https://blogs.gentoo.org/ago/2020/07/04/gentoo-tinderbox/ Issue: app-editors/vim-core-<unset> installs more files with FEATURES=test. Discovered on: amd64 (internal ref: ci) NOTE: To reproduce this issue: 1) Install the package with FEATURES=test 2) Put the list of the installed files into a file ( qlist -e PACKAGE:SLOT | sort | uniq > /tmp/qlist_test ) 3) Install the package without FEATURES=test 4) Put the list of the installed files into a file ( qlist -e PACKAGE:SLOT | sort | uniq > /tmp/qlist_notest ) 5) See the diff between the two files ( diff -ru /tmp/qlist_test /tmp/qlist_notest ) On this machine: - the installed files with FEATURES=test are 2153 - the installed files without FEATURES=test are 2149 This type of report comes from a check that is a bit outside of the tinderbox, that's the reason because build.log was not provided (however in cases like these the build.log doesn't help much) but if you need further help to diagnose this issue feel free to ask. In any case I'm providing a diff.
DIFF: --- /dev/fd/63 2023-08-07 07:56:00.141974138 +0200 +++ /dev/fd/62 2023-08-07 07:56:00.141974138 +0200 @@ -103,6 +103,10 @@ /usr/share/man/ja/man1/vimtutor.1.bz2 /usr/share/man/ja/man1/xxd.1.bz2 /usr/share/man/man1/evim.1.bz2 +/usr/share/man/man1/ex.1.bz2 +/usr/share/man/man1/rview.1.bz2 +/usr/share/man/man1/rvim.1.bz2 +/usr/share/man/man1/view.1.bz2 /usr/share/man/man1/vim.1.bz2 /usr/share/man/man1/vimdiff.1.bz2 /usr/share/man/man1/vimtutor.1.bz2
That sounds rather odd considering there's no USE=test and: src_test() { :; } aka FEATURES=test should be a no-op, I can't reproduce anyway.
(In reply to Ionen Wolkens from comment #2) > That sounds rather odd considering there's no USE=test and: > > src_test() { :; } > > aka FEATURES=test should be a no-op, I can't reproduce anyway. I looked deeply into it and, atm, I got always 2149 file installed with and without FEATURES="test" In any case, when I launched a build with FEATURES="test" (while test is completely unrelated), 'ci' collected 2153 files and compared it with the subsequently build (that registered 2149 files) there was a difference. Since there are no test, I suppose that +/usr/share/man/man1/ex.1.bz2 +/usr/share/man/man1/rview.1.bz2 +/usr/share/man/man1/rvim.1.bz2 +/usr/share/man/man1/view.1.bz2 are installed randomly?
found the trick! with MAKEOPTS="-j1" the files installed are 2153, so with higher -j value there is a race condition that do not install the manpages reported in the diff.
(In reply to Agostino Sarubbo from comment #3) > Since there are no test, I suppose that > > +/usr/share/man/man1/ex.1.bz2 > +/usr/share/man/man1/rview.1.bz2 > +/usr/share/man/man1/rvim.1.bz2 > +/usr/share/man/man1/view.1.bz2 > > are installed randomly? Possibly noteworthy is that these are all symlinks to (the non-translated) vim.1.bz2. (In reply to Agostino Sarubbo from comment #4) > found the trick! > > with MAKEOPTS="-j1" the files installed are 2153, so with higher -j value > there is a race condition that do not install the manpages reported in the > diff. I can repro it when running `ebuild ... clean install` and comparing listings with `find /var/tmp/.../image -type f -o -type l | sed -e 's|^/var/.*-9.0.1677/image||'` with/without MAKEOPTS='j1' prepended to it. What's even more interesting is that I most definitely have those symlinks in my root, yet they don't show up in `qlist -e vim-core`.
