[qb42]

View all comments

This is a rough draft inspired by @ChrisDenton in #150501 (comment). I won't have much time to work on it until next month but I wanted to get it out there.

This is implemented as an mdbook preprocessor that pulls data from the rustc_target crate, so it will always be accurate.

Assuming this works as intended (I am not set up to test locally), it's still missing the following:

• Links to target detail pages
• Footnotes

[Noratrieb]

This is very cool! I started a similar effort a while ago in #120745 but never got to completing it. Maybe some of the code in the linked PR thee may be useful

[qb42]

@Noratrieb Thanks, I'll take a look.

[qb42]

Alright, I'm set up properly now so no more failing tidy or pushing obvious mistakes.

Unfortunately it looks like depending on rustc_target directly is a dead-end, I'm getting a variety of errors related to it being a compiler-internal crate that is supposed to be compiled as part of bootstrap. The commit I just pushed (8c1125f) is my last attempt in that direction unless someone else has a solution.

I'm going to switch to pulling data from rustc -Z unstable-options --print all-target-specs-json and see how that goes.

[Noratrieb]

Executing rustc like that seems like the best approach, yes.

[Dylan-DPC]

@qb42 any updates on this? thanks

[qb42]

@Dylan-DPC IIRC, the last commit I pushed worked on my machine (though without links or footnotes). I put this to the side after failing to figure out what the issue with the filepath was and forgot about it. I'll revisit in the next couple of days and see if I can get it working now I'm not in school.

[qb42]

Hey @Noratrieb, I've been working on an updated version of your PR from 2024 and I was hoping you could take a look at it. It's still mostly yours, I just brought it up to date, cleaned it up a little, and added a few features (like optional sections). I believe the only blocker listed on your PR (not working on stage 0) is no longer relevant, since docs don't seem to be generated at all on stage 0 anymore.

The high-level of the important changes I made are:

• Generating the table headers as well as the contents, optionally replacing the whole thing with a message if the table would be empty (I added that bit before i686-pc-windows-msvc got demoted).
• Moving the file replacement logic in render_static to its own function, which also removes the marker comments when check_only=true. This allows the markers in SUMMARY.md to enclose the existing list, meaning that TARGET_CHECK_ONLY now fully controls whether the old or new docs are shown (instead of old or old and new smashed together).
• Create a page listing targets without a maintainer on record.
• Switch to using '-Zunstable-options --print all-target-specs-json' instead of querying each target individually, this sped up that stage by ~10 seconds on my machine. Collecting rustc information is still easily the slowest part, unfortunately.
• Added support for optional sections, which you had mentioned in a comment. I'm not sure which ones should be optional so there aren't any yet and the optional constructor is commented out.
• Allowed headers other than h2, for use as subsections. A quick grep showed some existing uses that seemed reasonable to me, but I can drop this if you prefer.

If you can remember any other features you were planning, I'll take a crack at implementing them. Otherwise, I'll get started on converting the existing docs to target_infos.

[rust-bors]

☔ The latest upstream changes (presumably #165002) made this pull request unmergeable. Please resolve the merge conflicts by rebasing.