Print out at-exit ASSERT FAILURES too

Print out at-exit ASSERT FAILURES too
Long term solution is to tweak Dr. Memory to print out such errors to stderr

TBR=bruening,rnk
Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=111138

[Timur Iskhodzhanov, 2011-11-22 12:48:24]

TBR

[Reid Kleckner (google), 2011-11-22 14:30:55]

On 2011/11/22 12:48:24, Timur Iskhodzhanov wrote:
> TBR

LGTM

[Derek Bruening, 2011-11-22 15:49:45]

On 2011/11/22 12:48:24, Timur Iskhodzhanov wrote:
> Print out at-exit ASSERT FAILURES too
> Long term solution is to tweak Dr. Memory to print out such errors to stderr

I don't follow what this is solving: Dr. Memory today prints assertion messages
to stderr (unless -quiet), and to results.txt.  I thought drmemory_analyze
already looked for asserts in results.txt?  If you want them to stderr you
should remove the -quiet (and perhaps replace with -no_summary).

[Reid Kleckner (google), 2011-11-22 15:53:54]

On Tue, Nov 22, 2011 at 10:49 AM,  <bruening@chromium.org> wrote:
> On 2011/11/22 12:48:24, Timur Iskhodzhanov wrote:
>>
>> Print out at-exit ASSERT FAILURES too
>> Long term solution is to tweak Dr. Memory to print out such errors to
>> stderr
>
> I don't follow what this is solving: Dr. Memory today prints assertion
> messages
> to stderr (unless -quiet), and to results.txt.  I thought drmemory_analyze
> already looked for asserts in results.txt?  If you want them to stderr you
> should remove the -quiet (and perhaps replace with -no_summary).

This code is reading results.txt.  I'm going to send out a CL to
remove -quiet and -no_results_to_stderr, but also turn on -no_summary.
 This still prints version info and the app command for every
subprocess, though, which IMO we should find a way to suppress.

Reid

[Derek Bruening, 2011-11-22 15:57:10]

On 2011/11/22 15:53:54, Reid Kleckner (google) wrote:
> This code is reading results.txt.

But drmemory_analyze.py already pulls asserts out of results.txt.  So why is
this diff needed?  Is there an i# or email exchange I missed?  What's not
working in the existing solution?

[Timur Iskhodzhanov, 2011-11-22 16:02:21]

the old version of drmemory_analyze was not catching the assertions happening at
the end of the run (after the ERROR SUMMARY line).

The stderr printing is not enabled on the bots yet (except for the Win7/Dbg bot
where Reid enabled it locally, w/o commiting)

On 2011/11/22 15:57:10, Derek Bruening wrote:
> On 2011/11/22 15:53:54, Reid Kleckner (google) wrote:
> > This code is reading results.txt.
> 
> But drmemory_analyze.py already pulls asserts out of results.txt.  So why is
> this diff needed?  Is there an i# or email exchange I missed?  What's not
> working in the existing solution?

[Derek Bruening, 2011-11-22 16:05:33]

On 2011/11/22 16:02:21, Timur Iskhodzhanov wrote:
> the old version of drmemory_analyze was not catching the assertions happening
at
> the end of the run (after the ERROR SUMMARY line).

OK, I see.  LGTM, though if Reid is about to remove -quiet perhaps this diff is
unnecessary and should instead remove the other code looking for ASSERT (will
abort by default so will be at end of output).