This is a small follow up fix related to 01be34a706.
Previously the script modified a list while iterating through it, which
is always a bad idea, but it didn't throw an error. My last changes
switched it to iterating through a list, which surfaced the flaw as a
RuntimeError. Using a set comprehension is a cleaner way to generate
the set we need to return.
Signed-off-by: Carl George <carlwgeorge@gmail.com>
Originally attempted in 01be34a706, but it
didn't work correctly and needed to be reverted. Some notable
differences this time around:
- Previously the script predicted branch and version based on the tag.
I had this switched to get both of those from the bodhi release, but not
all tags have a corresponding bodhi release. So this now takes a hybrid
approach of using bodhi release if available, and falling back to the
prediction approach otherwise.
- Retired package lists from the lookaside are now skipped if they don't
exist, both during the retrieval and when checked later.
- The get_project_branches function now returns a set.
Signed-off-by: Carl George <carlwgeorge@gmail.com>
The new epel10 release in bodhi is confusing this script, so
lets exclude it for now until we get the script sorted out.
Signed-off-by: Kevin Fenzi <kevin@scrye.com>
We had the syntax wrong here, it's space seperated, not ,
so the entire thing was getting checked and of course failing.
Lets try again now with space seperator
Signed-off-by: Kevin Fenzi <kevin@scrye.com>
We changed around compression as part of a Fedora change and move to
createrepo_c > 1.0, but the epel8 repos are causing a bit of problem for
users who use mergerepo_c on them from older (<1.0) createrepo_c.
They can't understand the primary.xml being zstd compressed.
So, lets try and set this to switch primary back to xz.
Signed-off-by: Kevin Fenzi <kevin@scrye.com>
You can't nest jinja's templating markers like this, it turns
out, they just wind up as literal strings. You also can't use
f-strings, sadly (tried it locally, it errors out). We have to
use older-style `.format()`, that works. I went with a fairly
explicit syntax to make it clear what's going on.
Signed-off-by: Adam Williamson <awilliam@redhat.com>
Our compose attempt with the containers being built under
Everything didn't go as planned because the filenames of the
container images turn out to be determined by the variant, so
they conflicted and only the last one built survived. So, we
really need to build them under unique variants for now. A
previous commit added the variant definitions.
We don't *need* to build the ostrees under the variants, but it
feels more consistent this way, both with the containers and
with how it works in the Rawhide/Branched composes.
Signed-off-by: Adam Williamson <awilliam@redhat.com>
We repeat this exact same repo definition block eight times,
let's just make it a variable (at the ansible template level).
Signed-off-by: Adam Williamson <awilliam@redhat.com>
bodhi is failing to finish a f40-updates-testing push because it's
looking for the ppc64le repomd on the primary location instead of the
secondary one.
This should fix it, but I don't understand how it worked before.
Did we just get lucky and always check aarch64 before anything else?
I can't see anything that changed in config around this off hand. ;(
Will try this and see if it fixes it for now...
Signed-off-by: Kevin Fenzi <kevin@scrye.com>
As is currently the case for the ostrees themselves. The variants
don't exist in this compose config.
Signed-off-by: Adam Williamson <awilliam@redhat.com>
Since Fedora 40, we've built native OCI containers for the atomic
desktops in the pre-release nightly and release composes. We
should also build native containers in the post-release nightly
composes that Bodhi runs. This should allow us to get rid of the
on-the-fly OCI image generation in the sync-ostree-base-containers
script.
Signed-off-by: Adam Williamson <awilliam@redhat.com>
