From: Keith W. <kei...@go...> - 2010-03-21 18:46:54
|
On Sun, Mar 21, 2010 at 5:57 PM, George Sapountzis <gsa...@gm...> wrote: > On Sun, Mar 21, 2010 at 7:43 PM, George Sapountzis > <gsa...@gm...> wrote: >> On Sun, Mar 21, 2010 at 7:22 PM, Keith Whitwell >> <kei...@go...> wrote: >>> George, >>> >>> This is basically a reproduction of a facility I had in target-helpers >>> previously (swrast_screen.c or similar), which I removed after >>> feedback from Jose that supporting it with scons created more ugliness >>> than we saved in the C code. >>> > > Ah ok, i was not aware of this > >>> This change breaks layering in the build system by making code in the >>> utility libraries conditionally built depending on which targets we're >>> supporting. Ideally the code in auxilliary wouldn't have any idea >>> whether softpipe, llvmpipe or any other driver is out there. >>> > > This commit did not build soft_screen.c with auxiliary but symlinked > it at each target and build it there, I guess this is not pretty > either. For the record, the scons build seemed to work (after getting > broken) as reported by Xavier. OK, that would overcome the issue I was thinking of, but agree it's not pretty either... >>> For the meantime, I'd say just duplicate the function in the few >>> places which use it. There aren't many currently. Longer term, I >>> think we probably want a little targets/common or similar, rather than >>> trying to bundle this into auxilliary. >>> > > Ah ok, so it will be build once and all places will use the same > software renderer by default. That's the hope... >> >> Ok, I just reverted the commit and the commits that fix it. It seems >> that there will be three places using it (libgl-xlib, drm/sw and >> drm/swrast). I'll give it a try some other day when I can read and >> type. >> > Thanks George. Keith |