Bug 13753

Forgot to mention: this is 7.0.2, and I'm seeing the bugs in src/mesa/shaders/shader_api.c.

(In reply to comment #0)
> These are all bugs I noticed while trying to track another bug through the Mesa
> source. I'd be happy to work on a patch during the next few weeks, but I'd to
> first check which are genuine bugs (i.e., oversights) and which are just due to
> things being incomplete and are being worked on.
> 
> - glGetUniformLocation doesn't allow array elements to be queried (there are
> some comments in the code that seem to indicate that this is known)

A known issue.

> - glUniformMatrix ignores the count argument totally

a bug

> - glUniform1i only allows 1 sampler to be set in an array

known issue

> - glUniform* does not validate the type from the entry point against the type
> of the uniform

bug

> - glUniform* raises an error if location == -1, although this is defined as a
> no-op in the standard

bug

> - glUniform* does not convert int/float to boolean when setting a boolean
> uniform

bug


> - glUniform* does not allow count to exceed the size of a uniform array (which
> is defined as legal in the standard, with excess elements ignored)

bug


> - in glUniform*, a count of around 2^30 and 4-element vectors may lead to an
> integer overflow when checking for array overrun

bug

> - many of the bugs in glUniform* also apply to glUniformMatrix*

I probably won't get to fix these until next year.  Feel free to post patches...

I've checked Mesa out of anonymous git and written 4 patches to address some of the bugs above. What's the best way to get them to you?

I think Brian will be offline until the new year, but you can just attach the patches to this bug.

Created attachment 13294 [details] [review]
Patch 1/5

Created attachment 13295 [details] [review]
Patch 2/5

Created attachment 13296 [details] [review]
Patch 3/5

Created attachment 13297 [details] [review]
Patch 4/5

Created attachment 13298 [details] [review]
Patch 5/5

I've attached a series of 5 git patches (never really used git before so let me know if I stuffed it up). They fix most of these bugs that can be fixed locally (i.e. without messing with other files and causing unintended consequences). They also finish off Brian's fix to bug 13751, and add a prog/tests/shader_api regression test (maybe should get renamed to bug_13751, that's up to someone else). The regression test doesn't work on everything because some of the bugs are masked by other bugs (e.g. the shader compiler asserts when given an array of samplers).

The problems still outstanding in shader_api.c
- glGetUniformLocation and glGetActiveUniform are still brain-damaged because the compiler doesn't provide the right information.
- passing count > 1 on a non-array is now silently accepted, because again the compiler doesn't provide enough information to distinguish between a 1-element array and a non-array (unless I missed something).
- glGetActiveAttrib almost always gets the type wrong (but less wrong than before), again because the compiler isn't providing the necessary information.

*** Bug 13751 has been marked as a duplicate of this bug. ***

I've commited your patches to git (both master and mesa_7_0_branch).

I may not get around to the remaining issues for a while though.

I think these are all fixed now. At least GetActiveUniform and count > 1 should be fine, and I think shaderAPI is testing attribute types.