4.16: no longer possible to package invalid symlink #1159
Comments
You can just ship a tarball of this stuff instead right? |
|
I'm sure we could find a workaround. It will be challenging because the same tests will need to work in CI as well as rpm-packaged, root and non-root. I will of course do that work if necessary. I filed this issue in hopes that the rpmbuild change was inadvertent/unintentional and could be reverted. |
|
Le mer. 1 avr. 2020 à 21:54, Ed Santiago <notifications@github.com> a
écrit :
We have a package, buildah <https://src.fedoraproject.org/rpms/buildah>,
that includes a symlink loop
<https://github.com/containers/buildah/blob/master/tests/bud/dockerignore2/subdir/circular-link>
as part of its -tests subpackage:
lrwxrwxrwx. 1 esm esm 13 Oct 21 05:53 circular-link -> circular-link
This has packaged up just fine, up to and including
rpm-4.15.1-2.fc32.1.x86_64
As of rpm-4.15.90-0.git14971.1.fc33.x86_64, rpmbuild fails with:
RPM build errors:
Recognition of file "/root/rpmbuild/BUILDROOT/buildah-1.15.0-0.22.dev.git912ca5a.fc33.x86_64/usr/share/buildah/test/system/bud/dockerignore2/subdir/circular-link" failed: mode 120777 broken symbolic link to circular-link (Too many levels of symbolic links)
My lame guess is that c6ff614
<c6ff614>
is responsible.
I realize this is an obscure corner case and that in fact it may be
desirable in most cases to catch this condition. This is, however, a
regression and places us in a difficult situation wrt packaging our tests.
—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHub
<#1159>, or
unsubscribe
<https://github.com/notifications/unsubscribe-auth/AAXGVEKVHJ5PNUEZKIFDU63RKOLXXANCNFSM4LZQBLHQ>
.
You can always package it as %ghost and actually create it in %post …
|
|
Sorry but benefits catching what is broken by definition at build time outweighs the special case needs of any single project, by far. The fact that its only caught the dependency generator is mostly a hint that we're lacking sanity checks someplace else. |
|
I suppose this does deserve a mention in the release notes though - will add. |
edsantiago
added a commit
to edsantiago/buildah
that referenced
this issue
Apr 2, 2020
rpm 4.16 alpha in rawhide has broken our koji build: containers#2264 rpm-software-management/rpm#1159 TL;DR we can no longer package the 'circular-symlink' file used in tests. This PR removes that and other symlinks from the git working tree, and reworks the test so it runs from a temporary directory and creates all necessary symlinks. Also, since the test as written was simply checking "did it run" without actually testing anything, add a test to confirm that the files added to the image are exactly what we expect. Signed-off-by: Ed Santiago <santiago@redhat.com>
edsantiago
added a commit
to edsantiago/buildah
that referenced
this issue
Apr 2, 2020
rpm 4.16 alpha in rawhide has broken our koji build: containers#2264 rpm-software-management/rpm#1159 TL;DR we can no longer package the 'circular-symlink' file used in tests. This PR removes that and other symlinks from the git working tree, and reworks the test so it runs from a temporary directory and creates all necessary symlinks. Also, since the test as written was simply checking "did it run" without actually testing anything, add a test to confirm that the files added to the image are exactly what we expect. Signed-off-by: Ed Santiago <santiago@redhat.com>
bors bot
added a commit
to containers/buildah
that referenced
this issue
Apr 2, 2020
2267: Bump 1.14.6 r=rhatdan a=TomSweeneyRedHat As the title says 2268: dockerignore tests : remove symlinks, rework r=rhatdan a=edsantiago rpm 4.16 alpha in rawhide has broken our koji build: #2264 rpm-software-management/rpm#1159 TL;DR we can no longer package the 'circular-symlink' file used in tests. This PR removes that and other symlinks from the git working tree, and reworks the test so it runs from a temporary directory and creates all necessary symlinks. Also, since the test as written was simply checking "did it run" without actually testing anything, add a test to confirm that the files added to the image are exactly what we expect. Signed-off-by: Ed Santiago <santiago@redhat.com> <!-- Thanks for sending a pull request! Please make sure you've read and understood our contributing guidelines (https://github.com/containers/buildah/blob/master/CONTRIBUTING.md) as well as ensuring that all your commits are signed with `git commit -s`. --> #### What type of PR is this? /kind other #### What this PR does / why we need it: Work around behavior change in rpmbuild #### How to verify it buildah will build in koji on rawhide (it currently does not) #### Special notes for your reviewer: #### Does this PR introduce a user-facing change? ```release-note None ``` Co-authored-by: TomSweeneyRedHat <tsweeney@redhat.com> Co-authored-by: Ed Santiago <santiago@redhat.com>
|
Comment added: rpm-software-management/rpm-web@1fddfd1 Closing, thanks for the report. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
We have a package, buildah, that includes a symlink loop as part of its
-testssubpackage:This has packaged up just fine, up to and including rpm-4.15.1-2.fc32.1.x86_64
As of rpm-4.15.90-0.git14971.1.fc33.x86_64, rpmbuild fails with:
My lame guess is that c6ff614 is responsible.
I realize this is an obscure corner case and that in fact it may be desirable in most cases to catch this condition. This is, however, a regression and places us in a difficult situation wrt packaging our tests.
The text was updated successfully, but these errors were encountered: