emit error when doc generation fails

[euclio]

Fixes #41813.

The diagnostic looks something like this:

error: couldn't generate documentation: No space left on device (os error 28)
  |
  = note: failed to create or modify "/path/to/crate/target/doc/src/lazycell"

[rust-highfive]

r? @steveklabnik

(rust_highfive has picked a reviewer for you, use r? to override)

[TimNN]

Ping form triage @steveklabnik / @rust-lang/rustdoc: This PR requires your review.

[steveklabnik]

@QuietMisdreavus or @GuillaumeGomez should be r+'ing this, not me

[GuillaumeGomez]

Looks nice! Please add a rustdoc-ui test and then I r+.

[euclio]

I'd like to, but I'm not sure how to simulate filesystem errors in a UI test.

[GuillaumeGomez]

Ah that's a good point...

@rust-lang/infra Does anyone knows how to simulate a full disk?

[kennytm]

what about write to a location which you have no permission? this will need to be changed to a run-make test though.

[GuillaumeGomez]

Hard to write an all-OSes test...

[QuietMisdreavus]

It may be enough to make it run on just Unix, though. The trick is just to make sure it outputs the right kind of error message when it encounters a filesystem error, even if the nature of that error doesn't matter much. Plus, plenty of other run-make-fulldeps tests skip running on Windows, so at least there's precedent there. At its core, we basically need to make sure that it doesn't say "internal compiler error" (or whatever the right capitalization should be) when it hits this error, right? If you want, i could try to sketch it out.

Otherwise, i quite like this PR. (I'm also not opposed to landing it as-is, since it's effectively just changing the way rustdoc prints certain kinds of errors.)

[euclio]

@QuietMisdreavus pushed a run-make test that checks that it doesn't ICE.

[QuietMisdreavus]

Does this save the exit status into $(.SHELLSTATUS) for later? I'm not familiar with this syntax.

[euclio]

Yes. https://www.gnu.org/software/make/manual/html_node/Shell-Function.html#Shell-Function

[QuietMisdreavus]

Cool, thanks for the link!

[QuietMisdreavus]

Excellent, thanks so much! r=me when travis is green.

[GuillaumeGomez]

@bors: r=QuietMisdreavus

[bors]

📌 Commit e2fa3c17ebf51aac41801a040ea0e7857e9da63f has been approved by QuietMisdreavus

[kennytm]

@bors r-

The test did not pass on Windows. See #56531 (comment)

---- [run-make] run-make-fulldeps\rustdoc-io-error stdout ----
error: make failed
status: exit code: 2
command: "make"
stdout:
------------------------------------------
make[1]: Entering directory '/c/projects/rust/src/test/run-make-fulldeps/rustdoc-io-error'
make[1]: Leaving directory '/c/projects/rust/src/test/run-make-fulldeps/rustdoc-io-error'
------------------------------------------
stderr:
------------------------------------------
make[1]: *** No targets.  Stop.
------------------------------------------
thread '[run-make] run-make-fulldeps\rustdoc-io-error' panicked at 'explicit panic', src\tools\compiletest\src\runtest.rs:3284:9
note: Run with `RUST_BACKTRACE=1` for a backtrace.
failures:
    [run-make] run-make-fulldeps\rustdoc-io-error
test result: FAILED. 192 passed; 1 failed; 0 ignored; 0 measured; 0 filtered out
thread 'main' panicked at 'Some tests failed', src\tools\compiletest\src\main.rs:503:22

[euclio]

The test was missing an else branch that runs an empty target. Fixed.

[QuietMisdreavus]

r=me pending travis

[QuietMisdreavus]

@bors r+

[bors]

📌 Commit c359f98 has been approved by QuietMisdreavus

[bors]

⌛ Testing commit c359f98 with merge 14997d5...

[bors]

☀️ Test successful - status-appveyor, status-travis
Approved by: QuietMisdreavus
Pushing 14997d5 to master...

[michaelwoerister]

I'm wondering why this caused a visible performance regression:
https://perf.rust-lang.org/compare.html?start=b866f7d258a2428e004039eb0f3fabd73cc9d6ae&end=14997d56a550f4aa99fe737593cd2758227afc56&stat=instructions:u

cc @rust-lang/wg-compiler-performance

[eddyb]

@michaelwoerister That makes no sense, I'd suspect an issue with the perf setup.