On Wed, 2023-03-01 at 13:05 -0600, Satish Balay wrote: > Perhaps my minimal test code is not an exact representation. However > this usage does work in our code-base. > > > > > > [balay@pj01 petsc]$ make --version |head -1 > GNU Make 4.3 > [balay@pj01 petsc]$ grep ^MAKE_NP configure.log > MAKE_NP = 6 > [balay@pj01 petsc]$ (ccache -c -C && make clean) > /dev/null && time > (make all) > /dev/null > > real 1m15.591s > user 6m4.442s > sys 1m9.179s > [balay@pj01 petsc]$ (ccache -c -C && make clean) > /dev/null && time > (make -j1 all) > /dev/null > > real 6m5.378s > user 5m5.263s > sys 0m59.175s > [balay@pj01 petsc]$ (ccache -c -C && make clean) > /dev/null && time > (make -j12 all) > /dev/null > > real 1m8.758s > user 10m58.865s > sys 1m41.094s > <<< > > Note: '-j6' is the value set in the sub-makefile for all the above > runs.
All of this behavior makes sense and aligns with what I wrote in my previous email message, with one caveat I didn't think of. What happens is this: In the first instance where you don't provide -j at all, the sub-make (that adds -j6) will create a new jobserver with 6 tokens and that make and all its sub-makes will share that jobserver. In the second instance where you provide -j1, the arguments passed to the sub-make (after you've modified them) will be: "-j6 -j1" because "- j1" appears in makeflags. I didn't consider this idea of explicitly passing "-j1" on the command line. In the third instance where you provide -j12, a new jobserver will be created with 12 tokens and the top-level make and all its sub-makes will share that jobserver, regardless of the value of -j added by the sub-make. So, the only time that the MAKEFLAGS="-j6 $$MAKEFLAGS" has a different behavior than MAKEFLAGS="$$MAKEFLAGS -j6" is in the case where someone invokes "make -j1" at the top level; in the first case this will prevent the sub-make from starting a jobserver, while in the second case it will still do that. Otherwise it has no effect. You can prove this to yourself with this simple makefile: $ cat Makefile .RECIPEPREFIX := > top: > @echo "$@: $$MAKEFLAGS" > $(MAKE) middle middle: > @echo "$@: $$MAKEFLAGS" > MAKEFLAGS="-j4 $$MAKEFLAGS" $(MAKE) bottom bottom: 1 2 3 4 > @echo "$@: $$MAKEFLAGS" 1 2 3 4: > @echo $@ start; sleep 2; echo $@ end First we run make with no -j at all (this is GNU make 4.3 FYI): $ make --no-print-directory top: --no-print-directory make middle middle: --no-print-directory MAKEFLAGS="-j4 $MAKEFLAGS" make bottom 1 start 2 start 3 start 4 start 1 end 2 end 3 end 4 end bottom: -j4 --jobserver-auth=3,4 --no-print-directory You can see from the output that no jobserver was available in the "middle" invocation, and that the "bottom" invocation was run with a jobserver with 4 slots available (all four jobs started in parallel). Now if we run with -j1: $ make --no-print-directory -j1 top: -j1 --no-print-directory make middle middle: -j1 --no-print-directory MAKEFLAGS="-j4 $MAKEFLAGS" make bottom 1 start 1 end 2 start 2 end 3 start 3 end 4 start 4 end bottom: -j1 --no-print-directory Now, the invocation of the jobserver in "middle" was ignored because the -j1 option on the command line overrode the -j4 in the "middle" rule. And we can see the "bottom" targets were all run serially, not in parallel. Finally, add -j2 to the parent: $ make --no-print-directory -j2 top: -j2 --jobserver-auth=3,4 --no-print-directory make middle middle: -j2 --jobserver-auth=3,4 --no-print-directory MAKEFLAGS="-j4 $MAKEFLAGS" make bottom 1 start 2 start 1 end 2 end 3 start 4 start 3 end 4 end bottom: -j2 --jobserver-auth=3,4 --no-print-directory Here you can see that the "-j4" option in "middle" was ignored. But if you reverse the MAKEFLAGS assignment: > MAKEFLAGS="$$MAKEFLAGS -j4" $(MAKE) -f /tmp/x8.mk bottom You'll get the same results: 1 start 2 start 1 end 3 start 2 end 4 start 3 end 4 end It still only runs with -j2 even though -j4 "should" take precedence.
