Revert "Merge pull request #165 from sched-ext/reduce-rust-build-load"

[htejun]

This reverts commit a7b39f2, reversing changes made to cf7404f.

The PR doesn't do what the description says. It instead limits the number of rustc instances to 1 for each cargo build making rust builds extremely slow. Let's revert and try again.

[arighi]

Hm... yeah that's definitely not what I wanted to do, will do more tests, sorry for the noise.

[htejun]

No worries. Hopefully, there's a way to do this with meson.

[arighi]

No worries. Hopefully, there's a way to do this with meson.

Asked the meson guys and apparently there's not a way (yet), unless we pass --jobs=1 to the meson compile command line.

meson 1.4.0-rc1 will have builtin Rust support, basically it can run rustc directly without cargo and it'll be able to account the proper amount of jobs, handle dependencies, etc. but relying on that doesn't seem very portable (not all distro will immediately move to the new meson).

Probably for now the best way is to use --jobs=1 directly in our CI scripts (that seems to properly parallelize rustc across all the available CPUs, serializing the cargo build instances) and maybe also add --jobs=1 to the README, so people may just copy/paste that and build all the Rust stuff without overloading their system...

[htejun]

We're also having trouble integrating libbpf build. I'm half-way convinced that meson isn't a great option for this repo and might switch to something which is more expressive and less restrictive. If you have some ideas on which build system would be better, please share.

[htejun]

So, possibly a better way of serializing rust builds is introducing serial dependencies across all of them so that e.g. scx_layered depends on scx_rustland which depends on scx_rusty and so on, so that the build system is forced to serialize them.

[arighi]

I wish there was a better way than adding direct dependencies like that, but if that works I guess it can be a reasonable workaround.

If there was a way to set --njobs=1 by default, unless specified otherwise, everything would work perfectly. I can't believe meson doesn't have something like that.

Anyway, I'll investigate a bit.

[htejun]

--njbos=1 is okay but not perfect in that it'd serialize compilation of all c scheds. They tend to be pretty small and we don't have many of them, so that's mostly okay, I guess.