[PATCH 00/14] GLX cleanup, round two

[Adam Jackson]

Largely a resend of the last series, but patches 1 and 14 are novel.

The first is _wrong_, wrong wrong wrong, but it makes piglit actually
runnable in indirect contexts so in the spirit of showing one's work
it's included.  I don't know how I'm going to fix that particular issue
yet, but I believe it's sufficiently orthogonal from the rest of the
changes that they can be vetted for correctness anyway; nothing else in
the series has anything to do with context or drawable lifecycle, so.

Patch 14 makes a pile of clutter conformance tests start passing while
introducing no regressions.  That it's a portability win is also kind
of cool.

The entire series survives piglit/quick up to the point of
glean/texCombine, whereupon llvmpipe segfaults.  Simply applying the
first patch gets only as far as glean/blendFunc before llvmpipe
segfaults.  If we limit the test results to everything up to blendFunc:

	Before	    After
Pass	   345        368
Fail	    22          7
Skip	   209        202
Total      576	      577

Where "before" obviously includes patch 1; it's all sadness and despair
if you don't.

Thanks to James Jones for checking that my reading of the GLX spec
appears to match what NVIDIA implements.

- ajax