Upstream generally considers every .0 a testing release, but if there's a reason, it might be OK for us to switch early without waiting for .1. (We have stable branches after all.)

Main reason for the bump is just for Iris improvements so waiting for .1 is fine though.

Can iris be used for X/Glamor easily now?

Without the environment variable? It depends on the CPU you're using. [1]

[1] https://cgit.freedesktop.org/mesa/mesa/commit/?id=cfbde3282d9b881a8cb09dc7a9c2b25b69872e32&utm_source=anzwix

Without the environment variable? It depends on the CPU you're using. [1]

[1] https://cgit.freedesktop.org/mesa/mesa/commit/?id=cfbde3282d9b881a8cb09dc7a9c2b25b69872e32&utm_source=anzwix

Ah thanks for the pointer :-)

btw newer graphics cards require mesa 19.2 with llvm9. see https://gitlab.freedesktop.org/mareko/mesa/commit/594010e366f911581ca0a4471a9d9fa68116514f

llvm9 is in available in nixpkgs master now. I tried to build this PR with it, but I haven't tested these in any other way. I expect we'll want to switch to support these GPUs by default.

I have tested it. x11/wayland/kmscube doesn't start without llvm9 on navi10 gpus.

This breaks SDL2: https://bugzilla.libsdl.org/show_bug.cgi?id=4797

I fixed it here with the following overlay:

  SDL2 = super.SDL2.overrideAttrs (attrs: {
    postPatch = ''
      substituteInPlace include/SDL_opengl_glext.h \
        --replace "typedef ptrdiff_t GLsizeiptr;" "typedef signed long int khronos_ssize_t; typedef khronos_ssize_t GLsizeiptr;" \
        --replace "typedef ptrdiff_t GLintptr;" "typedef signed long int khronos_intptr_t; typedef khronos_intptr_t GLintptr;"
    '';
  });

looks like there's already a bugfix release for 19.2:

https://lists.freedesktop.org/archives/mesa-announce/2019-October/000548.html

ok I'll do a new PR that passes llvm9

@jb55: I did these together, see 53a97c3.

@vcunat awesome! thanks