeing-est distro out there, lagging maybe a week behind upstream releases at most. The maintainers put a lot of work into making sure everything works together even when it shouldn’t (Python 3.7 + CUDA 10 + Tensorflow 1.14 not even close to officially supported, at yet it worked (not flawlessly but it did do stuff most of the time)). I had Arch Linux dual-booting with Windows on my old laptop and enjoyed it a bunch.

I’ll get into the reasons why I am OS shopping in the next post, as well as what I like and don’t like about the OSes.

Fighting off disease

Sat, 13 Jul 2019 12:00:00 GMT

Well shoot. I remember feeling exactly like this the last time I was vaccinated (1 month ago) but it always is a shock to feel just how bad it is. My shoulder is stiff, my joints ache, and overall I’m feeling super tired. (Man, just like an old man, huh?). I know I’ll come out the other side stronger and more disease-resistant, and wouldn’t even consider dropping vaccination, but still, why’s my body have to be so bad when fighting off disease?

That’s an easy question: when your body does fight off disease, the best thing for it to do is convince you to do is to expend as little energy elsewhere as possible. That means aches and pains to prevent physical movement and exhaustion to prevent mental exertion. Really, quite logical. It also makes perfect sense that when fighting off a disease analog that, mind you, whose whole purpose is to trigger the immune system so your body will react in the exact same way. The higher order functions of the brain simply can’t do much about it.

And here I am, lying in bed, rationalizing why I haven’t moved yet. Time to get up, I guess.

Morning vs Night

Fri, 12 Jul 2019 12:00:00 GMT

American culture certainly seems to favor staying up late. I can’t count how many times my friends bragged about staying up until 2am to finish a project. I can count how many times I’ve heard of someone getting up at 4am to do something, though: one. My ultimate frisbee captain last year would apparently wake up early to do homework and lift weights, and it showed. He was ripped and getting all the good grades.

Maybe that case can be attributed to night owls being more braggadocious than early birds, but still. There’s something about staying up late that’s easier for most people and looked on as the cool thing to do. Not me, unfortunately. At 10pm, right on schedule, it feels like my brain shuts down. Sometimes even before that. I go from moderately productive to a tired, forgetful, blathering mess. It’s partly a mental block but I really cannot stay up late very well. I would much prefer to get up early, where I can wake up more as I do more stuff. For me, at least, waking up at 7am vs 9am gives about the same level of exhaustion later in the day, only in the first case I have two more hours to do stuff.

So, uh, the point is, I didn’t really enjoy writing these posts at 10:30pm every night so I will try to write them in the morning instead. We’ll see how it goes. Sorry for slacking off since the end of vacation; unlike art, I didn’t immediately pick this habit (or Duolingo D:) back up.

Vacation End Post

Sat, 06 Jul 2019 12:00:00 GMT

Heyyy, I’m finally back from vacation! Yes, I had a great time, and yes, I saw lots of family and did the classic American things like play cards and eat barbecue and shoot fireworks and water slalom ski and yeah it was pretty awesome I’ll tell you what. I’ll elaborate more on some aspects later (little cousins + wakeboarding probably).

Looking back at my older posts… yeah uh I was really not kidding about that art thing. I still am not kidding and want to practice again today even though I’ve sat 9 hours in the car and am feeling blergh. Don’t have too many deep thoughts right now since my brain is friend from a Legend of Korra marathon in the car, so yeah I’ma just end this post here. See y’all bots tomorrow!

Vacation Start Post

Thu, 27 Jun 2019 12:00:00 GMT

So this will be my last post for a while. I’m going on vacation to Michigan with my family so I not get to use my phone before bed anymore. Really not much at all for that matter, because I will be prioritizing time with extended family instead.

Going up to the lake in Michigan has been a family tradition for who knows how long, at least since Dad was a baby even. This year we’ll me making a bit of a detour first in New York to attend my maternal grandfather’s funeral, which is tough, no one is prepared to deal with that kind of emotion with the loss of a family member. It’ll be strange to be all mournful and then cheery as we drive to the lake the next day but I digress.

Expect another post detailing my thoughts on my vacation after I get back. I hope the habits I worked so hard to build this summer, like Duolingo practice in the morning, Art practice in the evening, and blogging in the late evening aren’t disrupted by this week of other stuff. Anyways, see ya later!

QOTD: 11

Thu, 27 Jun 2019 12:00:00 GMT

“YESSSSS!!”

-Quinn, after magically guessing who Hitler was in the game Secret Hitler (i went over to a friend’s house today, we had a blast with a bunch of card games. Skipped art but totally worth it to keep these friendships intact)

Streaming Art

Wed, 26 Jun 2019 12:00:00 GMT

Eh I missed a post again, whops.

Anyways so yeah I’ve been streaming art the past couple of days very regularly. Most of the time I get my friend on in there, and we have a blast chatting, but for the past 2 nights in a row somehow I’ve had a new random dude jump in chat. I advertise my stream as a “chill stream” like 90% of the other streams on the platform, and somehow these guys chose me to converse with. I don’t know why; it can’t be the art quality, even those both seemed to be pleased with it; it can’t be my setup, the thumbnail isn’t amazing most of the time I don’t think. But I pride myself on audience interaction so I guess they stay around and even follow.

All of this is very nice for one big reason: having someone to talk to while I do art is very motivating. It a habit now, if I don’t have someone looking on then it just becomes boring. When I’m reading chat I can let my mind wander to other things besides how bad I think my art is (pretty bad, which it isn’t compared to professionals although pretty good compared to where I started out (I’m not done improving yet, though, my goal is to be able to quickly draw relatively high quality fan art like the more popular artists I see)). That puts me in a state of flow, or zen, or whatever you want to call it where your mind is able to focus on the task at hand without even considering outside distractions. Checking chat is just part of that flow for me now I guess. Cool.

