[patch] Use "set -o pipefail"

[Stephen Lin]

Just curious, what's the reason for changing the behavior of 'not' on
a crashing program? Do we currently have tests depend on programs
"crashing" in a particular way?

On Thu, Jul 4, 2013 at 10:27 AM, Rafael Espíndola
<rafael.espindola at gmail.com> wrote:
> Ping.
>
> An updated patch is attached. Testing it found all the problems I just
> fixed in existing tests. This patch updates both the internal and the
> shell based testing system. It also changes 'not' so that 'not
> a_crashing_program' return 0.
>
> On 17 June 2013 10:16, Rafael Espíndola <rafael.espindola at gmail.com> wrote:
>> ping.
>>
>> On 12 June 2013 14:03, Rafael Espíndola <rafael.espindola at gmail.com> wrote:
>>> It is fairly inconvenient for tests like
>>>
>>> opt ... | FileCheck
>>>
>>> to pass if opt returns false after printing what FileCheck was looking
>>> for. The attached patch adds "set -o pipefail" to the testscript that
>>> lit creates.
>>>
>>> The patch adds a test that only passes if pipefail is enabled and
>>> deletes to old tests:
>>>
>>> * test/CodeGen/X86/2007-05-07-InvokeSRet.ll: It had invalid IL.
>>> * test/CodeGen/X86/2008-02-08-LoadFoldingBug.ll: it was running "grep
>>> andpd | not grep esp" and passing because we were not printing any
>>> andpd.
>>>
>>> I have tested that it works on linux and OS X. Is it OK if I also
>>> update clang and lld tests?
>>>
>>> Cheers,
>>> Rafael
>
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>