People don't seem to get VA-API working out of the box on Intel. They assume a global USE=vaapi will take care of that, and assume as such also from the presence of the USE flag on media-libs/mesa. To better serve our users out of the box experience, I would like us to consider adding a PDEPEND on xf86-video-intel to x11-libs/libva-intel-driver when USE=vaapi and/or perhaps reworking how USE=vaapi works on media-libs/mesa In case of mesa, it currently forces gallium and llvm even when VIDEO_CARDS only contains intel. I think if there's VIDEO_CARDS=intel, it should pull in libva-intel-driver too and not force gallium/llvm stuff when radeon and such isn't in VIDEO_CARDS, or at least tell in an elog or similar that to actually get vaapi for intel to install libva-intel-driver. But primarily as a quick improvement I suggest the xf86-video-intel PDEPEND.
I'd rather not want to see this added to xf86-video-intel. People might use the modesetting drivers and not get this dep then. As a "quick fix", how about adding a postinst message to mesa?
I don't think this can be expressed well in terms of USE flags for either mesa or xf86-video-intel. Maybe a metapackage for video acceleration is the better solution?
I don't think it would be all that awful to just go with that PDEPEND. It will cover most users who have intel for now (though only on X11), the rest probably figure it out from various documentation. I'll see about a proposed patch to mesa a bit later on for discussion, got distracted with other matters for today. Other than that, any improvements to some documentation on wikis and whatnot can be useful. This bug was motivated by someone on phoronix comments not enabling the functionality right (assuming USE flag will do), and even me myself not getting it right - I needed you to tell me on IRC that USE=vaapi on mesa is only for gallium stuff, not intel.
Enabling USE=vaapi pulls in libva, which has a VIDEO_CARDS=i965 that PDEPENDS on x11-libs/libva-intel-driver. So... this all looks fine to me.