On Mon, Jun 27, 2011 at 12:26 PM, Norbert Thiebaud <nthieb...@gmail.com> wrote: > On Mon, Jun 27, 2011 at 12:03 PM, Tor Lillqvist <tlillqv...@novell.com> wrote: >>> If we do that, we definitely should then also add built-in mkdir and cp >>> commands in it, >> >> Hmm, or actually, I don't think that will be such a great win after all, as >> the gbuild recipies where tons of mkdir commands are being run typically are >> in a shell expression with && anyway, so they couldn't be run as "built-in" >> simple make commands anyway. Forget it. > > Yeah, but maybe there is something to be investigated to avoid fork > when running recipies... I've read somewhere that spawn was much more > performant than fork under cywin (note: I don;t know if make already > do that or not, nor what are the implication...) > > Another thing: I think most of these mkdir could be avoided at the > cost of another layer of dependencies: create rules for every target > so that the parent directory is a pre-req target and have rules for > directories to build them... that should put most of the the workload > on make itself an limit drastically the number of mkdir...
Another solution is a quick and dirty path to make to have ot try to create the base directory of a target before running a recipe for it. i don't like it because it will never be accepted upstream and would prevent building without a patched version... But since there is a platform/* support we could have conditional on windows to not do the mkdir if we have the 'right' make. (iow maintain the buidability with a vanilla make, but still improve perf when a 'lo-make' is available.) Norbert _______________________________________________ LibreOffice mailing list LibreOffice@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/libreoffice
