Arbitrarily large output can't be used in an exception.

Arbitrarily large output can't be used in an exception.

It seems like we are seeing whacky errors with sys.stderr
dying when trying to print out a process and its exception.
I think controlling how the output is printed in RunCommand
is better than passing it with the exception.

Change-Id: I3711575e5d9e710c3aea6ff19d43d149fd83f44c

BUG=chromium-os:14398
TEST=Ran it with au_test_harness

Committed: http://chrome-svn/viewvc/chromeos?view=rev&revision=237b686

[dgarrett, 2011-04-28 22:27:21]

The code change LGTM, but with a couple of general questions....

However, I'm surprised. Are the large exceptions causing the problem, or is it
related to the pickling for interprocess communication that happens later on?

Also, if multiple parallel tests fail, will the outputs end up interleaved with
each other?

[sosa, 2011-04-28 22:31:32]

The actual errors seem to occur when the buffer for sys.stderr becomes
full.  It's hard to repro so I'm hoping that flush stderr directly
after printing the larger output, we actually clean the buffer.  The
other alternative we could consider is running the au_test_harness w/
PYTHONUNBUFFERED

On Thu, Apr 28, 2011 at 3:27 PM,  <dgarrett@chromium.org> wrote:
> The code change LGTM, but with a couple of general questions....
>
> However, I'm surprised. Are the large exceptions causing the problem, or is
> it
> related to the pickling for interprocess communication that happens later
> on?
>
> Also, if multiple parallel tests fail, will the outputs end up interleaved
> with
> each other?
>
> http://codereview.chromium.org/6902129/
>

[sosa, 2011-04-28 23:26:09]

Actually reading the docs on sys.stderr, it's already UNBUFFERED.  So
my other solution wouldn't have worked.  As far as whether the errors
would be interspersed, this would be true.  I think in general this is
happening to us because we're printing huge blobs to sys.stderr in
multiple processes where it's usually expecting one line at a time.

On Thu, Apr 28, 2011 at 3:31 PM, Chris Sosa <sosa@chromium.org> wrote:
> The actual errors seem to occur when the buffer for sys.stderr becomes
> full.  It's hard to repro so I'm hoping that flush stderr directly
> after printing the larger output, we actually clean the buffer.  The
> other alternative we could consider is running the au_test_harness w/
> PYTHONUNBUFFERED
>
> On Thu, Apr 28, 2011 at 3:27 PM,  <dgarrett@chromium.org> wrote:
>> The code change LGTM, but with a couple of general questions....
>>
>> However, I'm surprised. Are the large exceptions causing the problem, or is
>> it
>> related to the pickling for interprocess communication that happens later
>> on?
>>
>> Also, if multiple parallel tests fail, will the outputs end up interleaved
>> with
>> each other?
>>
>> http://codereview.chromium.org/6902129/
>>
>