QOTD: 10

Wed, 26 Jun 2019 12:00:00 GMT

—FAQ THREAD ABOUT THE NEW #Discord VIRUS—

-Well meaning twitter user @splitsplatted (It was more of a phishing attack than a virus, but the whole this is pretty interesting. Link here)

I read a book today

Mon, 24 Jun 2019 12:00:00 GMT

I read a book today. Started reading, to be precise. I should read more often; it was very enjoyable. First, though, I should probably find more active things to do so I’m not feeling bored and lethargic enough to start reading randomly. Or maybe that’s a bad way to think: reading is actually really good apparently, crucial even to a well-balanced life. So maybe I’ll do it more often now. Cool.

On Learning

Sun, 23 Jun 2019 12:00:00 GMT

(This is that essay I talked about. Copy-pasted in its entirety.)

As I transition from high school to college, I’ve done some thinking about how I best learn. After all, so I’m told, college is where students are challenged to pace themselves, the bulk of the work being done outside the classroom. This in and of itself doesn’t bother me; I can time-manage and slog through mountains of homework with the best of em. No, what bothered me is this: if I don’t know my own strengths, then I can’t play to them to save time or shore up my own weaknesses.

As an example, the other day, I tried to play the piano. It’s just notes on a page, those are easy to read, I thought. Individually, yes, after a good ten seconds of counting up or down from middle C. Then came the unanticipated challenge of getting my fingers to press the right notes without looking. Then came stringing all that together at a reasonable pace. I pressed through for a bit longer, learning to play a few bars, but then I got discouraged by lack of immediate progress. It was just another frustrating riff on other unfamiliar subjects I’d tried before, like art, Tetris, and basketball.

QOTD: 9

Sat, 22 Jun 2019 12:00:00 GMT

“I have another friend who’s also learning Japanese, and they already know all the Katakana and that bugs me.”

-My friend who is concurrently doing Japanese Duolingo with me (he’s a really nice dude, great friend, always watches my art stream when I invite him. We do lots of stuff together, expect more quotes)

Am I smart?

Sat, 22 Jun 2019 12:00:00 GMT

Signs seem to point to yes. No bragging, but I did get As in every class except B+ freshman Spanish, even in classes most people did poorly in. But there’s a predicament: I don’t feel smart most of the time. Maybe it’s me or maybe it’s a human problem, but I can’t really fully comprehend learning things any faster or slower than I already do. And maybe because I’m impatient, I feel like I learn things too slowly, leading me to believe I’m not smart.

And maybe that’s true, too. That I am not smart. That especially seems to be the case when I compare myself to some classmates who knew more science, had better insights in humanities, and could comprehend higher math. I don’t study very often, but when I do it’s the most intense, “memorize this or die” feeling ever. I can’t tell if my memory is awful (I’ll easily forget deadlines that are more than 2 weeks out) or fantastic (what are the main quirks that differentiate Python 2 and 3? Easy, the print statement is now a function instead of a keyword, range has been replaced by it’s faster brother xrange, float division is now on by default, etc.). Or maybe just extremely specialized for handling certain facts and concepts (math?) but not others (vocabulary, except weirdly enough most SAT vocab words).

Anyways the point I was trying to make but forgot to lead up to is that maybe the differentiator isn’t intelligence, but experience. And that scares me. Because right now, I may have the experience lead over some people, but everyone will catch up given time. So I don’t want to get caught on my butt, and I need to keep working to improve myself in the areas I most care about. Yeah.

I have an essay I wrote a while ago for school about something similar, I’ll post that tomorrow b/c I can’t keep doing this staying up late thing, I wake up too late gosh darn it.

Why short post?

Fri, 21 Jun 2019 12:00:00 GMT

Short post today because I stayed up too late programming. It was very fun but took longer than I thought to make a python script that fetched the currently playing song from Spotify. Also bummed there weren’t any working existing solutions. So yeah at least now I have something going, and ready to be used on my next art stream. I should’ve actually practiced art today tho. Oh well.

QOTD: 8

Thu, 20 Jun 2019 12:00:00 GMT

“Sour Skittles are actually legit, really sour unlike most things that have the more sugary sour crystals.”

-My brother, over some not-really-sour pool ice cream (We have very similar tastes when it comes to food, except I can handle more bitterness (vinegar, dark chocolate) and spice (wasabi, jalapeños).

Born

Thu, 20 Jun 2019 12:00:00 GMT

There’s a pretty philosophical question that’s been on my mind for a while: are people naturally born good or evil? The obvious noncommittal answer is that it varies, and the end result is also dependent on the state of society upbringing and such too. Then my next question is this: What, then, would be the natural tendency for someone in my same place to be, good or evil? Taking my parents and siblings into account, it seems the answer is good.

But backing up a bit, there still seems to be an unanswered question here: what exactly makes a person good or evil? There are two parts to the English ambiguity of “make”: the first is definitions, and the second is causes. For the purposes of this thought train, I’d like to define “good” as always doing things for the benefit of others, and “evil” as always doing things for the benefit of self. I realize this might be slightly different from a tradition definition of good and evil as moral vs anti-moral but maybe I’ve shifted the morals around a bit, no?

