Code review comment for lp:~sinzui/juju-release-tools/validate-streams

Hi Aaron. Thank you for the review. your question has raised a scenario I had not considered.

Retract is like a mode. We are removing a set instead of adding one. The version being published is likely to be the previous version. The version is always passed because Jerf will always be installing that version to create the streams. I assumed that the version is the previous good version.

I don't think your example is real. I cannot think of retracting a later version, and publishing an earlier non-existent version, but...

I can image waking up and find an email from Ian that says there is a security issue with version n, his team have prepared version q. We need to retract earlier n and since something is ready, publish later q. I would certainly want to do one publication, not two to achieve this. I am certain this test would fail because there are two expected differences.

old_tools = make_tools_data('trusty', 'amd64', ['1.20.7', '1.20.8'])
new_tools = make_tools_data('trusty', 'amd64', ['1.20.7', '1.20.9'])
check_expected_tools(old_tools, new_tools, '1.20.9', '1.20.8')

I am going to change the rules to accommodate this case. This is tricky I think the rule is that when doing a retractions, we need to ask if the version being published is new and need to be included in the expected differences.