[Piglit] [PATCH 2/2] framework: Don't try to parse GL/GLSL version from wflinfo.

[Jose Fonseca]

On 07/12/15 18:12, Dylan Baker wrote:
> On Mon, Dec 07, 2015 at 02:26:00PM +0000, Jose Fonseca wrote:
>> It's making many assumptions about the wflinfo which are not true.
>>
>> So completely disable it as a workaround.
>>
>> Though I wonder if there's really any merit in adding a depending on
>> wflinfo.  IMO, if piglit cares for the advertised GL/GLSL versions, it
>> should have its own internally utility program to dump every thing needed.
>> Parsing output from external utilities is begging for trouble.
>
> If you don't have wflinfo, it'll go on and run anyway.

Yes, that was the workaround I took for my test slaves.

 > It's also
> designed to be as conservative as possible with versions, and we should
> absolutely try/except the float if a driver doesn't return a sane
> version.
 >
> Out of curiosity what driver are you using, I tested this extensively
> against both i965 and against RadionSI. I had no issues at all, when
> anything failed it would simply fall back to running the test and
> letting the binary skip itself, as it was designed to do.

It happened with all drivers I normally:


- On NVIDIA it crashes because version string is "4.4.0 ...", i.e., 
"major.minor.patch".  I believe this is fine.

   Traceback (most recent call last):
   File 
"/home/jfonseca/work/vmware/tests/piglit/framework/test/base.py", line 
181, in execute
     self.run()
   File 
"/home/jfonseca/work/vmware/tests/piglit/framework/test/base.py", line 
234, in run
     self.is_skip()
   File 
"/home/jfonseca/work/vmware/tests/piglit/framework/test/opengl.py", line 
314, in is_skip
     if (self.__info.gl_version is not None
   File "/home/jfonseca/work/vmware/tests/piglit/framework/core.py", 
line 204, in __get__
     value = self.__func(instance)
   File 
"/home/jfonseca/work/vmware/tests/piglit/framework/test/opengl.py", line 
182, in gl_version
     raw.split('\n'), 'OpenGL version string').split()[3])
ValueError: invalid literal for float(): 4.4.0


   $ wflinfo --platform glx --api gl --profile core
   Waffle platform: glx
   Waffle api: gl
   OpenGL vendor string: NVIDIA Corporation
   OpenGL renderer string: Quadro K1000M/PCIe/SSE2
   OpenGL version string: 4.4.0 NVIDIA 355.11
   OpenGL context flags: 0x0


- With Mesa Xlib libGL.so.1:

Traceback (most recent call last):
   File 
"/home/jfonseca/work/vmware/tests/piglit/framework/test/base.py", line 
181, in execute
     self.run()
   File 
"/home/jfonseca/work/vmware/tests/piglit/framework/test/base.py", line 
234, in run
     self.is_skip()
   File 
"/home/jfonseca/work/vmware/tests/piglit/framework/test/opengl.py", line 
322, in is_skip
     if (self.__info.gles_version is not None
   File "/home/jfonseca/work/vmware/tests/piglit/framework/core.py", 
line 204, in __get__
     value = self.__func(instance)
   File 
"/home/jfonseca/work/vmware/tests/piglit/framework/test/opengl.py", line 
214, in gles_version
     raw.split('\n'), 'OpenGL version string').split()[5])

   $ wflinfo --platform glx --api gles3
   Waffle platform: glx
   Waffle api: gles3
   OpenGL vendor string: WFLINFO_GL_ERROR
   OpenGL renderer string: WFLINFO_GL_ERROR
   OpenGL version string: WFLINFO_GL_ERROR

   Not sure why this happens.  Anyway, my interest is GL + GLX, not so 
much GLES.

> FYI:
> I asked Chad (who maintains waffle) before I started this, and he made
> the assertion that wflinfo output was designed for parsing by external
> utilities and was considered API stable.
>
> Also, I would prefer to have some kind of bindings for wflinfo rather
> than parsing the string output, the goal was to have JSON interface
> (google also wanted this interface for other uses), but neither they nor
> I have had time to send anything.

All our tests use Waffle/Glut/etc.  I think we could have simple "dummy" 
tests that dump the desired versions.  I don't see the need for external 
tools or complex bindings.  By doing something different than our tests 
do, we're risking creating contexts slightly differently, and end up 
making wrong assumptions.

Jose