So the causes of good and evil: unclear. Genetics of the brain and upbringing must have something to do with it (wow so deep that’s literally everything it could be). I have no idea why some people seem to be inclined to help others, and others only themselves. Maybe it even depends per-situation? See that’s why I’ve been thinking about this. I want to be good. But I end up falling into my own definition of evil (see Being selfish). Yes other people want to be my definition of evil, and truly believe that is the best thing. I understand it, I think I do, but I don’t know. (As a side note, a lot of times conflicts can even arise between good and good people depending on who they want to help).

So there you have it. Some of thoughts on naturally tendency and good and evil.

What I want posts to be

Wed, 19 Jun 2019 12:00:00 GMT

I let the morning slip away again, only this time because I slept in. Felt really good actually. I shouldn’t be staying up this late because I want to get up early but yeah not having an alarm for once was nice (my phone died).

Wait a minute this isn’t right. I don’t want these posts to be just a collection of diary entries. I want them to be deeper thoughts than that. While keeping a daily journal is nice, it’s be nicer to have a record of personal philosophy and stuff. The minutae can wait. Not even I will want to read that probably.

So, I apologize for the decline in quality, and will get some better stuff going soon. In the meantime, more sleep. No quote today.

Scanning cards

Tue, 18 Jun 2019 12:00:00 GMT

My morning got off to a rough start. I was very lazy and wasted a lot of time on my computer instead of doing anything productive. I’ll try and remember how this feels and do better tomorrow.

Part of the reason why I couldn’t focus was I was having friends over. Around 4:30 my favorite ground came over to help digitize our common card collection. Called 1KBWC, it’s a really fun game we’ve had going since Junior year of HS. We recently hit the thousand card milestone recently with the help of the rest of our club members (we were the founders). Anyways yeah we had this whole setup going, digital scanner going, green screen for pencil cards that didn’t scan well, sorting by who made the card, the whole shebang. We plan to host a MediaWiki server with all cards in picture form along with tags for who made them and maybe categories of cards. We also did other things, like eat pizza and play Puyo Puyo Tetris (that’s another post), but the gathering was focused around card digitization. Which was good because we’d been talking about it for a while and needed an excuse both to get together again and to accomplish that task.

All in all, another fantastic summer day, non-ruined by rain.

QOTD: 7

Tue, 18 Jun 2019 12:00:00 GMT

“Hey guys, did you know that ‘portmanteau’ is a portmanteau of ‘portmanteau’ and ‘portmanteau?’”

-Noah, over pizza (he’s a funny dude. Also really good a Tetris)

QOTD: 6

Mon, 17 Jun 2019 12:00:00 GMT

“Let’s call it…Operation Studfinder! points finger at me and beeps”

-Dad, while getting tools to mount a TV on the wall (thank you Dad. Btw we didn’t finish mounting that TV yet)

Being selfish

Mon, 17 Jun 2019 12:00:00 GMT

I am way too selfish. Nearly every thought I have centers around me. Only with a fair amount of effort can I logically think through what others would want and then act accordingly. I don’t like this. Right now, it just caught myself again: my first thought as to why I don’t like this was, “it makes me feel self-conscious and like a lesser human being,” but the non-selfish response should’ve been, “I’m not able to do enough for others because I only think of my own interests.” Both are true statements, but one shows better character than the other.

I love talking about myself and the things I like, and will never pass up a chance to flaunt some new or obscure knowledge I have. I worry that can be isolating, like, “ooh look at mr. smarty-pants over there flexing again.” Am I over-thinking it? Should I watch myself more? Maybe to both. I care too much about how I come across, because again, I’m selfish.

I won’t pretend like being aware of my own faults is making me a better person. Even correcting for my mistakes isn’t enough. But I’m scared. Because I don’t know how to change. My hesitation around strangers, my habit of blurting facts, my selfishness, all seem ingrained. And I don’t know who to talk to, because all the people I trust the most I want to think highly of me so they trust me back. So I probably won’t reveal my deepest insecurities, because I’m selfish. Which isn’t ok. So I’m typing them here, for potentially everyone to see. Which is ok. Maybe I’ll change eventually and come back and wonder what I was so worried about. I don’t know. It’s too late for this and my plan of just writing out thoughts has worked because I have these thoughts every night but don’t share them. Anyways gn.

QOTD: 5

Sun, 16 Jun 2019 12:00:00 GMT

“Mom’s happy and she doesn’t even know it: I was this close to buying a new 4K TV we didn’t need, but none of them were the right size…”

-My Dad at Best Buy (he’s a rascal sometimes :P)

Father's Day

Sun, 16 Jun 2019 12:00:00 GMT

My dad is pretty awesome, not gonna lie. He works hard, imparts good life advice, is always patient with us rambunctious kids, and just is a great person. I hope my family was able to provide a good enough father’s day to keep up with his awesomeness. Me and my brother helped with fixing the TV sound setup downstairs and cleaning the old junk, we had lunch together while out shopping for electronics supplies, we watched the original “Mission Impossible” after we were done, and then we went to his favorite Indian restaurant for dinner. Then ice cream afterwards.

All around, a pretty productive and fun day spent entirely together. He seemed happy, but also kind of sad at times. Was he realizing that this summer will most likely be the last spent like this? Did he wish he could be doing other things but was too polite to say it? I don’t know. The first is definitely what’s sad about this summer to me, and the second is definitely within his nature. In any case, I should probably help finish mounting the TV tomorrow (we bought a new mount for the downstairs one). He deserves this kind of attention every day, as does every kind and loving father.

