run_processor / get_processor: do not pass (empty) ocrd_tool

[bertsky]

Follow-up to premature #999: there are more places where we pass an empty ocrd_tool around that will never be used.

Reasoning (from ocrd_tesserocr discussion):

That's despite ocrd==2.47.1. Where does the empty ocrd_tool come from???

Aha! It turns out that my #999 was premature: we also pass an unnecessary ocrd_tool in other places:

• run_processor kwarg:

  core/ocrd/ocrd/processor/helpers.py

  Line 38 in cff8dda

  ocrd_tool=None,

• run_processor calling get_processor:

  core/ocrd/ocrd/processor/helpers.py

  Line 91 in cff8dda

  ocrd_tool=ocrd_tool,

• get_processor kwarg:

  core/ocrd/ocrd/processor/helpers.py

  Line 306 in cff8dda

  ocrd_tool: dict = None,

• get_processor calling constructor:

  core/ocrd/ocrd/processor/helpers.py

  Line 325 in cff8dda

  ocrd_tool=ocrd_tool,

Now, the big story here is: none of these places is needed (or used) – neither currently, nor for #974 nor #884!

It's also clear now, why we never used this: you need runtime information (from the CLI) for this. Either we are already on the CLI (Processor CLI decorator in all processor modules) or we have to make foreign CLI calls and parse the JSON response from the concrete Processor subclasses (which don't live in our venv, or we don't know the class name of, or are bashlib anyway).

[bertsky]

Now also updated the tests to not pass on ocrd_tool for no reason.

BTW, we should try to cover cached processor instances in the processor tests, too.

[bertsky]

@kba as this modifies existing API args, will necessitate a new minor – 2.48. I have scanned all ocrd_all code and have found no other use cases throwing ocrd_tool kwarg at run_processor other than in core's own regression tests. It may be that outside repos do this, though.

[bertsky]

@kba please prioritise. If you find any arguments against this, I can still try to find a different solution in ocrd_tesserocr, but as it stands, I don't see a reason. And since after #972, ocrd_tesserocr stopped working, this is blocking the next ocrd_all release.

[joschrew]

My thoughts behind my approval: I see this PR as kind of reminder to not provide ocrd_tool=None in
a processor call (or in a run_processor invocation). This must be kept in mind / ensured for the
PR 972 and for other possible changes.

[bertsky]

@kba for the v2.48 changelog:

Changed

• 🔥 ocrd.run_processor / ocrd.processor.get_processor: rm unnecessary ocrd_tool kwarg ocrd_cli_wrap_processor: drop passing ocrd_tool=None #998