Date: Sun, 14 Nov 2004 17:42:04 +0200 From: Maxim Sobolev <sobomax@portaone.com> To: Poul-Henning Kamp <phk@FreeBSD.ORG> Cc: cvs-all@FreeBSD.ORG Subject: Re: cvs commit: src/usr.bin/make job.c job.h main.c make.c Message-ID: <41977CCC.8070607@portaone.com> In-Reply-To: <200411120858.iAC8w7ng088956@repoman.freebsd.org> References: <200411120858.iAC8w7ng088956@repoman.freebsd.org>
next in thread | previous in thread | raw e-mail | index | archive | help
Great job! However I wonder if use of pipe(2) would be better way, which will not leave any traces behind it in the case when all makes die in a way that prevents them from taking clean up actions (e.g. kill -9). -Maxim Poul-Henning Kamp wrote: > phk 2004-11-12 08:58:07 UTC > > FreeBSD src repository > > Modified files: > usr.bin/make job.c job.h main.c make.c > Log: > Add code to enforce the paralleism count (-j N) for the entire tree > of submakes spawned during processing. > > We create a fifo and stuff one character into it for each job we are > allowed to run. The name of the fifo is passed to child processes > in the MAKE_JOBS_FIFO environment variable. > > A make which finds this variable on startup will open the fifo and > only spawn jobs when it managed to read a token from the fifo. > When the job completes a token is writen back to the fifo. > > Slave make processes get one token for free: the one their parent > make got in order to run them. This makes the make processes > themselves invisible in the process counts. > > The net effect is that "make -j 12 -s buildworld" will start at > most 12 jobs at the same time, instead of as previously up to > 65 jobs would get started. > > Revision Changes Path > 1.57 +114 -28 src/usr.bin/make/job.c > 1.26 +1 -1 src/usr.bin/make/job.h > 1.96 +2 -0 src/usr.bin/make/main.c > 1.25 +2 -2 src/usr.bin/make/make.c > > >
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?41977CCC.8070607>