Today's Results

Sat, 15 Jun 2019 12:00:00 GMT

What I did today, in bullet point form:

Woke up late
Made oatmeal
Watched YouTube while eating said oatmeal
Played video games
Went over to a friend’s house for their graduation party. It was very fun, got to see old friends and make new ones. All in all a great experience with lots of physical activity, card games, and the staple of any teenage party: video games. Lunch and dinner were had there, tacos.
Did some art. Worked way too long on it for something I’m not terribly proud of but hey I practiced.

Result: I wish I could’ve gotten up earlier but that was another great summer day. Won’t have too many of these left nope. Gotta get together with friends soon.

QOTD: 4

Sat, 15 Jun 2019 12:00:00 GMT

“Yet in view of wider adoption of Mixed Reality (MR) technologies, surprisingly little attention has been given to the critical nature of the privacy and security implications of MR.”

-Microsoft, https://www.microsoft.com/en-us/research/blog/envisioning-privacy-preserving-image-based-localization-for-augmented-reality/

(that was actually a pretty good read, even if it boils down to, “turns out neural networks can extrapolate useful information from feature-labeled 3D point clouds, so we turned the point clouds into line clouds and now they can’t. Fancy visuals!”)

QOTD: 3

Fri, 14 Jun 2019 12:00:00 GMT

“I misread ‘volt padmount’ and thought it said ‘Voldemort’ instead”

The person who started the “Voldemort Transformer Moment” meme (some discord servers I’m in are purely to view their stupid funny memes)

Another story about trying

Fri, 14 Jun 2019 12:00:00 GMT

Hey I felt better today. Had a wacko dream but hey no more constant weakness. I even biked over to gamestop to pick up an Xbox controller, that was fun. Anyways yeah cool day, weather was perfect, spent some time outside with the dogs in the cool summer breeze, at the pool eating $3.50 hamburgers with my sister, and inside cleaning my desk. The desk is still cluttered but much less so. The pencil holder 3D thingy (yes that’s what that was) coming in clutch.

Art Hour (9:00-10:00 most nights) was disrupted by my sister’s dance recital (she was awesome, wouldn’t have missed it), but I thought I would make up for it my transcribing a few notes from one of my favorite songs in the 15 minutes I had remaining. Turns out no, my musical ability is super rusty. Even at my (still quite low) prime, I probably wouldn’t have made more progress. That is, to say, zero. I started out with the background chords, realized I don’t know any chords except for the standard major/minor 5th (where you press the 1,3,5 keys, is that right?), which these complicated ones were not. I then moved on to trying to decipher the melody. It should’ve been easy, there appeared to be only 3 notes, all within 1 whole step of each other, but even finding the “base” note was beyond me. In conclusion, I was woefully underprepared. Guess I’ll take another shot at it tomorrow and/or ask my friend for help.

QOTD: 2

Thu, 13 Jun 2019 12:00:00 GMT

“See you later, Hazel.”

Mr. White, Syslab Director TJHSST (I might cry, thinking about leaving TJ is too emotional for me rn)

A good day

Thu, 13 Jun 2019 12:00:00 GMT

With everything extremely sore, and feeling quite weak from the large immune response the vaccine yesterday prompted, today was not set to be a good day. But it was anyways, I’ll tell you why.

Today was the last day I set foot in my high school for a long while probably. I handed out thank-you notes to my favorite teachers, got to talk for an hour with my favorite-est teacher, and generally re-feel that nostalgic feeling of loving something but knowingly you are lettting it go but promising to never forget all the good times.

I also got to experience that mystical feeling of “flow” you get when you lose all track of time doing something you really love. For me today it was programming and art. Both sedantary activities, yes, but with my body screaming “no” to simply existing I didn’t want to risk it. When it got really bad (around noon) I just loaded up a few glasses of water and binge-watched 4 episodes of RWBY volume 6. Good show. Later, after programming but before art, I made dinner with my mom and sister. That was fun.

Anyways yeah it turns out mild physical discomfort can be mitigated by doing fun stuff. yey.

Something about Doctors idk

Wed, 12 Jun 2019 12:00:00 GMT

Got caught up on my Meninngococcal vaccines today. The bruises still hurt, as per usual, but I think in another 48hr I’ll be completely fine. The lady doing them was still in training but she was very nice. She talked about how she, at one point, worked full-time (60hr week!) while also going to community college full time, only quitting when work made her move 2hrs away from college! She said at that point she had already given up on sleep but the extra commute made it impossible. That’s hardcore, and again reinforces how lucky I am and how easy I have it. She’s now working for the medical center and taking online classes over the weekend to finish off her degree.

Oh yeah now I remember why I wanted to talk about doctors. The replacement one I had was a real kook, no better way to describe him. He seemed to be around 70, had a killer white mustache and beard, and constantly cracked pretty bad dad-jokes (I could keep up with his torrent, even firing a few back). He certainly knew what he was doing though, and fortunately the checkup went flawlessly. Turns out I am a healthy boiyo with no health problemos. yey. Let’s keep it that way.

QOTD: 1

Wed, 12 Jun 2019 12:00:00 GMT

Tell your parents your doctor introduced you to sex, drugs, and rock-and-roll, hah!

The doctor I had today (trying out a new thing at the suggestion of my friend. Basically posting a memorable quote from each day in addition to the usual post)

Sleep

Tue, 11 Jun 2019 12:00:00 GMT

I’m not entirely sure how I keep forgetting to post. I wasn’t that tired when I fell asleep. Or so I thought, before I ended up snoozing for 11 hours straight. I guess that means I maybe caught up on the sleep I missed during the ANGP on Saturday. Maybe not, thought, and now I’m used to just used to waking up late. Hmm.

