Cadey is coffee
<Cadey> Hello! Thank you for visiting my website. You seem to be using an ad-blocker. I understand why you do this, but I'd really appreciate if it you would turn it off for my website. These ads help pay for running the website and are done by Ethical Ads. I do not receive detailed analytics on the ads and from what I understand neither does Ethical Ads. If you don't want to disable your ad blocker, please consider donating on Patreon or sending some extra cash to xeiaso.eth or 0xeA223Ca8968Ca59e0Bc79Ba331c2F6f636A3fB82. It helps fund the website's hosting bills and pay for the expensive technical editor that I use for my longer articles. Thanks and be well!

The carcinization of Go programs

Read time in minutes: 11

hero image crab-invasion

Image generated by Waifu Diffusion v1.3 (float16) -- crabs, invasion, beach, palm trees, green hill zone, studio ghibli, xenoblade chronicles 2, pokemon, ken sugimori, thick outlines, ink

Sometimes you just need to embed this one library written in another language into your program. This is a common thread amongst programmers time immemorial. This has always been a process fraught with peril, fear, torment, and lemon-scented moist towelettes for some reason.

Normally if you want to call a Rust function from Go, you have to go through some middleman like cgo. This works and is somewhat elegant for how utterly terrible a hack cgo is.

However, the main problem is that when you use cgo to link a Rust function into a Go program, you need to copy around the shared object that Rust generates. You can't check this shared object into your source tree (it needs to be unique per OS distribution per OS per CPU architecture, much like normal dynamically linked binaries are). It does work, but overall the developer experience is poor. Your build is no longer one simple go build. Now you have to remember to run cargo build --release and ensure that the resulting .so, .dll, or .dylib is in the right path for the OS' dynamic linker to read from. It's a mess.

Mara is hacker
<Mara> This is such a big problem that at a generic level that this is why Nix and NixOS exist. Imagine how complicated this is when you get general-purpose OS components into the mix. It's astounding that anything works at all.

So what if I told you there was a way that we could ship one binary from Rust, have that work on every platform Go supports, and not have to modify the build process beyond a simple go build? Imagine how much easier that would be. It's easy to imagine that such a thing would let users not even know that Rust was involved at all, even if they consume a package or program that uses it.

I've done this with a package I call mastosan and here's why it exists as well as how I made it.

Why

Mastodon stores toots in HTML and presents that HTML to API consumers. HTML is very nice for a browser to display, but this is not as useful for a bot. Especially if your goal is to send toots to a Slack webhook.

When you look at a toot such as this in the API:

the profile picture for cadey
Xe :verified: @cadey
M11 22 2022 18:50 (UTC)

test mention @xe so I can see what HTML mastodon makes

Link

Its content looks something like this:


<p>test mention <span class="h-card"><a href="https://vt.social/@xe" class="u-url mention">@<span>xe</span></a></span> so I can see what HTML mastodon makes</p>

Ideally we'd like it to look semantically identical in Slack, maybe something like this:


test mention <https://vt.social/@xe|@xe> so I can see what HTML mastodon makes

This will display the link in Slack like any other hyperlink. As things get more elaborate, Mastodon will do more semantic weirdness like invisible spans and other things that make displaying things on Slack annoying. Imagine the difference between these two things:


https:// tailscale.com/blog/introducing -tailscale-funnel/

https://tailscale.com/blog/introducing-tailscale-funnel/

One is much more easy to understand for humans than the other.

How

One of the core features of the UNIX philosophy is the idea that programs are simple filters that do one thing well and then allow you to compose them into new and interesting ways. If you've ever used curl and jq together to do things like read data from a JSONFeed, you know how this is in practice:


$ curl https://xeiaso.net/blog.json -qsSL | jq .items[0].title -r
The birdsong persists

I made a little program in Rust that uses lol_html to take incoming Mastodon-flavored HTML and emit slack-flavored markdown. Usage is simple:


$ echo '<p>test mention <span class="h-card"><a href="https://vt.social/@xe" class="u-url mention">@<span>xe</span></a></span> so I can see what HTML mastodon makes</p>' | ./testdata/mastosan.wasm
test mention <https://vt.social/@xe|@xe> so I can see what HTML mastodon makes

That's it. It takes input on standard input and returns the result on standard output. This doesn't cleanly map to the WebAssembly flow, except if you use WASI to bridge the gap. WASI gives WebAssembly programs enough of a POSIX-like environment that most basic things can work, but here we are only really using two major parts of it: standard input and standard output.

