rustdoc's treatment of cross-crate ("imported") files is suboptimal.

[eddyb]

@bjorn3 noted in #70025 (comment) that single files missing disables [src] links for the entire crate.
That's not ideal UX, we should render as many source files as we can possible load.

---

And there's also the paths rustdoc uses (see #70025 (comment)), e.g.:

• https://doc.rust-lang.org/nightly/nightly-rustc/rustc_span/def_id/struct.CrateId.html
  • links to https://doc.rust-lang.org/nightly/nightly-rustc/src/rustc_span/checkout/src/librustc_index/vec.rs.html#110-112
  • but it should use https://doc.rust-lang.org/nightly/nightly-rustc/src/rustc_index/vec.rs.html#110-112

If we use the CrateNum of the original source file and reuse the existing rustdoc logic to get the relative path of that file in that crate, I believe we can even support linking to the source correctly even when the absolute paths are not usable (e.g. /rustc/$hash/src/libstd/thread/local.rs when using thread_local! and the rust-src rustup component isn't installed)

Because the downstream crate uses an exported macro from the original crate, the rendered source file should always have been generated, so we should never really need to re-generate it.

cc @rust-lang/rustdoc @Aaron1011

[bjorn3]

Because the downstream crate uses an exported macro from the original crate, the rendered source file should always have been generated, so we should never really need to re-generate it.

It is possible to document a single crate.

[eddyb]

@bjorn3 I think then they get generated at the same paths but by the current rustdoc (this logic already exists, it's just only used for the "cross-crate doc inlining mode", not "cross-crate source files").

[ollie27]

#70642 may have fixed the compiler docs but it didn't fix the underlying regression caused by #66364. rustdoc is now trying to load files belonging to external crates which even without --remap-path-prefix, it won't always have access to those files.

If we use the CrateNum of the original source file and reuse the existing rustdoc logic to get the relative path of that file in that crate

Yeah, that look like it's the easiest way to actually fix the regression.

I've submitted #70828 to fix this.

[jyn514]

For future readers: this is mostly fixed but was left open to track the edge case in #70828 (comment).