We have two - two! - fake xlib-backed GLX implementations. The classic one uses classic swrast, the gallium one uses softpipe or llvmpipe. As a result the gallium one is almost uniformly superior, and this is one more consumer of classic swrast that we can hopefully get rid of. But I say "almost" for a reason, there are a few piglit regressions in gallium relative to classic.
This MR addresses piglit's glx-create-context-current-no-framebuffer test. classic escapes with a warning here because swrast maxes out at GL 2.1, and the test (and functionality) requires GL 3.0 or better.
The other failures are in actual rendering interactions, specifically glx-copy-sub-buffer, glx-multi-context-single-window and glx-tfp. For the last one I think the test is probably broken (see piglit#21), for the other two I suspect y-inversion shenanigans.