Post

Log in
About
Code of conduct
Privacy
Users
Instances
About Bonfire

@algernon@come-from.mad-scientist.club · activity timestamp 4 hr. ago

I wanted to remove minification from #iocaine 4, because it's expensive, achieves comparatively little, and the dependencies required for it are big too.

But I'm having second thoughts. Minifying every request after garbage generation is definitely going away, yes, but... what if I minified the template? That might be worth something, as the price is only paid twice: once in dependencies, and once on startup.

On the other hand, one can easily minify the template outside of iocaine, and then I can drop the dependencies.

#iocaine

Flag this post
Block

5 Replies

0 Boosts

Luddicus Mus

@algernon@come-from.mad-scientist.club · activity timestamp 4 hr. ago

I guess... one way to approach this would be to remove it in iocaine 4.0, and if there's a demand for it, and a good reason, it can always be re-introduced in a later 4.x version.

One of the advantages of a major version is that I can break things. I plan to use this advantage to remove dependencies. I can always add things back, if they're needed, that doesn't break SemVer.

Flag this comment
Block

Luddicus Mus

@algernon@come-from.mad-scientist.club · activity timestamp 4 hr. ago

Out of curiosity, lets do a "quick" comparison! Lets compare the size of target/release/iocaine after a cargo build -r:

iocaine 3.5.1 (unmodified): 29Mb
iocaine 3.5.1 (without minify-html, toml, and serde_yaml): 11Mb

It's also worth looking at the number of dependencies. In iocaine 3.5.1's case, there are 611 crates to compile, but with the above three removed, that's down to 466.

I also plan to remove JSON support, figment, exn, and turn clap_mangen into a dev dependency rather than a runtime one, and those would decrease the size of the dependency tree, and the final binary too.

But the impact of removing minify_html is by far the biggest win. Less than half the size, and a whole lot of dependencies gone.

Flag this comment
Block

Luddicus Mus

@algernon@come-from.mad-scientist.club · activity timestamp 3 hr. ago

I can't easily test the impact of removing serde_json and figment, because they're currently closely tied to various places in iocaine. Removing them requires changing how the entire handler configuration works.

Chances are, the savings aren't as dramatic, probably a few tens of kilobytes or so?

I'm also looking at removing some other dependencies, like axum, console-subscriber, paste, pkg-config (some of which are dev dependencies already).

The goal is to go below 10Mb for the release binary on x86-64. With the planned dependency drops, and the roto/lua runtimes becoming a whole lot simpler, and a lot of code being dropped, that feels achieavable.

Flag this comment
Block

Luddicus Mus

@algernon@come-from.mad-scientist.club · activity timestamp 3 hr. ago

Hm, looking at cargo bloat output, maybe below 10Mb is too ambitious.

There's 606KiB of iocaine-powder, and I will likely be able to chisel quite a bit off that (200KiB, maybe), axum at 105KiB, figment at 88KiB... that's still far from shaving more than a megabyte off.

Hm. There's zstd-sys at 115KiB, and miniz_oxide at 28KiB. They're used to compress the embedded stuff, but... that's ~143KiB to compress ~524KiB. That sounds worthwhile, but... disabling compression cuts dependencies further down from 466 to 420, and the binary size is maybe ~22KiB larger.

46 dependencies less is a good trade for +22KiB larger binary.

Flag this comment
Block

1+ more replies

Ariel (🐿 arc)

@arichtman@eigenmagic.net · activity timestamp 3 hr. ago

@algernon is the aim to make it better sized for router distros like openwrt and *sense?

Flag this comment
Block

1 more reply