Doc links should count as a "use" with respect to the unused_imports lint

[sfackler]

use std::collections::HashMap;

/// [`HashMap`]
pub fn foo() {}

Compiling this will warn about the HashMap import being unused, but it is required for the doc link to work. It would be nice if rustc handled this like Java's compiler does, where a doc link acts as a use of an import even when not building documentation. You can work around this with [HashMap](std::collections::HashMap) but long imports in the doc block can get a bit gross.

[jyn514]

This is tricky because the unused_import lint runs outside of rustdoc (and in general, the interaction between rustdoc and resolve is kind of a mess). I don't know if this is possible without first fixing #68427 and #72381.

[Mark-Simulacrum]

I think this is likely unfixable in the general case without pulling all of the doc-comment parsing and evaluation machinery into rustc and running it.

That said I think you can work around this with e.g. cfg(doc) or whatever the cfg is?

[sfackler]

Yeah, it would require inlining the doc-comment parsing, but that'd be nice as well to get the broken_intra_doc_links lint into normal builds rather than just doc builds. Definitely appreciate that there's a ton of work in the way of doing that though!

At one point IIRC there was an idea of having rustdoc just talk to rust-analyzer which seems like it could potentially unblock this as well?

[jyn514]

That said I think you can work around this with e.g. cfg(doc) or whatever the cfg is?

This breaks cross-crate re-exports: #75855.

[jyn514]

At one point IIRC there was an idea of having rustdoc just talk to rust-analyzer which seems like it could potentially unblock this as well?

I would be very nervous of regressions doing this, especially for associated items. IIRC rust-analyzer has some trouble with associated items still.

[jyn514]

that'd be nice as well to get the broken_intra_doc_links lint into normal builds rather than just doc builds

I would want to improve the performance a lot more before doing this by default. #78761

[Lonami]

Not a solution, but note you can work around:

You can work around this with [HashMap](std::collections::HashMap)

With the following instead:

/// [`HashMap`]
///
/// [`HashMap`]: std::collections::HashMap
pub fn foo() {}

Which is cleaner (rather than use you type out the link name again, but it's about the same amount of typing), and doesn't clutter the actual documentation that much.

[petrochenkov]

pulling all of the doc-comment parsing and evaluation machinery into rustc and running it

Please, no. Rustdoc's link resolution is an ad-hoc mess very remotely related to path resolution rules in the language.
I don't want to pull it into the compiler and provide the compiler's stability guarantees for it.

[jyn514]

I agree with that actually - there are still some edge cases I'm hoping to change around primitives (on the theory that no one is using prim@ outside the standard library - maybe that should have been feature gated?) and it's easier to make changes like that in rustdoc than in rustc, since it will never break a working build.
#77875 (comment)