get source_file of wrapped functions

[tmeyier]

I added a function get_source_file to doc_ast.py that tries to get the source file from obj.wrapped if inspect.getsourcefile(obj) doesn't succeed.
I used the same cached approach as in doc_ast.get_source.

closes #656

[tmeyier]

As an alternative, I thought about unwrapping cached functions in Function.__init__. But I wasn't sure about possible side effects and I found it a bit creepy to test for a cached function with

if isinstance(func, functools._lru_cache_wrapper):

[tmeyier]

I haven't implemented test coverage for the two new methods in doc_ast yet.

What kind of test would you like to have?

• unit tests inside test/test_doc_ast.py
• an integration test with a custom template using the function source_file (I have one that gives the source_file in the caption of the source code box)?

[mhils]

As an alternative, I thought about unwrapping cached functions in Function.__init__. But I wasn't sure about possible side effects and I found it a bit creepy to test for a cached function with

if isinstance(func, functools._lru_cache_wrapper):

Does that fix source extraction? That seems like a reasonable approach here. We heavily depend on Python internals anyhow. If we're concerned about this being an internal API, we can also try the import in _compat and fall back to a placeholder class.

[tmeyier]

I removed the cached version of get_source_file from doc_ast.py and cached functions are unwrapped like this, now:

        elif hasattr(func, "__wrapped__"):
            unwrapped = func.__wrapped__

According to the source code of functools, wrapped is part of the public API of functools.lrucache, so that should be reasonably safe and every function decorator using @wraps should be covered.

[mhils]

Thanks, this looks great now. Can we get a test + changelog entry for this? :)

[mhils]

Thanks for the super quick turnaround! 🍰