Some projects use cmake_autogen to generate moc/uic/rcc jobs.  The max
number of parallel jobs to run is AUTOGEN_PARALLEL[0], itself
initialized by CMAKE_AUTOGEN_PARALLEL.  By default it's empty, which
means cmake_autogen will start up to $ncpus jobs on the system.  Those
$ncpus are children of the cmake_autogen process and aren't accounted
in the global pool of ${MAKE_JOBS} jobs.  So if you build an affected
port with MAKE_JOBS=1 on a 16 cores machine, you end up with 1
cmake_autogen job + 16 moc/uic jobs = 17 jobs.  That's not nice as we
may have other ports building in parallel.

The diff below attempts to limit the max number of autogen jobs to
${MAKE_JOBS}, so that the max number of jobs running at any time
doesn't exceed 2 * ${MAKE_JOBS}.  Looks like there is no way to get
exactly ${MAKE_JOBS} jobs in total, but hopefully the behavior should
now be saner.

Issue spotted while looking at qgis.  Thoughts?  ok?

[0] https://cmake.org/cmake/help/latest/prop_tgt/AUTOGEN_PARALLEL.html
[1] https://cmake.org/cmake/help/latest/variable/CMAKE_AUTOGEN_PARALLEL.html


Index: cmake.port.mk
===================================================================
RCS file: /home/cvs/ports/devel/cmake/cmake.port.mk,v
diff -u -p -r1.84 cmake.port.mk
--- cmake.port.mk	29 Jul 2023 11:53:23 -0000	1.84
+++ cmake.port.mk	1 May 2024 18:49:32 -0000
@@ -13,6 +13,10 @@ CONFIGURE_ENV += MODCMAKE_USE_SHARED_LIB
 MAKE_ENV += MODCMAKE_USE_SHARED_LIBS=yes
 .endif
 
+# Limit the number of moc/uic processes started by cmake_autogen
+# (default: number of CPUs on the system)
+CONFIGURE_ARGS += -DCMAKE_AUTOGEN_PARALLEL=${MAKE_JOBS}
+
 USE_NINJA ?= Yes
 
 .if ${USE_NINJA:L} == "yes"


-- 
jca