Spurious incremental mkdir error

[alexcrichton]

Seen on https://buildbot.rust-lang.org/builders/auto-linux-32-nopt-t/builds/10408/steps/test/logs/stdio

failures:

---- [codegen-units] codegen-units/partitioning/local-transitive-inlining.rs stdout ----

error: compilation failed!
status: exit code: 101
command: i686-unknown-linux-gnu/stage2/bin/rustc /buildslave/rust-buildbot/slave/auto-linux-32-nopt-t/build/src/test/codegen-units/partitioning/local-transitive-inlining.rs -L i686-unknown-linux-gnu/test/codegen-units/ --target=i686-unknown-linux-gnu -L i686-unknown-linux-gnu/test/codegen-units/partitioning/local-transitive-inlining.stage2-i686-unknown-linux-gnu.codegen-units.libaux -C prefer-dynamic -o i686-unknown-linux-gnu/test/codegen-units/partitioning/local-transitive-inlining.stage2-i686-unknown-linux-gnu --cfg rtopt -C rpath -L i686-unknown-linux-gnu/rt -Zprint-trans-items=lazy -Zincremental=tmp/partitioning-tests/local-transitive-inlining
stdout:
------------------------------------------

------------------------------------------
stderr:
------------------------------------------
error: Could not create incremental compilation crate directory `tmp/partitioning-tests/local-transitive-inlining/local_transitive_inlining-7tx6mqwyh4r5`: No such file or directory (os error 2)

error: aborting due to previous error


------------------------------------------

thread '[codegen-units] codegen-units/partitioning/local-transitive-inlining.rs' panicked at 'explicit panic', /buildslave/rust-buildbot/slave/auto-linux-32-nopt-t/build/src/tools/compiletest/src/runtest.rs:2359
note: Run with `RUST_BACKTRACE=1` for a backtrace.


failures:
    [codegen-units] codegen-units/partitioning/local-transitive-inlining.rs

Seems relatively serious?

cc @michaelwoerister
cc @nikomatsakis

[alexcrichton]

Ah and the spurious failure was #35074 (comment)

[michaelwoerister]

@alexcrichton That's strange, the directory is created using your own create_dir_racy function which has worked quite well for a few months now, right?

[alexcrichton]

@michaelwoerister yeah that function has worked well, but also assumes that no directories are being deleted. Perhaps there's a deletion racing with a creation here?

[michaelwoerister]

As far as I can tell, the directory the error is about should never be deleted, only sub-directories of it. Maybe compiletest does something it didn't do before?

[alexcrichton]

Hm yeah, maybe some directory is being deleted? Haven't seen it since yet unfortunately :(

[michaelwoerister]

So, we have a case on 32bit Linux and one on 64bit Linux, so far.

[alexcrichton]

Haven't seen this in awhile, and we're off buildbot, so let's see if it's gone.