[Piglit] [PATCH] egl: test that EGL_BAD_PARAMETER is returned for an invalid EGLImage attrib

[Emil Velikov]

On 12 January 2017 at 05:43, Tapani Pälli <tapani.palli at intel.com> wrote:
> On 01/11/2017 05:24 PM, Emil Velikov wrote:
>>
>> On 10 January 2017 at 10:51, Tapani Pälli <tapani.palli at intel.com> wrote:
>>>
>>> This test does not work correctly for me (skips because
>>> EGL_KHR_image_base
>>> is unsupported even though driver supports it). I believe you either
>>> would
>>> need to utilize egl-util framework (check the other tests in egl
>>> directory)
>>> or initialize EGL manually ground up like some tests do. Or is this
>>> depending on some compilation flags of Piglit?
>>>
>> Things are initialized. The issue is that the default/mixed_glx_egl
>> heuristics are "fun".
>>
>> Namely: most likely on your setup GLX gets picked thus things go bonkers.
>> All we need is to check that we're using EGL and bail our accordingly.
>>
>> The egl-utils helpers are quite dare I say it ... broken, so please
>> don't add any tests that use those.
>
>
> I voted to egl-utils instead of current logic as current logic is broken. It
> is true that egl-utils works only with X but currently we don't have other
> way to be sure we get EGL (with a test written for EGL!). We should probably
> start to work on changes to this so that one can reliably know what backend
> will be picked. It is crazy that EGL tests will be simply skipped on systems
> like Fedora 25.
>
Agreed it is pretty crazy. The !egl-utils codepath is iffy. With the
egl-utils one even more so - it effectively [part] duplicates core
piglit functionality - from input parsing to winsys setup.
The latter handles only X on the default screen. So things like gbm,
CI and alike are out of the question.

On any winsys in piglit fall a bit short - I have a sizeable WIP to
resolve that. We can use eglGetCurrentDisplay() [to determine "am I
EGL"] temporary until I polish that work ?

Emil