beangulp-style hook detection is very fragile

[alarsyo]

I have just spent the last hour fighting with Fava to get it to correctly pass the arguments my hooks expect:

Fava detects if the user wants beangulp-style hooks by looking at the type annotations, however there are several problems with the current implementation:

1. get_annotations(hook_fn) will only work for functions, however one of my hooks is actually a callable object:

   class MyHook():
       def __init__(self, regex_list):
           self.regex_list = regex_list
   
       def __call__(self, extracted, existing) -> list[tuple[str, data.Entries, str, Importer]]:
           pass

   This case is easily fixable by also checking get_annotations(hook_fn.__call__), however I'm not sure that this covers all possible cases.

   For now, an easy workaround on my side is to wrap my callable object with a function with the correct return type annotation.

2. The current code assumes that annotations are stringly-typed (uses "Importer" in annotation as a test), however this is not the case in all instances: I had to add from __future__ import annotations to my file for this test to work. Otherwise the test fails even though the type annotations are present. I guess this could be fixed with "Importer" in str(annotation), but that doesn't feel great.

I can submit a PR if you think the fixes I mentioned are good enough, but the overall approach feels fragile to me, and could break in other scenarios as well. I don't have a better solution though 😅

[yagebu]

I agree that it's a fragile approach. I see this mainly as a compatibility layer to ease the migration from Beancount 2 to beangulp and mid-term we should get rid of the "old style" of hooks altogether.