On Mo, 2014-07-21 at 13:02 -0400, Joey Hess wrote: > Sebastian Dröge wrote: > > Ok, mine apparently fails to build dll-split properly again. Although it > > looks like every call to gcc was with -fuse-ld=gold and ld.gold was used > > as a linker everywhere. > > > > Joey, how did you build your version successfully? > > Edited rts/OldARMAtomic.c to remove the ifdefs around the spin lock > code Why? > and used the ld shimming in the github gist. Right, that's what I was trying to get around with my attempts as we can't really change /usr/bin/ld to point to ld.gold... and even if we do it temporarily during the build the resulting compiler would use the wrong linker in the end. I was trying to have the resulting linker use ld.gold all the time, and then also have usage of ld.gold as default in the "settings" file. It looked like it all worked but it seems like I missed at least once place where ld is called. What I did is described in that GHC trac ticket, but I'll check a bit more what went wrong in the next days.
Attachment:
signature.asc
Description: This is a digitally signed message part