run spellcheck as a tidy extra check in ci

[lolbinarycat]

This is probably how it should've been done from the start.

r? @Kobzol

[rustbot]

Some changes occurred in HTML/CSS/JS.

cc @GuillaumeGomez, @jsha, @lolbinarycat

[Kobzol]

I would suggest preinstalling the typos binary in the tidy Dockerfile (by downloading a pinned version from GH releases), and then also adding some download functionality from GitHub releases for local execution (or just use cargo install probably easier), same as how tidy installs Python/JS deps for the user.

[klensy]

Can you keep previous logic: check version first and only if not found/wrong build it? I have typos in PATH downloaded from github and not build via cargo, so current logic will build second one for me.

P.S. in case of building typos, it will be nice to drop note in log like building typos or smth.

[okaneco]

Suggested change

	/// If the given executible is installed with the given version, use that,
	/// If the given executable is installed with the given version, use that,

[lolbinarycat]

you would really think that enabling typo checking would catch that ^^;

[Kobzol]

I think that tidy is being executed from some other directory. You should specify the working directory when running the typos binary to be the source root which is passed to tidy.

[Kobzol]

The comment on top of SPELLCHECK_DIRS is outdated, could you remove it please?

The build of typos seems to take ~1 minute on CI. We could try to build it with --debug, or preinstall it in the Dockerfile, but in any case 1 minute isn't so terrible.

[Kobzol]

Let's move this to tidy. It's only used in one place now, no need to increase bootstrap's (and other tools') build time by putting it into build_helper.

[lolbinarycat]

The build of typos seems to take ~1 minute on CI. We could try to build it with --debug, or preinstall it in the Dockerfile, but in any case 1 minute isn't so terrible.

this seems like the exact usecase for opt-level 1 or 2, where you care about minimizing run + build time. i'll do local testing to see what works best.

UPDATE: local testing seems very noisy, opt level 0 was fastest, but then weirdly 3 was next fastest, with 1 and 2 being almost twice as slow as the others. this seems strange, as the non-rebuild runtime of tidy seemed about the same no matter the opt level. I guess we can just stick -Copt-level=0 in RUSTFLAGS, since it doesn't seem like typos relies much on optimizations being enabled. preinstalling in the Dockerfile also seems like a good idea, but I don't know enough about docker or github releases to do that properly (we do need to make sure the versions stay in sync, tho, probably with a .version file)