Process 1.6.13 changed observable behaviour

[angerman]

Reproducer is https://github.com/angerman/repro-process-behaviour-change-1.6.13

Will produce the following output:

$ make all CABAL=$(which cabal) GHC=$(which ghc)
Building against process < 1.6.13 ...
Running against process < 1.6.13; with an empty PATH, this should result in a git not found message
<no git executable found>
Building against process >= 1.6.13 ...
Running against process >= 1.6.13; this now throws an exception instead.
process-crash: git: readCreateProcessWithExitCode: posix_spawnp: failed (Undefined error: 0)
make: *** [Makefile:9: all] Error 1

We are now throwing an exception where we didn't before.

[bgamari]

Thanks @angerman, I'll investigate!

[bgamari]

Which operating system is this reproducible on?

[bgamari]

I cannot reproduce this on Linux.

Speaking to @angerman, it seems that they only see this on Darwin.

[bgamari]

@angerman says that posix_spawnp returns 2. However, the log output clearly suggests that errno==0. The problem appears to be that Darwin's posix_spawnp implementation does not set errno whereas glibc's appears to. This is the source of the discrepancy.