sys-libs/talloc-2.0.1-r1 did not create the soname symlink 'libtalloc.so.2', and so a QA notice was fired and and that file was created by portage duing the qmerge process, via the function '_post_src_install_soname_symlinks()' located in /usr/$(libdir)/portage/pym/portage/package/ebuild/dobuild.py sys-libs/talloc-2.0.1-r2 fixes the QA issue by creating the symlink during src_install, but at qmerge time the package reports a collision with the already-installed file /usr/$(libdir)/libtalloc.so.2 (the one auto-created when emerging -r1). Running equery b libtalloc.so.2 _does_ report it as being owned by talloc-2.0.1-r1, but for some reason this collision is still reported. I do not know the operational details of the collision check (nor would I understand them without significant study) but it would seem to me that the '_post_src_install_soname_symlinks()' function does not register the files it creates in a way that is compatible with the collisions check. This was tested on a stable x86 and a stable amd64 system; the issue seems to be independent of profile or arch. Steps to reproduce: emerge -1 =sys-libs/talloc-2.0.1-r1 && \ emerge -1 =sys-libs/talloc-2.0.1-r2
(In reply to comment #0) > Running equery b libtalloc.so.2 _does_ report it as being owned by > talloc-2.0.1-r1, but for some reason this collision is still reported. Maybe equery de-references symlinks, while portage does not. You can use portageq to see if portage thinks it belongs: portageq owners / /usr/lib/libtalloc.so.2 > I do > not know the operational details of the collision check (nor would I understand > them without significant study) but it would seem to me that the > '_post_src_install_soname_symlinks()' function does not register the files it > creates in a way that is compatible with the collisions check. Since portage-2.1.10.41, a warning is issued, but no symlink is created (see bug 387053). So, I think that there really is a collision, and that equery b is misleading because it's de-referencing the symlink. Note that the collision should not cause emerge to bail out unless FEATURES=collision-protect is enabled, which is not the default behavior. By default, FEATURES=protect-owned is enabled, and it will only warn about this case.
.. I cannot reproduce the case where libtalloc.so.2 is reported as owned by anything with 'equery b'. I guess the real issue is that this QA check on older portage's did not register the file it created as part of the package, but there isn't anything that can be done about it now (from the portage point of view, at least) so the point is moot. Closing bug as RESO/INVA
