Add optimize(size) to some particularly large functions

[clubby789]

When built as part of libstd, this saves a few kb in hello world binaries.

The comments are formatted like they are so that if this feature is stabilized they can be identified and used unconditionally.

[clubby789]

Hm. Is there a way to check that we're actually built as part of libstd, without enabling for as-if-std 🤔

[workingjubilee]

I mean, I think that's kind of the point, no?

[workingjubilee]

It does seem more than slightly annoying though that we aren't really testing building this in std under actually-as-if-std conditions (e.g. nightly features), to be clear.

[clubby789]

Changed the as-if-std jobs to only run when using a nightly toolchain. The binary size job failure seems unrelated

[workingjubilee]

It is indeed. I am of half a mind to give that CI job the knife, but it certainly won't block this. It is supposed to help answering questions like "is this exact kind of PR an improvement?" but I can manually validate things.

Small questions first before I go looking: Did you evaluate these functions for size based on their source size or the actual compiled code size? Did the reduction happen only in those functions or was the entire compiled libstd smaller? I'm mostly wondering to make sure we didn't just hit some weird inlining thresholds or anti-thresholds.

[workingjubilee]

I'm mostly wondering to make sure we didn't just hit some weird inlining thresholds or anti-thresholds.

I think those "anti-thresholds" are called "limits", Jubilee.

[clubby789]

I identified these functions with cargo bloat iirc, then recompiled stage1 rustc with these changes, then a hello world binary with that compiler, and compared the size of those binaries

[workingjubilee]

Neat!

[workingjubilee]

Reproduces!

[clubby789]

rust-lang/rust#140735 (comment) (includes backtrace update with these changes)
Some small but nice binary size wins 🙂