The GPU info api we have is meant to be a general facility for use anywhere within Cogl where structured version, vendor and architecture information is useful. The plan was even to make something like that api public at some point too.

Although the -feature-functions.h only a basic mechanism for automatically enabling certain features according to extensions which handles common cases, when that isn't flexible enough we can do more special case checks in cogl-winsys-glx.c:update_base_winsys_features() to enable/disable features.

Since the GpuInfo can't be initialized until we have a GL context (until then we can't call glGetString) we can also add special case checks in cogl-winsys-glx.c:update_winsys_features() which lets us hook in when there is a context and the gpu info is initialized by the call to _cogl_context_update_features() at the top of that function.

My guess a.t.m is that we should update the checks where we currently apply the workaround in update_winsys_features() so it goes something like like:

if (glx_renderer->pf_glXCopySubBuffer || context->glBlitFramebuffer)
{
  CoglGpuInfo *info = &context->gpu;
  CoglGpuInfoArchitecture arch = info->architecture;

  COGL_FLAGS_SET (context->winsys_features, COGL_WINSYS_FEATURE_SWAP_REGION, TRUE);
  if (info->driver_package == COGL_GPU_INFO_DRIVER_PACKAGE_MESA &&
      info->driver_package_version < COGL_VERSION_ENCODE (10, 1, 0) &&
      (arch == COGL_GPU_INFO_ARCHITECTURE_LLVMPIPE ||
       arch == COGL_GPU_INFO_ARCHITECTURE_SOFTPIPE ||
       arch == COGL_GPU_INFO_ARCHITECTURE_SWRAST))
    {
      COGL_FLAGS_SET (context->winsys_features, COGL_WINSYS_FEATURE_SWAP_REGION, FALSE);
    }
}

The GpuInfo should already be detecting when you are running with a Mesa driver and its version so hopefully you don't need to change anything there.

Created attachment 265440 [details] [review]
winsys-glx: Reenable swap_region for llvmpipe and swrast

The bug that prevented MESA_copy_sub_buffer to work for
swrast / llvmpipe got fixed in mesa 10.1 git so enable
it for mesa 10.1+.

---

OK, so lets do it this way then.

(In reply to comment #2)
> Created an attachment (id=265440) [details] [review]
> winsys-glx: Reenable swap_region for llvmpipe and swrast
> 
> The bug that prevented MESA_copy_sub_buffer to work for
> swrast / llvmpipe got fixed in mesa 10.1 git so enable
> it for mesa 10.1+.
> 
> ---
> 
> OK, so lets do it this way then.

+          (arch == COGL_GPU_INFO_ARCHITECTURE_LLVMPIPE &&
+           arch == COGL_GPU_INFO_ARCHITECTURE_SOFTPIPE &&
+           arch == COGL_GPU_INFO_ARCHITECTURE_SWRAST))

I think this should be using || instead of && here.

I think it should be ok also to remove the following marker in the comment now that the workaround has essentially been resolved:

/* XXX: ONGOING BUG:
 * (Don't change the line above since we use this to grep for
 * un-resolved bug workarounds as part of the release process.)

Created attachment 265451 [details] [review]
winsys-glx: Reenable swap_region for llvmpipe and swrast

The bug that prevented MESA_copy_sub_buffer to work for
swrast / llvmpipe got fixed in mesa 10.1 git so enable
it for mesa 10.1+.

Comment on attachment 265451 [details] [review]
winsys-glx: Reenable swap_region for llvmpipe and swrast

Thanks I have pushed the patch to master and the cogl-1.18 branch. I made one tweak to #include "cogl-version.h" since I got a warning about COGL_VERSION_ENCODE being implicitly defined when building.

For reference, the original is bug #674208.