In Go, if you were running this as a normal OS subprocess, you'd probably write some code like this:


package foo

import (
    "bytes"
    "os/exec"
    "strings"
)

func HTML2Slackdown(input string) (string, error) {
    loc, err := exec.LookPath("mastosan")
    if err != nil {
        return "", err
    }
    
    fout := &bytes.Buffer{}
    cmd := exec.Command(loc)
    cmd.Stdin = bytes.NewBufferString(input)
    cmd.Stdout = fout
    if err := cmd.Run(); err != nil {
        return "", err
    }
    
    return strings.TrimSpace(fout.String()), nil
}

However this still depends on the program being compiled for your native OS and distribution as well as present in a folder in your $PATH. This works, but this is not ideal in the slightest.

Rust lets you build a binary that targets WASI with this compiler flag:


cargo build --target wasm32-wasi --release --bin mastosan

This will emit a several megabyte binary file in ./target/wasm32-wasi/release/mastosan.wasm. When you run it, it will do what you want.

Now you need to use it from Go. There's many choices for this, but I chose to use wazero. The overall flow of using this is similar to using a subprocess with os/exec, but slightly different because we're embedding WebAssembly. It will look like this:


//go:embed testdata/mastosan.wasm
var mastosanWasm []byte

func HTML2Slackdown(ctx context.Context, text string) (string, error) {
    // create wazero runtime
	r := wazero.NewRuntime(ctx)
	defer r.Close(ctx)
    
    // load wasi environment into runtime
	wasi_snapshot_preview1.MustInstantiate(ctx, r)

    // set up standard output and standard input
	fout := &bytes.Buffer{}
	fin := bytes.NewBufferString(text)

    // create runtime configuration
	config := wazero.NewModuleConfig().WithStdout(fout).WithStdin(fin).WithArgs("mastosan")

    // compile the WASM module
	code, err := r.CompileModule(ctx, mastosanWasm)
	if err != nil {
		log.Panicln(err)
	}

    // run the WASM module
	if _, err = r.InstantiateModule(ctx, code, config); err != nil {
		return "", err
	}

	return strings.TrimSpace(fout.String()), nil
}

This is mostly the same thing. You set up the environment, load the WASM module and then run it. The main difference is that instead of loading the binary as machine code from the disk, I use go:embed to embed the precompiled WebAssembly module into the binary. This means that the resulting Go program will Just Work as long as the WebAssembly module is present in the place it expects.

Moar faster

One main drawback to this implementation is that it's a bit slow. It has to compile the WebAssembly module every time the function is called.

The wazero runtime and compiled WebAssembly module code can be lifted into a package-level variable, like with this patch. The main advantage this gives you is speed. After this patch, the WebAssembly module is only ever compiled once, on application boot. Before this patch, each invocation took about 0.2 seconds per run. Here's the benchmark results after this patch:


BenchmarkHTML2Slackdown             1221            938774 ns/op
BenchmarkHTML2Slackdown-2           2293            488032 ns/op
BenchmarkHTML2Slackdown-6           3555            305505 ns/op
BenchmarkHTML2Slackdown-12          3897            297974 ns/op

It's gone down from 0.2 seconds in the best case to 0.3 milliseconds in the best case. This is at least a 1000x increase in performance, with most of the time probably being spent in the HTML parser rather than being spent in anything else.

I think this is going to more than meet my needs both personally and at work. I'm going to have to try this a bit more against random Mastodon messages to see if it does what I want. It's cool to be able to merge two incompatible worlds together and I'm excited to see what I can do in the future with this.

Cadey is coffee
<Cadey> Darn you shitposting coworkers nerd sniping the poor DevRel on their well-earned week off!

This article was posted on M11 22 2022. Facts and circumstances may have changed since publication Please contact me before jumping to conclusions if something seems wrong or unclear.

Tags: cursed wasm go rust

The art for Mara was drawn by Selicre.

The art for Cadey was drawn by ArtZorea Studios.