Re: parallel-tests: work around Tru64/OSF 5.1 sh read bugs.

[Stefano Lattarini]

On Sunday 16 January 2011, Ralf Wildenhues wrote:
> * Stefano Lattarini wrote on Sun, Jan 16, 2011 at 02:37:55PM CET:
> > On Sunday 16 January 2011, Ralf Wildenhues wrote:
> > > http://autobuild.josefsson.org/automake/log-201101150748140530000.txt
> 
> > > This is due to this portability issue:
> > > http://thread.gmane.org/gmane.comp.sysutils.autoconf.patches/7522
> > > 
> > > Here's a patch to fix the instances I could find in the Automake tree.
> > > Unfortunately, I don't see an easy way to expose the first two hunks in
> > > check.am, they would require making a log unreadable between the test
> > > running and the summary generation.
> > >
> > What about the attached testcase?  Unfortunately, I cannot really test
> > it because I lack access to a Tru64/OSF 5.1 system.
> 
> Thanks for writing the test case, and good job, it exposes the first
> read issue.
>
Oh, I thought it exposed also the second one.  After all, if the
shell bails out on a failed read, I'd expect the `$(TEST_SUITE_LOG)'
file not to be created, and us to be left simply with the temporary
file `$(TEST_SUITE_LOG).tmp'.  But maybe the Tru64/OSF 5.1 shell is
one of those shell which execute a redirected compound command
{ ...; } in a subshell?  In that case, the second `test -r' check
wasn't really warrented (altough IMHO it's good to have it, for
consistency), so no big deal not to have it covered.

> I'm squashing it in, adding a 'cat stdout' to it in the
> non-error case,
>
Oops, sorry about that.

> you as author to the patch, and pushing to maint.
> 
> Cheers,
> Ralf
> 
Thanks,
  Stefano