Add real-world case-based tests #33

New issue

Open

opened 2026-04-18 19:17:32 +00:00 by adamwill · 0 comments

adamwill commented

2026-04-18 19:17:32 +00:00

Owner

Of course, they're quite large. So an idea I had is to resurrect the lxml (or pygixml?) based parsing as a standalone script to minify real-world metadata: parse it and rip out bits that don't matter, like the summary, description, packager, url, time, size, location, license, vendor, group, buildhost etc etc. If dnf breaks without some we can replace their values with empty strings at least. It could also allow you to filter out chunks of packages - perhaps by initial letter, so you can just leave out packages that are irrelevant to the scenario. This should allow us to reduce the sizes quite substantially, I hope, so we can keep a few real-world scenarios in the repo without overly bloating it. Possibly we could use a separate repo for this, though.

The synthetic tests are OK, but pretty limited and hard to extend to cover more cases. I had an idea that we can just add real-world based tests as well: we can literally just download the real-world metadata for specific interesting results (the base repo metadata, and the metadata from the tested package repo) and use those as test cases. Of course, they're quite large. So an idea I had is to resurrect the lxml (or pygixml?) based parsing as a standalone script to minify real-world metadata: parse it and rip out bits that don't matter, like the summary, description, packager, url, time, size, location, license, vendor, group, buildhost etc etc. If dnf breaks without some we can replace their values with empty strings at least. It could also allow you to filter out chunks of packages - perhaps by initial letter, so you can just leave out packages that are irrelevant to the scenario. This should allow us to reduce the sizes quite substantially, I hope, so we can keep a few real-world scenarios in the repo without overly bloating it. Possibly we could use a separate repo for this, though.