Building sys-libs/ncurses-5.9-r3 can result in "make: INTERNAL: Exiting with 1 jobserver tokens available; should be 3!" which can be worked around by de-parallelizing the build: cd ${EPREFIX} mkdir -p etc/portage/env echo 'MAKEOPTS="-j1"' > etc/portage/env/deparallelize echo '=sys-libs/ncurses-5.9-r3 deparallelize' >> etc/portage/package.env
Created attachment 382294 [details] build log
Created attachment 382296 [details] emerge info
Created attachment 382298 [details] emerge pqv
Doesn't look like a prefix problem to me.
Could be a make problem as well.
make[1]: *** read jobs pipe: No such file or directory. Stop. make[1]: *** Waiting for unfinished jobs.... Makefile:109: recipe for target 'all' failed Same error as in all your previous parallel make bugs. Please stop filing new ones. Something's wrong with your make or system.
Comment on attachment 382294 [details] build log FWIW: Similar problems were seen on Interix. For AIX, I had to mask gmake-4.0 for stability problems for now. But even on Linux sometimes I do see parallel build issues - which feel related to portage's stream-redirection to some degree... Also, there's this one as the first output of configure: configure: line 1109: 33371 Segmentation fault: 11 ( PATH=".;."; conftest.sh ) 2>&5
@Fabian and @Michael, so it looks like the theory that this is a make version 4 issue may be partially but not completely correct. Core i7 hw.ncpu: 8 MAKEOPTS="-j5" CFLAGS="${CFLAGS} -O2 -pipe" 1. qlop -l | awk '{print $NF;}' | sort | uniq | wc -l 453 2. I've needed to deparallelize app-portage/eix-0.30.1 dev-lang/ruby-2.1.1 net-misc/rsync-3.1.0 app-arch/libarchive-3.1.2-r1 dev-vcs/git-2.0.0-r2 dev-libs/gmp-5.1.3-r1 sys-devel/gettext-0.18.3.2 3. qlop -l | awk '{print $NF;}' | sort | uniq | while read atom; do for cp in app-portage/eix dev-lang/ruby net-misc/rsync app-arch/libarchive dev-vcs/git dev-libs/gmp ; do [[ "$atom" =~ .*"$cp".* ]] && echo "$atom" ; done ; done app-arch/libarchive-3.1.2-r1 app-portage/eix-0.30.1 app-portage/eix-0.30.2 dev-lang/ruby-1.9.3_p448-r1 dev-lang/ruby-1.9.3_p545 dev-lang/ruby-2.0.0_p247-r1 dev-lang/ruby-2.0.0_p451 dev-lang/ruby-2.1.1 dev-libs/gmp-5.1.3-r1 dev-vcs/git-1.9.1 dev-vcs/git-1.9.3 dev-vcs/git-2.0.0 dev-vcs/git-2.0.0-r1 dev-vcs/git-2.0.0-r2 dev-vcs/git-2.0.2 dev-vcs/git-2.0.4 net-misc/rsync-3.1.0 net-misc/rsync-3.1.0-r1 net-misc/rsync-3.1.1 So out of 453, that would be about 19 affected (about 4%), assuming every version of each package was affected. 4. I just downgraded to sys-devel/make-3.82-r4 from sys-devel/make-4.0-r1 and attempted to emerge gmp, rsync, gettext, ruby, libarchive, git, and eix. Results: dev-libs/gmp-5.1.3-r1 failed net-misc/rsync-3.1.1 succeeded app-arch/libarchive-3.1.2-r1 succeeded sys-devel/gettext-0.18.3.2 succeeded dev-vcs/git-2.0.4 succeeded app-portage/eix-0.30.2 succeeded dev-lang/ruby-2.1.1 failed 5. So it looks like make 4 may be the problem for rsync, libarchive, gettext, git, and six, and that gmp and ruby have parallelization issues unrelated to make 4. @Fabian, while it appears that you are correct that some of my problems have actually been make specific, almost a third of the problems were not specific to make, or at least not specific to make version 4. Though I suppose you cannot rule out that there may be some other make problem affecting both make version 3 and make version 4, or that there may be some problem affecting my particular system, it seems more likely that the gmp and ruby issues are indeed specific to those packages. Therefore, assuming that the downgrade to make 3.8 is sufficient to work around the make-specific issues, it seems I should actually continue reporting parallelization issues I encounter going forward, or do you disagree? 6. I suppose someone needs to create a bug report about the make 4 problem.
Further analysis seems to indicate the problem may have nothing to do with make versions. I tried build 5 times with make4 and 5 times with make3. In no case did rsync or git fail, so let's ignore those. (make4 failures, make3 failures) eix (3, 2) gettext (1, 1) gmp (2, 3) libarchive (4, 2) ruby (3, 2) total (13, 10) and of course (make4 successes, make3 successes) eix (2, 3) gettext (4, 4) gmp (3, 2) libarchive (1, 3) ruby (2, 3) total (12, 15) make4 13 failures, 12 successes make3 10 failures, 15 successes Note that no package shows 0 failures with make3. So not sure where this leaves us.
(In reply to iloveosxzfs from comment #9) > make4 13 failures, 12 successes > make3 10 failures, 15 successes Does it make a difference with make-3.81 eventually? AFAIR, there's one problem with Solaris and NFS we don't carry the patch for, as it is fixed in 3.82 (https://savannah.gnu.org/bugs/?18680).
I was comparing sys-devel/make-3.82-r4 and sys-devel/make-4.0-r1
is this still an issue w/make-4.1 ?
we've a bit further now, haven't been able to reproduce the problem