verdi code delete falsely claims it actually deleted a node

[sphuber]

It claims that the code is deleted, but it is not actually the case. Seems that the problem is with the PsqlDosBackend.delete_nodes_and_connections that does perform the deletion but changes are not persisted. This is most likely related to recent refactorings of the PsqlDosBackend.transaction method.

[sphuber]

There is actually a test for this and it passes. I suspect the reason is that the test doesn't actually go exactly through the same route as when calling verdi code delete but it uses the click.TestRunner. This has bitten us in the ass before as these two pathways are not exactly identical. We should really have an alternative for run_cli_command to run through subprocess to simulate the actual pathway that a user would experience.

[sphuber]

The scope of this problem is way bigger than thought. The problem is that changes to the session in the PsqlDosBackend are no longer committed correctly. So essentially most (if not all) verdi commands that mutate state (create groups, delete nodes, etc.) do not persist the changes. The same goes for an interactive shell. If you run Group(label='test').store() in a verdi shell and then exit, the group won't exist.

[sphuber]

This is fixed in #5851 by reverting the offending commits and will be released with v2.2.1