Idk if I’ve mentioned this in a previous post, but I, like most humans, absolutely adore sleeping. I’m reminded of one meme where someone says, “You’ll spend about 1/3rd of your like unconscious, unable to do anything except think about things you have no control of. And the best part is, that’ll be your favorite part.”

That’s a bit of an exaggeration, but still, you’d think sleep would be useless. My classmates most certainly do. One of my particularly sleep-deprived friends measures his “sleep deficit” from 6:20hrs a night. And his chart still went into the red. Personally, I’ve found sleep to be like a secret weapon for learning things fast, better than staying up to study in fact. Not that you shouldn’t study at all, you most certainly should, but like after a certain point you’ve reviewed all that you can and sleeping on it and being well rested for the test the next day is the better option.

Anyways. I hope I’m both able to break my current summer habit of late to bed and rise and shift it to early, as well as keep that up in college. We’ll see.

Elementary School Reunion

Tue, 11 Jun 2019 12:00:00 GMT

Yup you read that right. My elementary school had a reunion party tonight to celebrate our current graduation from high school. And I went to it.

It was actually pretty fun. I had one friend, son of a family friend, who I had kept in touch with, but that was pretty much it. The rest I had never reached out to again (it was 5th grade in 2012, idt we had phones yet), let alone remembered. I was probably the worst there are remembering names and faces, so I’m grateful to my friend for busting out the old 5th grade yearbook and giving me a tour.

Most of them went to the same or a nearby high school, while I had moved in for just 5th grade and moved out the year after. I was an antisocial little kid, only making a few friend and sticking to them. Still am, come to think of it. I emphasize the point of having Friends, but in reality my mind just goes blank when I talk to strangers. I can think of anything to talk about, don’t know how they’ll react to what I say at all if I do say something random, and yeah. A couple people were nice enough to make conversation, girls even, and I got to meet up with an old friend who I’ll keep in touch with now again, so I won’t consider the night a complete waste. Plus the free Mexican was nice.

Anyways yeah if this is the first post you read, uh, I’m sorry, but a lot of the rest of the blog is like this. Check out the projects section if you want something a bit more professional.

ANGP

Sun, 09 Jun 2019 12:00:00 GMT

I just set foot in TJHSST for the last time in probably a long while. The All Night Graduation Party was held there, and I went, despite not staying up all night. (That’s also why I missed yesterday’s post). I had a great time, but my friend William had a good observation: despite all the wild revelry we had been having, everything was tinged by remorse and nostalgia and wandering thoughts about what had been.

I would’ve liked another “last time,” and will probably count the final time I stepped foot in the Electronics Lab and talked to Mr. Bell as my final moment in TJHSST. Gosh darn was that also sad, me and Alexander had to fight back choked throats and blink away the starts of tears as we left the wonderful room filled to the brim with any imaginable breadboard component, where so many memories of true freedom reigned, the place anyone could unequivocally point to and say, “this opportunity is what’s so special about TJHSST.” Reliving the experience is hard.

HOLY COW I ACTUALLY TOOK A PICTURE OF THE 3D THINGY FINALLY here it is:

Friends!

Thu, 06 Jun 2019 12:00:00 GMT

I highly recommend having some. I pretty miserable at putting myself out there to meet new friends, but once I find people with common interests and compatible personalities, I think we can hit it off fast. Maybe that’s my imagination, but the making a lot of friends part isn’t. Throughout nearly all of high school, I’ve had one cadre of 4 other really close friends. Today, we all met together at one of the group’s house for a small school-is-done party, complete with live-streaming sandwich making, playing Frisbee, Minecraft, and Monopoly. Really fun!

Friendships, like anything else, I think, take time. It takes time to learn how to make friends, and it takes investment of time to keep friends. Unfortunately I’ve slacked off on the first (already having good friends) and haven’t put as much time into side-friendships as I should have (again, due to a group of already amazing friends). My general strategy going into college is this: keep in touch with all of my best friends and some other good friends, join a lot of clubs and activities, and just meet as many people as I can under circumstances where we can just do stuff together with no pressure to be friends and hopefully by virtue of just time spent together talking about stuff something will develop. Very hazy, I know, but what can I do? Spend the summer trying to make friends with people that I won’t be seeing ever again? I don’t know how best to make friends yet (although I’ve gotten a lot of good advice), so we’ll see what happens. Later.

Ah yes here’s the paragraph where I talk about the 3D thing you aren’t quite sure exists anymore due to how long I’ve been putting off showing a picture of it. Well I spent so long at the party today I didn’t have time to take a picture when I got home. AND I forgot to do that in the morning rush. So there. I promise it’s really cool, maybe not worth this much wait but still cool nonetheless. See ya tomorrow.

A true eclectic collection of thoughts

Wed, 05 Jun 2019 12:00:00 GMT

The game: only reveal my thoughts online, and nothing physical. If close friends find out about this blog they will know me even better because I usually don’t share what I think.

Yay I was able to practice art today, along with everything else (including programming!), but at the cost of sleep. Thank you, friend who did art with me.

Wait a minute I have to get up early tomorrow. How am I going to fit it my morning Duolingo practice? Going to friend’s house in the afternoon, might be too late. We’ll see.

Oh yes the 3D printed thing. Good news: it finished! Bad news: I forgot to take a picture. Tomorrow!

