xf86-video-2.4.2 (and possibly some earlier versions) leaked out xf86mm.h, so if 2.4.2 finds libdrm-2.4.0 it will try to include xf86mm.h for drmBo and fail, because the actual libdrm-2.4.0 release didn't get xf86mm.h. All versions that check for libdrm-2.4.0 in configure.ac for the purpose of including xf86mm.h and using that API should be patched to not do so, so that libdrm-2.4 could eventually be added to the tree. Edited excerpt from #intel-gfx: <leio> when I downgraded to xf86-video-i810-2.4.2 (with some patches on top from 2.4 branch) with libdrm-2.4.0 on system it failed <leio> apparently can't find xf86mm.h header <leio> I can't see any references or existence of that in neither 2.3.1 or 2.4.0, bizarre <airlied> leio: xf86mm.h leaked out.. the driver shouldn't have depended on it. <leio> right, just chased this down too <leio> the 2.4.2 release defines XF86DRI_MM only if libdrm >= 2.4.0, but 2.4.0 doesn't have it <leio> now just need to figure out a painless upgrade path.. <leio> patching 2.4.2 to not enable XF86DRI_MM ever maybe <airlied> yup.. that is probably right.
On a slightly unrelated note, libdrm-2.4.0 and xf86-video-intel-2.4.98 seem to work fine for me (I have ebuilds with slight dependency mods according to changes). Just text drawing performance is worse than before with xorg-server-1.5.2 (110k glyphs per second with intel-2.4.2, 46k glyphs per second with intel-2.4.98). Apparently the new driver version more desperately begs for Owen Taylor's glyph cache implementation from xorg-server master than the 2.4.2 version did, as I read out from discussion with KeithP - xorg-server-1.5.2 doesn't have that glyph cache. I already backported few pieces of them for my r200 long ago, I will probably backport again and see what happens then.
2.4.2-r3 is libdrm 2.4.0 ready. I'll try building other drivers in the morning. Thanks
All other drivers I've tried are fine. libdrm 2.4.0 has been unleashed. Thanks