[Piglit] Silent Python exceptions

[Jose Fonseca]

On 07/12/15 13:55, Ilia Mirkin wrote:
> On Mon, Dec 7, 2015 at 8:26 AM, Jose Fonseca <jfonseca at vmware.com> wrote:
>>
>>
>> Is there any way to tell piglit framework to not trap and hide Python
>> exceptions, so they can be debugged?
>>
>>
>> Also, I'd much prefer that Python exceptions would cause the abortion of
>> whole testsuite, instead of being captured into the results.
>>
>> IMO, the framework should be (or aim to be) rock solid -- exceptions in the
>> framework are not the same as test failures -- but rather catastrophic
>> failures which should never happen in practice.
>
> +1
>
> I was pretty opposed to this sort of "user-friendly" hiding as well.
> Although I think this one's gone a step further -- normally you at
> least see "BUG: I know what the bug is but I'm not going to tell you"
> on the console. I guess this was in a different bit of code. We're all
> developers here... just show us the stacktrace :)

I finally managed to get a stack trace.  Basically all the GL/GLSL 
version parsing in piglit/framework/test/opengl.py is totally hosed.  At 
least with wflinfo that comes with Ubuntu 15.10.


But I'm really frustrated that, thanks to catching exceptions, and 
redirecting stderr, it took me almost two hours just to get stack 
backtrace trace -- I had to figure out where the exception handler was, 
then open /dev/stderr to actually get it out somehow.


This is madness  -- 2 hours to get basic information that should have 
been there from the start.


For our sanity sake, could we please agree on sticking to the following 
basic principles in the framework:

- though shall not catch exceptions
- though shall not override stderr


Jose