Op 17 mei 2010, om 22:58 heeft Warren Young het volgende geschreven: > Most of the tests in a configure script don't actually have to run in series. > > Writing the code to analyze all the tests and build a dependency graph would be pretty tricky, though. Getting portable shells to launch N tests in parallel (where N is probably cores * 1.5 or so) at a time and integrate the results would be even trickier. You could do it easier in, say, Erlang, but then you've thrown portability out the window. GNU make with option -j comes to mind. Portability of GNU make is relatively OK, and the concurrent job execution there is pretty decent (OK load balancing even on make invocations). Another benefit of GNU make is that the dependency graph can be exposed to GNU make directly (by means of makefile rules), so the entire scheduling could be delegated to it. The key point from a technical perspective is not how to *run* the concurrency, it's to find it in the first place. The analysis itself will be the trickiest, agreed. Cheers -- Raphael 'kena' Poss · r.c.poss@xxxxxx · +31 20 525 7559 UvA · FNWI · IvI · CSA · 1098XG 107 Amsterdam (NL) z = \f. (\x. f (\y. x x y)) (\x. f (\y. x x y)) There is no spoon, only fork() and vfork(). _______________________________________________ Autoconf mailing list Autoconf@xxxxxxx http://lists.gnu.org/mailman/listinfo/autoconf
