rustdoc: make --emit and --out-dir mimic rustc

[notriddle]

The behavior in the test case matches rustc's:

test-dingus % ls
main.rs
test-dingus % mkdir foobar
test-dingus % rustc --emit=dep-info main.rs --out-dir=foobar
test-dingus % ls
foobar  main.rs
test-dingus % ls foobar
main.d
test-dingus % rustc --emit=dep-info=testfile.d main.rs --out-dir=foobar
test-dingus % ls
foobar          main.rs         testfile.d
test-dingus % ls foobar
main.d

CC #146220 (comment)

[rustbot]

r? @camelid

rustbot has assigned @camelid.
They will have a look at your PR within the next two weeks and either review your PR or reassign to another reviewer.

Use r? to explicitly pick a reviewer

Why was this reviewer chosen?

The reviewer was selected based on:

• Owners of files modified in this PR: rustdoc
• rustdoc expanded to 9 candidates
• Random selection from GuillaumeGomez, camelid, fmease, lolbinarycat

[notriddle]

r? rustdoc

[fmease]

Thanks! I have some questions about some behavioral differences from rustc but once they're cleared up this PR is good to go.

View changes since this review

[fmease]

Hmm, rustc & rustdoc (except the json backend) actually interpret the - passed to --out-dir literally. So under "normal execution" both place the artifacts into a file called - (I can't say I like that; I think ideally we would reject such a 'weird' path). Consequently, rustc tries to write to -/$FILESTEM.d given --emit=depinfo --out-dir=-.

I fear it would be more consistent to ignore output_to_stdout here (which is currently only used by the json backend). What do you think?

[notriddle]

The json backend is usable with --emit=depinfo, though.

The problem is that --out is aliased to --out-dir in rustdoc, and it isn't in rustc. If #149365 was merged, there'd be a more principled way to handle this, but it isn't.

[fmease]

For some unknown reason, rustc doesn't actually try to create the outdir for depinfo files, not sure if that's intentional. E.g., --out-dir=nonexistent --emit=dep-info fails. This might warrant a quick investigation. Unless it's a bug in rustc, we should probably be consistent with rustc and not attempt to create the outdir? I'm split.

I don't know why but write_dep_info specifically contains a check to prevent create_dir_all from being called if DepInfo is the only requested emission type. However that runs after write_out_deps which by that point would've already bailed out with an error if the directory didn't exist...

rust/compiler/rustc_interface/src/passes.rs

Lines 844 to 855 in 67aec36

	write_out_deps(tcx, outputs, &output_paths);
	let only_dep_info = sess.opts.output_types.contains_key(&OutputType::DepInfo)
	&& sess.opts.output_types.len() == 1;
	if !only_dep_info {
	if let Some(ref dir) = sess.io.output_dir {
	if fs::create_dir_all(dir).is_err() {
	sess.dcx().emit_fatal(errors::OutDirError);
	}
	}
	}

[notriddle]

I can remove it.

I put it in there because it seemed more consistent with rustdoc's behavior (we implicitly create the directory, but we do it after emitting depinfo while emitting html), but it's easy for build systems to create the directory themselves (cargo does).