Experiment with the `mostly-unused` Cargo hint to optimize compile times for downstream users

[alice-i-cecile]

Suggestion

As explained in this call for testing, this hint can reduce the amount of time spent compiling and optimizing code that is thrown away.

If you maintain a crate that has a large API surface, and you expect that the typical user might use only a fraction of it, try setting hints.mostly-unused in your Cargo.toml

windows is a great candidate for this! This is especially true for downstream users (like Bevy) who depend on multiple copies of the windows crate transitively (see bevyengine/bevy#11244).

As the blog post says, if you do this, please report your findings in rust-lang/cargo#15644 <3

[ChrisDenton]

See #3660

[kennykerr]

This was added in #3660 - the benefits are not conclusive at this point but I'll leave it there for now to enable early adopters to test. Do let me know if you find some compelling benefits.

[tim-weis]

I'd love to supply numbers, but I don't fully understand the ins and outs of running before-and-after tests. Here's what I tried:

1. Updated all [dependencies] under the "windows-rs" umbrella to reference git = "https://github.com/microsoft/windows-rs"
2. Switched to the +nightly toolchain
3. Toggled between "before" and "after" via the -Zprofile-hint-mostly-unused command line option

Given the inconclusive results posted in #3660 (which align with mine), I'm beginning to wonder whether:

• the -Z switch actually toggles the behavior under scrutiny (rather than being enabled by default on +nightly), or
• hyperfine produces reliable results on Windows

The project I tested does ingest the windows crate, and it's moderately complex. Am I just not using it correctly, or does the mostly-unused hint simply not have a measurable impact on the windows crate for some reason?