On Mon, Aug 3, 2009 at 9:27 PM, Peter Hutterer<[email protected]> wrote: > On Tue, Aug 04, 2009 at 01:51:52PM +1000, Peter Hutterer wrote: >> On Mon, Aug 03, 2009 at 07:51:02PM -0700, Dan Nicholson wrote: >> > On Mon, Aug 3, 2009 at 7:39 PM, Peter Hutterer<[email protected]> >> > wrote: >> > > libman_ref are all the refpages that are just links to the actual man >> > > page >> > > (e.g. XIUngrabButton.man is a refpage to XIGrabButton.man). >> > > >> > > This patch forces the all actual man pages to be complete before the >> > > refpages target can be started. >> > > >> > > Previously, some ref pages could have started parallel to the man pages >> > > still being generated. The ref pages rely on their respective man pages >> > > but >> > > they may not have existed yet, leading to build errors. >> > > >> > > Signed-off-by: Peter Hutterer <[email protected]> >> > > --- >> > > I think this should be the correct approach to address the parallel build >> > > errors we see in libXi, at least I can't reproduce it anymore with >> > > different >> > > -j settings. >> > >> > I don't see how it makes any difference, really. Take >> > XChangeDeviceControl.man, for instance. Rather than just waiting for >> > XGetDeviceControl.man to show up, it will now wait for all the real >> > manpages to show up. It really shouldn't need to do that for any >> > reason. Now consider the situation where XChangeDeviceControl.man >> > doesn't exist for some reason. We want to trigger >> > XGetDeviceControl.man to get rebuilt, but now the rule has all the >> > real pages as prerequisites. This will cause all the pages to get >> > removed and rebuilt. It's a pretty big hammer, but it would work if >> > you're not confident in the approach that's in there. >> > >> > Also, I played around quite a bit this morning on this and I'm pretty >> > sure that what's in their now works. >> >> It doesn't. MAKEFLAGS=-j 5 gives me consistent build errors without this >> patch. try 'git clean -dxf && ./autogen.sh && make'. it's just the first run >> of make that breaks. >> >> anyway, I found the real problem while writing this email. some ref targets >> didn't have the necessary .man, so the target was >> XIUndefineCursor: XIDefineCursor.man >> instead of >> XIUndefineCursor.man: XIDefineCursor.man >> >> bloody thing was hard to spot, I read over that a zillion times. >> XDeviceTimeCoords was affected by this too. >> >> So nevermind about the patch before, I'll just fix up the two targets. > > argh. nevermind. I just found you fixed this yesterday and I didn't notice. > thanks for fixing it though.
Oops. I figured you'd noticed my commit and that's why you were taking another stab at this. It was tough to figure out, too. :) If you have the stomach for the firehose it produces, "make -d" can usually help find things like this. -- Dan _______________________________________________ xorg-devel mailing list [email protected] http://lists.x.org/mailman/listinfo/xorg-devel
