Re: Multiple as-installed test suite failures under different circumstances with 5.1-test-1

[Peter Stephenson <p.stephenson@xxxxxxxxxxx>]

On Wed, 9 Sep 2015 16:10:27 +0200
Axel Beckert <abe@xxxxxxxxxxxxxxx> wrote:
> We though had a test suite failure at _build_ time (i.e. "make test"
> directly after the build) once and only once withing 5 build runs so
> far, so it may be a kind of race condition:
> 
>...
> Test ../../Test/E01options.ztst failed: error output differs from expected as shown above for:
>...
> Was testing: SH_NULLCMD option
>...
> ../../Test/E01options.ztst: test failed.
> ../../Test/E02xtrace.ztst: starting.
> + fn:echo:2: write error: broken pipe
> + fn:2: write error: inappropriate ioctl for device
>...
> I wonder if that's another case of a missing "2>/dev/null" like we
> fixed a few before 5.1?

OK, I suppose

  echo hello | >foo

in the case of SH_NULLCMD, which is equivalent to

  echo hello | : >foo

is indeed liable to the race in question, which is entirely benign for
this test.  It shouldn't fail for the case where this is doing a cat,
but the stdout test output already checks we picked that up, so
suppressing the error looks reasonable.

How about changing it to

  { echo hello | >foo } 2>/dev/null

?  Is that good enough?

pws