Adjust workaround for mipmap-related OpenGL implementation bug on Mac OS X.

gpu/command_buffer/service/gles2_cmd_decoder.cc

Index: gpu/command_buffer/service/gles2_cmd_decoder.cc
===================================================================
--- gpu/command_buffer/service/gles2_cmd_decoder.cc	(revision 106436)
+++ gpu/command_buffer/service/gles2_cmd_decoder.cc	(working copy)
@@ -3008,8 +3008,11 @@
   }
   // Workaround for Mac driver bug. In the large scheme of things setting
   // glTexParamter twice for glGenerateMipmap is probably not a lage performance
-  // hit so there's probably no need to make this conditional.
-  glTexParameteri(target, GL_TEXTURE_MIN_FILTER, GL_NEAREST_MIPMAP_LINEAR);
+  // hit so there's probably no need to make this conditional. The bug appears
+  // to be that if the filtering mode is set to something that doesn't require
+  // mipmaps for rendering, or is never set to something other than the default,
+  // then glGenerateMipmap misbehaves.
+  glTexParameteri(target, GL_TEXTURE_MIN_FILTER, GL_NEAREST_MIPMAP_NEAREST);

[greggman, 2011/10/20 23:27:37]

It seems like there are 2 possibliies

1) The state needs to be GL_NEAREST_MIPMAP_NEAREST
2) The state needs to be dirty

If it's #2 this won't fix it in certain cases. That seems easy to test

genTex
bindTex
texParm(GL_NEAREST_MIPMAP_NEAREST
texImage2d(red
genMips
texImage2d(blue
genMips

should we test that?

[Ken Russell (switch to Gerrit), 2011/10/20 23:36:24]

I believe it's neither, but a third possibility: the filter mode needs to have
been changed from the default at least once, and also needs to be set to
something requiring mipmaps.

   glGenerateMipmapEXT(target);
   glTexParameteri(target, GL_TEXTURE_MIN_FILTER, info->min_filter());
 }