Any way to make the compilation/testing faster?

[juliohq]

I'm sorry for opening this issue, I'm not sure if this is the correct way to discuss about this topic.

Right now the compilation process looks a bit slow. Do you think there's any way to speed it up somehow? Maybe avoiding useless recompilation?

[fruitsbat]

have you already tried using mold as a linker?

it's a pretty safe way to speed up rust compilation but still has some issues on anything that's not linux iirc

[Bromeon]

Yes, there are several ideas:

• Get rid of bindgen when possible -- Pre-generate bindgen file #12
• Reuse Godot artifacts -- Godot compatibility guidelines #107
• Split godot-core crate up (it's too big, and Rust cannot efficiently compile code inside a crate in parallel)

Also, once that's all done, we an have a look at the flamegraphs emitted by cargo build --timings to get a more concrete idea of bottlenecks.

Bevy also has some ideas on build optimization, some might be applicable here. In general, one huge advantage that gdext has over other Rust projects is that the legwork is done by an external engine, which is dynamically linked -- so a lot of the problems are moved to the C++ side.

[juliohq]

have you already tried using mold as a linker?

it's a pretty safe way to speed up rust compilation but still has some issues on anything that's not linux iirc

Not actually, but I'm trying it right now and looks to be way faster. And yes, I'm on Linux.

[Bromeon]

@juliohq out of curiosity, where exactly do you experience slowness? During incremental compilation, or just the initial one?

[juliohq]

@juliohq out of curiosity, where exactly do you experience slowness? During incremental compilation, or just the initial one?

During incremental ones and ./check.sh. I have an i3-2120, which is pretty much low-end, but I don't think that should be so slow.

Also, commenting out those lines makes it cargo build 20% faster for a clean build:

[profile.dev.package.bindgen]
debug = 0
opt-level = 3

[profile.dev.package.nanoserde]
debug = 0
opt-level = 3

[profile.dev.package.quote]
debug = 0
opt-level = 3

[profile.dev.package.proc-macro2]
debug = 0
opt-level = 3

[profile.dev.package.venial]
debug = 0
opt-level = 3

[profile.dev.package.godot-codegen]
debug = 0
opt-level = 3

[lilizoey]

Also, commenting out those lines makes it cargo build 20% faster for a clean build:

do you know what impact it has on incremental build times?

[juliohq]

Also, commenting out those lines makes it cargo build 20% faster for a clean build:

do you know what impact it has on incremental build times?

2m 26s with opt-level = 3, 2m 11s with everything commented out. However I wouldn't take that as absolute, still need to check what the times are under different conditions I guess.

[Bromeon]

Very interesting, thanks for the feedback. Might be worth removing the special handling of those crates. Seems like the faster runtime does not make up for the extra compilation time needed for release mode.

Regarding check.sh, one problem is that the different subtasks may be compiled with slightly different compiler config, which requires needless recompilation. We can probably improve that too.

[Bromeon]

I removed those lines and allow for pre-generated bindgen in #211. There are still more places to optimize, like check.sh, but it might already help somewhat.

[juliohq]

@Bromeon sure, that has helped a lot though!

[Bromeon]

Good to hear! 🙂
I'll close this, we can address future improvements in separate issues.