TranslationBranchApprover trips over deleted templates.

[Henning Eggers]

The TranslationBranchApprover goes into great length comparing what templates are in the tree to what templates are in the database. if a template in the database is not found in the tree but other templates are new in the tree, the approver assumes that templates were renamed and does not approve. It is assuming too much here as the template might as well be deleted. This has resulted in lots of "Needs review" templates which could just have been created, putting more load on the manual approval.

I suggest the following simple (and obvious) behaviour:
- If a template name from the tree (deducted from path) is found in the tree, update the path in the database entry and approve the queue entry.
- If a template name is not found in the database, create a new entry.
- Do not care about the other way round (database entries not being found in the tree).

If the uploader renamed a template, this will result in a new template being created besides the old one. With the new ability to deactivate templates, the maintainer can rectify this manually.

Problems will arise with projects that already have templates that were imported from tarball and the human approver did not create the template name from the template path in the same way the branch approver does. When the project switches to imports from branch, this will lead to new templates being created without the template having been renamed. But I think we can fix these interactively with the maintainer.

[Robert Collins (lifeless)]

There is a separate bug about making template imports use bzr deltas rather than full tree; doing that would solve this implicitly. However for now, this specific bug isn't part of the LP roadmap, though we would happily accept patches.