Forgot to practice

Tue, 04 Jun 2019 12:00:00 GMT

Stuff I did get around to after school today:

Walking my dogs
Making dinner
Cleaning the kitchen
Exercise
College finances (a little bit, but it helps)
Color coding my Thunderbird calendars
Watching art videos

Stuff I did not get around to after school:

Practicing Super Smash Bros. (did a lot of that during school, shouldn’t matter)
Practicing what I learned in the art videos

I should be better about this…but I don’t want to sacrifice sleep. This post is usually the last thing I do before I go to bed (I know I know, unnatural light is bad, too bad).

Also, update on 3D print thing: what was supposed to be a 15hr print turned into a 30hr one (I knew I should’ve turned off infill), so it’s still going. I promise tomorrow.

Art Again

Mon, 03 Jun 2019 12:00:00 GMT

Haha yes Clip Studio Paint is truly a magical piece of Japanese software. All the brush and texture presets are very quality and emulate a lot of the classic manga stuff you see a lot.

Coming from Krita, there’s a palpable increase in professionality and polishedness of the whole system, even if the UI style isn’t exactly modern. I still need to get used to the key bonds but fortunately a lot carries over.

I know using a new art program won’t automatically improve my drawing skills, but at least now I don’t think I’ll have to try as hard to get certain things done. Plus I like the presets more. Too bad it’s paid, but I’m definitely willing to shell out a cool $50 for a professional program with (hopefully) lifetime updates.

Also, about the 3D thing I said I was going to show? Yeah uh the filament broke in the middle of the night around 1/3rd the way through the build, had to restart :(. Tomorrow then.

You FOOL

Sun, 02 Jun 2019 12:00:00 GMT

You thought I was done with these daily blog updates? HA, think again! I have so much more to complain about to no one in particular but am uncomfortable saying aloud because I have self-esteem issues and think my friends will think less of me if I bother them with inane problems!

Anyways, why the heck does it take so long to 3D print stuff? Cura says it’ll take 20 hours for this cool-looking pencil holder thing to get made. I don’t know if I can wait that long, but I already started the build, so we’ll see. Maybe tomorrow there’ll be a photo?

Senior Research

Sun, 02 Jun 2019 12:00:00 GMT

I don’t know if anyone else will read this paper, but here it is:

RLAIT.pdf

I spent a whole year on this. I learned a lot, but unfortunately don’t have a whole lot to show for it. Still cool and very fun and I would do it again.

Code: https://github.com/hazelduvall/RLAIT

Whops: the unnumbered

Thu, 30 May 2019 12:00:00 GMT

No, I haven’t been asleep since I last posted. I just forgot to post. Again. Yes I’m bad, thanks for asking. Making habits is hard.

Anyway, today’s topic: DRM. It sucks. I realized I’ve been trying to break it in a completely selfish way, but still, c’mon, please just yield to my halfhearted googling attempts already. I think I’ve got the full file (probably not, it was something stupid like removing the “Range” header on a request whose only authentication was in a temporary token and hmac included in the URL), but there’s a bunch of audio decoding errors after the first 10 seconds. Another stupid python script to get all the individual bits failed spectacularly (I think there was an extra bit returned by the server that complicated things? idk).

So yeah when people talk about how unhackable stuff like WideVine is, they really mean it. I was probably too lazy to even get to the part where encryption is a problem! Anyways yeah today I again learned just how much I don’t know.

Being tired

Sun, 26 May 2019 12:00:00 GMT

Oh boy am I tired right now. I can’t remember the last time I was this exhausted; it was probably during wrestling season sometime. I might have heat stroke, I don’t really know

The thing is, my “normal” tired is the tired you’d feel after you first wake up, that thing where you kind of want to go to sleep but your brain (if it’s unaddled enough) tells you that a little bit of movement will shake it off. I feel that in school all the time, sluggish a little bit but overall once engaged in something I’ll feel awake.

Not now. Now I have a fever, all my limbs are heavy, thinking is hard, I have the classic dehydration headache (as opposed to my other “screw you alergies” migrate), and overall just bleh. The weird part is I don’t have an appetite; I had to force myself to eat enough lunch and dinner. Physical activity seemed like it would make my body implode even more than it currently is, so I avoided it. I definitely spent too much time outside in the sun (it shows in my terrible-looking farmer’s tan too).

Anyway, game results: we only won our very last game in the tournament, clutching that sweet sweet 11th/12 place. Maybe I’ll talk about Wilson, the team that crushed us and probably won the entire thing, tomorrow.

Whops pt. 2: Boogaloo

Sat, 25 May 2019 12:00:00 GMT

Forgot again yesterday. Uhhh something about how short-hops in Suepr Smash Bros are hard to do in the heat of the moment.

Frisbee

Sat, 25 May 2019 12:00:00 GMT

I went to an Ultimate Frisbee tournament today. Played in, specifically. For those unfamiliar with the game, Ultimate Frisbee is a non-contact sport played with the classic Frisbee where teams of 7 players attempt to catch the Frisbee in the opponent’s end zone while blocking the opponent from doing the same to them.

It’s a good sport, and I had a lot of fun today even though now I’m exhausted. I like Frisbee because it’s something active to do with friends. Really any sport could be like that but the culture around UF is very welcoming compared to other, more competitive sports.

I guess tomorrow’s post will be about how we did in the tournament, we play 3 more games tomorrow too to determine our final placing.

Calendars

Thu, 23 May 2019 12:00:00 GMT

I don’t know why, but I haven’t gotten into any habit of using them. I know they’re useful, but I don’t check or put any dates on them.

Come to think of it, that’s probably a self-perpetuating cycle. Huh.

I’m usually fine with keeping all the important dates in my head. That’s worked out fine for quizzes and tests. Not so much for Senior Shirt Day (which was today) or other long-term planning things like internships. The former example is why I bring this up; missing the group photo, while kind of just novelty, sort of stings.

