In current Fedora Rawhide, the fprintd service fails to start after default Fedora Server package set install: https://openqa.fedoraproject.org/tests/744980 The failing service is fprintd. It fails on: Dec 21 04:48:31 localhost.localdomain systemd[1]: Starting Fingerprint Authentication Daemon... Dec 21 04:48:32 localhost.localdomain fprintd[793]: /usr/libexec/fprintd: error while loading shared libraries: libnss3.so: cannot open shared object file: No such file or directory Dec 21 04:48:32 localhost.localdomain systemd[1]: fprintd.service: Main process exited, code=exited, status=127/n/a examining the executable with ldd, it is linked against libnss3.so and libnssutil3.so , but the fprintd package does not depend on nss or nss-util. Nothing else in the default Server package set pulls in nss or nss-util, so the service startup fails. On other editions I think this bug is masked because nss and nss-util are pulled in via other dep chains. I'm not sure *exatly* how the executable comes to be linked against those libraries; I think the build flags get to include `-I/usr/include/nss3` because it's in the Requires(private) of libfprint and fprintd's src/meson.build and utils/meson.build both have `link_with: libfprintd_private`, but beyond that I'm not sure if this is the intended result or what the best fix should be. But as things stand it's certainly broken.
Proposing as a Final blocker per https://fedoraproject.org/wiki/Fedora_34_Final_Release_Criteria#System_services - "All system services present after installation with one of the release-blocking package sets must start properly, unless they require hardware which is not present."
Yeah, that seems odd. libfprint depends on nss and should pull it in, but fprintd should certainly not link directly against NSS.
Oh, hey, I know why this bug just appeared: it's because mstransky set Firefox to use internal nss because of https://bugzilla.redhat.com/show_bug.cgi?id=1908018 . Previously this bug was camouflaged by Firefox pulling in nss. Yes, even on Server - for some reason the default Server package set includes Firefox, that's odd, I'll have to try and figure out why. We're actually hitting this on Workstation too, because apparently Firefox was the only thing pulling nss into Workstation's default package set.
The other thing I didn't understand here is why no auto-generated dependency happens. Is it because these libs are unversioned, perhaps? libfprint doesn't depend on nss or nss-util either.
The dependency is actually there: https://koji.fedoraproject.org/koji/rpminfo?rpmID=24271679 The problem is that firefox provides libnss3.so()(64bit), but does not install the DSO in the directory searched by glibc: https://koji.fedoraproject.org/koji/rpminfo?rpmID=24364770 That's arguably a bug in the dependency generators: by default, they should only consider directories searched by default.
aha, yes, thanks Florian, I missed that. So the problem is still the change in Firefox, but for a slightly different reason. It's wrong (others have noted this already) that the Firefox build provides the libnss deps, since it's not intending its copy to be system-wide. The packaging guidelines say that such provides should be filtered out by the packager. There's already a bug report for that - https://bugzilla.redhat.com/show_bug.cgi?id=1908791 - so this ultimately can just be treated as a dupe of that, I guess. *** This bug has been marked as a duplicate of bug 1908791 ***
> The packaging guidelines say that such provides should be filtered out by the packager. Why is it better to rely on human factor to explicitly filter out non-applicable provides when the dependency generator script can be fixed not to produce them by default?
(In reply to Kamil Dudka from comment #7) > > The packaging guidelines say that such provides should be filtered out by the packager. > > Why is it better to rely on human factor to explicitly filter out > non-applicable provides when the dependency generator script can be fixed > not to produce them by default? I expect it will break a few packages which install shared objects in non-default paths, along with a configuration file in /etc/ld.so.conf.d. I don't quite understand the reason for that approach (rather than using /usr/lib64 directly).