On Thu, Jul 2, 2009 at 7:45 PM, Mike Frysinger<[email protected]> wrote: > On Thursday 02 July 2009 22:00:41 Garrett Cooper wrote: >> On Thu, Jul 2, 2009 at 6:57 PM, Mike Frysinger<[email protected]> wrote: >> > On Thursday 02 July 2009 21:21:06 Garrett Cooper wrote: >> >> Both sinclude and include are possible I suppose, but as far as I >> >> understand that actually does an fstat, tries to open the file, etc, >> >> but this completely ignores the include statement and reduces I/O a bit. >> > >> > yes, it'll remove a syscall or two, but i dont think that is significant >> > in the larger scheme of things >> > >> > the problem is that i dont want these kind of ifdef things propagating as >> > they'll get tweaked and go inconsistent all over the place >> > >> > how about include a top level .mk file that always exists and in there, >> > you have the conditional config.mk include >> >> I couldn't agree more :)... >> >> FWIW using include all over the place has footprint implications >> (somewhere on the order of 2~3 times longer when taxing the fs a bit >> more), so it could be a boon to multijob compiles to go with sinclude >> / -include. > > well, if you can see tangible differences, then it's worth discussing. i dont > think the include style would make a difference in this regard at all, it'd > only change the behavior when the file doesnt exist (warn/error/etc...).
Yeah, but I'm trying to make maximize compiles while maximizing repeatability. Hence, that's why there are comments in some of the files about srcdir :)... > how about we aim for avoiding the include completely regardless of target. > really config.mk is setting up the environment and recursive makes should be > inheriting those. so if our toplevel .mk exported something like: > LTP_CONFIG_INCLUDED = yes > > then all the fragments in the sub-makefiles should look like: > ifneq ($(LTP_CONFIG_INCLUDED),yes) > include $(topdir)/config.mk > endif > > then you could have the clean target lie by doing: > $(MAKE) -C testcases LTP_CONFIG_INCLUDED=yes clean > > which would address all of your concerns ? Yes. That's sort of the concern that I was going for with my previous spin, but then it turned into a gob of Makefile.inc files that were being added unnecessarily to the SCM. Considering that the include file is only needed for files that include config.mk (which are a small handful -- 10 to be exact), I figured that adding the necessary logic to the top would be wise. >> People already complain about compiling LTP in my group -- I'm just >> trying to give them less of a reason to complain :).. > > bah, tell them to build it in a tmpfs partition then ;) Best reasoning I've heard all week :D. There's an accidentally bug with pathing in my diff (that wasn't exposed because it was just using in the CFLAGS section) that I'm going to fix and resubmit patches for tomorrow after doing more extensive development and testing on my Gentoo box at home. More changes coming down the pike soon (tomorrow) as well... Thanks, -Garrett ------------------------------------------------------------------------------ _______________________________________________ Ltp-list mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/ltp-list
