The rationale is in the change itself.

Signed-off-by: Nicolai Haehnle <nhaehnle@gmail.com>

Upload a texture with random data, use it in rendering, and attempt to read
the texture data back with glGetTexImage.

This catches yet another r300+bufmgr regression that was found via Sauerbraten.

Signed-off-by: Nicolai Haehnle <nhaehnle@gmail.com>

Regression test for a particular crash of the r300+bufmgr driver, triggered
by Sauerbraten. The crash occured when an application uploads cubemaps in a
non-canonical order.

Signed-off-by: Nicolai Haehnle <nhaehnle@gmail.com>

This exceedingly silly bug caused false negatives in the test results...

Signed-off-by: Nicolai Haehnle <nhaehnle@gmail.com>

If we have the external st3c library we'll see a message saying so, ignore it.
This fixes confusing "warn" results.

Conflicts:

	tests/glean/tfragprog1.cpp

This catches issues with offset handling that were missed by the original path
that copied the whole teximage at once.

This patch fixes a few issues with piglet on Cygwin/X (which uses Mesa
for OpenGL):

1) Cygwin requires that all dependencies be explicitly passed in
dependency order during linking.  This means that static libraries must
precede the shared libraries on which they depend.

2) Cygwin defines log2 as a macro instead of a function.

3) Cygwin doesn't provide lspci(8).

This catches bug #17865 on i915.  It's not a terribly complete test, and the
allowance for the driver to count indirections differently from the spec for
its native numbers means we don't get to justifiably fail very often.

Tests SGIS_generate_mipmap versus glTexImage2D.  Catches a mesa bug with
updating GL_GENERATE_MIPMAP state without flushing.

Tests that SGIS_generate_mipmap occurs after TexSubImage.  Catches bug #17077.

This is a very basic test that creates 3D textures of varying sizes,
renders them and compares the results with expected results.

It's still quite incomplete, e.g. a test for TexSubImage and mipmapped
textures is still missing.

Test some texture redefinition scenarios: Switching between different
texture sizes & mipmapping enable/disable

This is a regression test for the recently fixed KWin corruption bug.

Perhaps it'll help someone else, some day.

This test currently fails on R500.

Reveals problems on i965 accelerated copytexsubimage when using i915 code.

Current r500 fragment program implementation fails this test.

Old implementation of r300 driver fails this test.

Since 1D textures are really one-dimensional objects, the T texture coordinate
must be ignored completely. This is not the case for the r300 driver, which
simulates 1D textures via 2D textures, with "interesting" filtering results
depending on the TEXTURE_WRAP_T texture parameter.