Implement --check-cfg option (RFC 3013)

[mwkmwkmwk]

This is my attempt at implementing rust-lang/rfcs#3013. I'm not quite sure I'm doing this right, so I'd appreciate some feedback.

One major problem I hit in implementation is finding a good node_id to attach the early lint to: neither the config condition itself nor (for #[cfg]) the item that it guards is suitable, because they won't survive the expansion. For now I've attached the lint to crate root, but this is clearly suboptimal — I think the best we can do is using the parent item's node_id, but this seems to require a lot of plumbing to get the node_id down to the place where the lint is emitted. Is there some better way to handle this?

Also, I haven't implemented the -Z check-cfg option specified in the RFC — it seems -Z unstable-options already works as a feature gate well enough, so there's not much point?

Other TODOs:

• add plumbing to pass --check-cfg from rustdoc (do we want this one?)
• document --check-cfg (or is this something to be done on stabilization?)
• not in the RFC, but sounds like a good plan: also add a builtin list of well-known values (for stuff like target_os, ...) — perhaps we can even enable the lint by default for those?

cc #82450

[rust-highfive]

Thanks for the pull request, and welcome! The Rust team is excited to review your changes, and you should hear from @petrochenkov (or someone else) soon.

Please see the contribution instructions for more information.

[jyn514]

add plumbing to pass --check-cfg from rustdoc (do we want this one?)

It would be useful, I think, it catches issues like cfg(doctst) or something (and in general I would like expansion to match rustc as closely as possible).

The code lives around

rust/src/librustdoc/core.rs

Line 222 in 6f608ce

// these are definitely not part of rustdoc, but we want to warn on them anyway.

; you may also have to add an unstable option to `config.rs, I don't remember if rustdoc inherits rustc's options automatically or not.

[Aaron1011]

One major problem I hit in implementation is finding a good node_id to attach the early lint to: neither the config condition itself nor (for #[cfg]) the item that it guards is suitable, because they won't survive the expansion. For now I've attached the lint to crate root, but this is clearly suboptimal — I think the best we can do is using the parent item's node_id, but this seems to require a lot of plumbing to get the node_id down to the place where the lint is emitted. Is there some better way to handle this?

You can use ExtCtxt.current_expansion.lint_node_id, which is specifically designed for this kind of early linting. To make it available where you need it, you could add a lint_node_id field to StripUnconfigured, and use it in the corresponding call to cfg_matches. For the other calls to cfg_matches, e.g:

rust/compiler/rustc_codegen_ssa/src/back/link.rs

Line 2445 in 50f9f78

Some(ref cfg) => rustc_attr::cfg_matches(cfg, &sess.parse_sess, None),

you could either avoid linting, or continue to use CRATE_NODE_ID. Feel free to message me on Zulip if you have any follow-up questions.

[petrochenkov]

it seems -Z unstable-options already works as a feature gate well enough, so there's not much point?

+1

[mwkmwkmwk]

add plumbing to pass --check-cfg from rustdoc (do we want this one?)

It would be useful, I think, it catches issues like cfg(doctst) or something (and in general I would like expansion to match rustc as closely as possible).

This option doesn't affect expansion at all, though — the only effect is that lints are emitted. I also don't see why we need specifically rustdoc to catch cfg(doctst), as this should already be caught by the main rustc run? After all, doctest is already included in the main "well-known conditions" list.

The code lives around

rust/src/librustdoc/core.rs

Line 222 in 6f608ce

// these are definitely not part of rustdoc, but we want to warn on them anyway.

; you may also have to add an unstable option to `config.rs, I don't remember if rustdoc inherits rustc's options automatically or not.

Wait, this seems like rustdoc specifically disables all lints except for this short list? Are the "unknown condition" lints somehow special enough to warrant adding here?

[petrochenkov]

Could you add one more non-error test as well - making sure that cfg(unknown_key = "value") produces a warning when names checking is enabled.

[sivadeilra]

I actually implemented this, last year, when I wrote the RFC. Let me dust off this old branch and see if there's anything worth rescuing from it...

[sivadeilra]

Btw, you're welcome to grab anything useful from my old branch: https://github.com/sivadeilra/rust/tree/user/ardavis/check_cfg If you do, just credit me.

I had it mostly working, but had to set it down for some other work. By the time I returned to this work, master had moved on so much that I couldn't just trivially rebase.

[bors]

☔ The latest upstream changes (presumably #89629) made this pull request unmergeable. Please resolve the merge conflicts.

[JohnCSimon]

ping from triage:
@mwkmwkmwk
Please address the merge conflict and adjust the label with @rustbot ready when you're ready for another review. Thank you.

[JohnCSimon]

Ping from triage:
@mwkmwkmwk
I'm closing this due to inactivity, Please reopen when you are ready to continue with this. Thank you.

@rustbot label: +S-inactive