> With these numbers I'm slightly in favor of using the one-automoc-target > per target approach (with a bit of cmake logic we could do this without > requiring the newer automoc4 version).
Alright, I see that the extra-target version is faster at make time. Since with cmake 2.6 running cmake is faster already there goes anther argument. My last argument for custom_command is that make target/fast would run automoc, but this can be helped by a little helper script that turns "make foo/fast" into "make foo_automoc/fast foo/fast"... I'll come up with a patch to keep compatibility where possible but switch to automoc targets when possible. > P.S. some time ago you removed some code in automoc which did something in > multiple parallel processes. Was this for running multiple moc's at once ? Yes. > Hmm, now that I have 4 cores I could make good use of that :-) > We could try to figure out the number of CPUs (e.g. using sysconfig()), and > then run that much mocs. Do you know whether the problem might be fixed in > current Qt ? When you call make -j5 then it can happen that you get 5 concurrently running automoc4 processes. The extra parallelization inside automoc4 isn't really necessary and for unexplainable reasons broke on FreeBSD and some other platforms. -- ________________________________________________________ Matthias Kretz (Germany) <>< http://Vir.homelinux.org/ _______________________________________________ Kde-buildsystem mailing list [email protected] https://mail.kde.org/mailman/listinfo/kde-buildsystem
