search

rpm-software-management / librepo

A library providing C and Python (libcURL like) API for downloading packages and linux repository metadata in rpm-md format
http://rpm-software-management.github.io/librepo/
GNU Lesser General Public License v2.1
74 stars 91 forks source link

Use nanosec precision for timestamp of checksum cache (RhBug:2077864) #256

Closed kontura closed 2 years ago

kontura commented 2 years ago

= changelog = msg: Use nanosec precision for timestamp of checksum cache type: bugfix resolves: https://bugzilla.redhat.com/show_bug.cgi?id=2077864

librepo already uses #define _POSIX_C_SOURCE 200809L in librepo/handle.c and librepo/util.c so I think it should be fine from compatibility point of view.

m-blaha commented 2 years ago

Nice. Given that it is RHEL bug, do we need a test for it?

kontura commented 2 years ago

It seems there is some test that discovered this problem mentioned here: https://bugzilla.redhat.com/show_bug.cgi?id=2077864#c3

But I am not sure if that is deterministic and for this behavior specifically, I could create a dedicated CI test that loads the cached mtime and then changes the file mtime manually with something like: touch --date '(cached mtime + 1nanosec)' bash.rpm?

m-blaha commented 2 years ago

Yes, something like this should work. The current test is definitely not deterministic.

m-blaha commented 2 years ago

Why is the PR blocked? Is it because of missing / failing tests?

kontura commented 2 years ago

Why is the PR blocked? Is it because of missing / failing tests?

While creating the CI tests (https://github.com/rpm-software-management/ci-dnf-stack/pull/1096 which are also failing but I guess that is unrelated?) I noticed there is a failing unittest. It can be fixed but I am confused how come the build for the PR succeed. I want to look into it.

kontura commented 2 years ago

I was also considering whether it would be better to store the nanoseconds part of the timestamp separately under a different key, for example: user.Librepo.checksum.mtime_nsec but I am not sure. Nanoseconds epoch is a very big number.

kontura commented 2 years ago

Why is the PR blocked? Is it because of missing / failing tests?

While creating the CI tests (rpm-software-management/ci-dnf-stack#1096 which are also failing but I guess that is unrelated?) I noticed there is a failing unittest. It can be fixed but I am confused how come the build for the PR succeed. I want to look into it.

Oh I see, it is the problem with support of extended attributes. The test wasn't running (failing) because the build environment doesn't support them.