Putting things on the Internet

Wed, 22 May 2019 12:00:00 GMT

I try to avoid it, mostly. I’m pretty sure only 3-4 photos of me exist publicly online. I’ve also tried to separate my “professional” accounts from my “personal” accounts, although with enough digging (no I’m not going to tell you where) it’s possible to link the two realms.

This blog is actually on the “professional” side. Kind of strange to think about, I don’t do this kind of random thoughts thing anywhere else, even in private chat servers (except occasionally in DMs to friends). It’s kind of a load off to be able to just say stuff, even if it is to the loneliness of a self-hosted blog on the public internet.

Whops

Tue, 21 May 2019 12:00:00 GMT

I can’t believe I forgot to post yesterday.

Anyways, here’s yesterday’s thought: getting good at things takes time. So that’s why I try to do a lot of things every day. To get good at them all.

Art stuff

Tue, 21 May 2019 12:00:00 GMT

So I like doing art. I’m not great at it, but apparently practice has elevated me to above-average now so that’s fun.

One thing doing more art than usual has taught me is that good art, especially really good art is super hard and I totally respect anyone who can do that sort of thing.

I’ll still definitely stick with programming and computer things (because I love it) but art I think is a good secondary thing to have going.

Will I post any art here? Probably, as soon as I figure out how to clean the metadata. Most stuff is either posted elsewhere already (under a moniker), or hand-drawn, and scanning is a pain. Anyways.

OpenCV Chessboard Test

Mon, 08 Apr 2019 12:00:00 GMT

This was for my Computer Vision class. This thing was pretty simple once you understood it all.

The cube is drawn in perspective on the chessboard

The hardest part was getting the chessboard pattern even when the tissue was obscuring part of it. That could be solved by finding the closest point from the previous frame, only taking the ones with a certain min distance, and calling cv::solvePnP on the filtered chessboard pattern.

The next challenge is to do the same thing without the chessboard, but it’s pretty hard. Links for future reference:

The Othello Server

Mon, 01 Apr 2019 12:00:00 GMT

Located at https://othello.tjhsst.edu/

Code at https://github.com/hazelduvall/othello_tourney

I’d thought I’d finally write about my biggest creation so far. This baby has been developed through hundreds of hours across 3 years now. Aside from a few awesome webdev components provided by William O’Connell, I wrote all the code from scratch.

Background

The reason for the Othello Server’s existence starts in TJHSST’s Artificial Intelligence class. Every year, students in the class are assigned to write an AI program that plays the board game Othello. When I took the class, no one actually knew how to test their AI. All we were given was a simple shell to play the AI ourselves through a console window. Sharing your code with others was difficult, involving flash drives and lost programming time.

So, when I had finished my own AI early, my teacher and I came up with a plan. As an extra-credit project, I would create a simple web platform for students to upload their code to, allowing them to quickly test their code against their classmates’.

Apparently, I underestimated how much work it would take.

The Design

Evolution of the Othello Server frontend:

Application that generated series of PNGs with PyGame
Javascript canvas, SocketIO, everything is rectangles
Javascript canvas, Websockets, and animated pieces

Evolution of the Othello Server backend:

Running it on the same machine
Running it remotely in a jailed process, Flask webserver
2-stage backend, one to coordinate the web clients (Django) and one to handle running the AIs (custom asyncio).

I am only going to explain in detail the current state because oh boy has everything changed a lot.

Frontend

The frontend uses websocket callbacks to trigger updates to a <canvas> element that displays the board. A few other HTML elements display the player names, score, and any error messages thrown by AI.

In the end, it looks pretty OK if I do say so myself.

All the board and stone images were provided by William O'Connell

Django

Django is a very nice web framework. I feel, though, it was meant for projects a bit bigger than mine. Pretty much the only reason I use it is for easy OAuth support and the best Websocket support. Doing either of those things in Flask was a lot harder and more bug-prone.

That’s not to say Django is without its bugs, either. Especially with Channels/ASGI, a lot of things that I thought should be obvious and “just work” didn’t, in fact, work. That was one of the major reasons I split up the backend into two parts. As it turns out, handling incoming clients, safely parsing what games they want to run or what games they want to watch, and actually running the games is best left split up.

Now, for how I actually structured my Django code. Django itself forces a pretty tight scheme already, but I did take some liberties, especially with the game running code placement:

manage.py
run_ai_jailed.py (for running the jailed AIs)
othello/
- settings.py
- ion_secret.py (contains secret keys, in .gitignore)
- urls.py
- routing.py (ASGI-specific, similar to above)
- asgi.py and wsgi.py
- static/
- templates/
- apps/
  - auth/ (handles all the user authentication views, integration with the OAuth library)
  - users/ (defines the User model. Could be reasonably combined with above)
  - games/ (where most of the code lives. Handles interface with clients, game server backend)
- gamescheduler/ (where the entire game server code lives)
students/ (where all the uploaded student code lives)
run_gamescheduler_server.py (script to start the regular game server)
run_tournament_gamescheduler_server.py (runs the game server in tournament mode, rejecting any external requests to play)
csvity_aztr.py (turns results blob from tournament into a simple csv file)

Asyncio

Asyncio is the culmination of the Python development team’s twisted aims to bring about insanity in all programmers who attempt to use it. Something akin to viewing an eldritch abomination is required to fully understand the internal machinations of Futures, Tasks, coroutines, and awaitables. Once the knowledge invades, your mind will never be the same.

Jokes aside, programming something asyncio is seriously tough. Somehow, though, I pulled through in a weekend-long coding trance to get a fully asynchronous, multithreaded, custom protocol server working. The server using a monolithic protocol to keep track of all client states, something channels could never do (it spawns a new Consumer every time a client connects). Quick example below:

if __name__ == "__main__":
    loop = asyncio.get_event_loop()
    gs = GameScheduler(loop)
    def game_scheduler_factory(): return gs
    coro = loop.create_server(
        game_scheduler_factory,
        host=SCHEDULER_HOST,
        port=SCHEDULER_PORT
    )
    server = loop.run_until_complete(coro)

    # Server requests until Ctrl+C is pressed
    log.info("Running server")
    try:
        loop.run_forever()
    except KeyboardInterrupt:
        pass

    # Close the server
    log.info("Stopping server")
    server.close()
    loop.run_until_complete(server.wait_closed())
    loop.close()

Ingoring all the stuff about setting up the event loop, let’s focus on how I instantiate the protocol on line 5. Asyncio expects the factory function to either be the class itself (which can be called like a function to create new members of the class) or a function that returns new members of the class with custom parameters. I am lazy and make a function that returns the same object over and over again, so only it is listening for all incoming connections.

This is so I can handle all the room processing from within one object, not having to rely on Django’s lock-prone ORM. I did encounter a few a lot way too many errors trying to get asyncio to work properly with handling many rooms at a time and their communication pipelines and all, but in the end it chugs along smoothly.

Firejail

Now for the part I had the least trouble with (surprisingly): sandboxing! Early on, I did my research and came to the conclusion that https://firejail.wordpress.com/ was probably the best solution to my problem of how to secure student AIs. It could be run from the command line, easily installed on Linux, and had robust whitelisting and capability limiting support. Here’s the file I use to sandbox AIs:

caps.drop all
net none
noroot
seccomp
whitelist /home/othello/www/venv
whitelist /home/othello/www/public
whitelist /home/othello/www/run_ai_jailed.py
whitelist /home/othello/www/run_ai.py
whitelist /home/othello/www/Othello_Core.py
whitelist /home/othello/www/othello_admin.py

Simple easy, done, all that’s left is to surpress firejail’s output with the -q flag so it doesn’t mess up the other IPC I do.

Conclusion

I don’t really know where I was going with this. I have a lot of other in-code comments about the Othello Server that I will probably add to this post in the future.

If I had to choose one big takeaway, it’s this: one person can’t write an enterprise-grade app by themselves. They can try, and partially succeed, but it’ll soon become extremely hard to maintain without the help of others.

I’m hopeful that another Sysadmins at TJHSST will take up my mantle as the lead Othello Server developer after I’m gone. Anyways, thanks for reading if you made it this far.

TL;DR: Othello Server wack cool

Hello, World!

Tue, 12 Mar 2019 12:00:00 GMT

Might as well start a blog, I guess

This is running on an old Sony Vaio Laptop, Ubuntu 18.04 LTS. Nothing special, but it gets the job done.

Hazel Duvall's Blog

This Is Why My Latest iOS CI Check Failed

Click-To-Open In Neovim With Kitty

The Basics

The Shell

The Terminal

The Complications

Notes On Getting $KITTY_TAB_ID

Conclusion

Footnotes

It's Very Hard To Get LLMs To _Not_ Respond To Messages

Things That Don’t Work

Hypothesis For Why

The Thing That Does Work

Footnotes

CSS Flex, Grid, and Containing Blocks

The Bug

Time To Do Some Reading

Synthesis

Solving This The Correct Way

Solving This The Cool Way

Footnotes

The AIs Seem To Like One Of My Posts; Why Is That?

Hypotheses For Why Algorithms Like It

Why I Like It

Footnotes

node-webrtc v0.9.0 Released!

No More Compiler Warnings

Unified Build Environments With Nix

Reducing Memory Errors With Rust-Inspired Lifetime Management

Future Plans?

Footnotes

More Photography Updates (and RSS)

NYC Photography

All Javascript Keyboard Shortcut Libraries Are Broken

What Is A Keyboard Shortcut?

More Than One Way To Describe A Key

Oh No All The Libraries Use code

But Wait There’s More

Even More Actually

What The Good Libraries Have To Do

What We Cross-Browser Plebians Have To Do

Footnotes

Rochester Photography

Montreal Photography

Tracking Down a "Heisenbug" in Chrome

The Entire <body> Element Was Shifted Up

Shifting 2 - Electric Boogaloo

Debugging Time ^w^

A CSS Footgun?

A CSS Bug

Resolution

Footnotes

I Made A Webring

Better Static File Handling

A Safari aspect-ratio CSS Bug

Problem Statement

Solution 1: Simple Boundaries

Safari’s Problem

Solution 2: Good ol’ margin: auto

Footnotes

A Third Browser Bug...

Virtual StyleSheets

Virtual Documents

innerHTML

Footnotes

More Photography!

You Won't Believe That I Found Another Chrome Rendering Bug

Narrowing Down A Culprit

Why Is That??

HDR Was A Known Problem

HDR Still Is A Problem

Take 1

Take 2

Conclusion

RSS Feed Is Live

Working Around A Chrome Bug (And Other CSS Limitations)

Footnotes

My Blog Has More CI Now

I Think I Found A Chrome Bug

Notes On Getting `$KITTY_TAB_ID`

Oh No All The Libraries Use `code`

The Entire `<body>` Element Was Shifted Up

Solution 2: Good ol’ `margin: auto`

`innerHTML`