DNF5 prototype #5847
DNF5 prototype #5847
Conversation
Use DNF5 in the Anaconda system installer for the package installation.
The option getters were renamed in dnf5 in commit: rpm-software-management/dnf5/commit/740fd279d283156b8e9c3525a16142f2fffda291 There is also another option and that is to change the DNFConfigWrapper. However, this class is described as a temporary wrapper, so further investigation will be needed around this. For now, renaming the options seems sufficient.
The dnf5 methods `Repo::fetch_metadata` and `Repo::load` were both removed from API. See: rpm-software-management/dnf5/commit/cc2ae250c30bdbd8fadb99a4eb9d817b074dc6b2 rpm-software-management/dnf5/commit/b2ece236b62d121c4e2ddd91272804db17ee7d12 Instead use `RepoSack::load_repos` to load all repositories. This method must be called only once.
The method was removed by commit: 0ff5e44
The test is adjusted because the base.setup() was already called. This might need to change if the place of calling base.setup() changes.
The method was removed by commit: rpm-software-management/dnf5@bfb6f32
There was a change regarding creation of the `ValidationReport`. It's no longer created in `CheckPackagesSelectionTask._resolve_selection()` but in `DNFManager.resolve_selection()`. The change lookgs good and it was probably even necessary because of changes in dnf reports from the transaction resolving, but it prevents testing the report messages, because the report is now also mocked. Since it's not a critical problem, I am disabling the message checks for now. The changes that caused the issue: poncovka@faccc47#diff-08489d103d5780dd5279cb390820b5eeb83963642783cd82773cef081803b1b7L620-L641 poncovka@faccc47#diff-57dffd56427be4eca900ee75e0a130414fb51527d206b00116ef9501f0784127L96-L114
The comps objects are owned by the queries, so when a query is destroyed, so are the contained objects. There is a bug for dnf: rpm-software-management/dnf5#1530 This is a temporary workaround.
The ValidationReport is now created within the mocked DNFManager.resolve_selection, so it must be mocked as well. Also, there are no longer exceptions raised from the DNF transaction, instead, the problems are stored in a report, so there are no exceptions to mock.
The base is no longer being closed and the resetting of base is tested in test_module_payload_dnf5_manager.DNFManagerTestCase.test_reset_base.
These tests probably didn't work at the time they were marked as skipped, but are working now.
Type 'void *' is expected as return from `add_new_download` and as `user_data` and `used_cb_data` arguments, so it cannot be easily used in Python. This is a workaround that fixes a segmentation fault, but a better solution is still needed.
The transaction can fail even if it doesn't contain any transaction items with an error status.
pkratoch
temporarily deployed
to
gh-cockpituous
GitHub Actions
Inactive
|
Hello @pkratoch! Thanks for opening this PR. We checked the lines you've touched for PEP 8 issues, and found:
|
No description provided.