Add typing_extensions.get_annotations

[JelleZijlstra]

Fixes #409

This is based on the implementation in python/cpython#119891. The version that eventually lands in CPython may look a bit different, but should be compatible with what's in this PR.

[AlexWaygood]

In #151, the consensus seemed to be that we shouldn't add new "data files" to the src directory for usage in tests, because these will end up in the site-packages directory for people who've installed typing_extensions. Should we just store these as strings in test_typing_extensions.py, like we did for the data files in that PR?

[srittau]

It might also be time to split the test file, as it's approaching 10k lines.

[AlexWaygood]

Maybe we could have an src/test_typing_extensons directory: then it's only a single importable module in site-packages, but we'd be able to split it up so that it doesn't get too huge.

[JelleZijlstra]

I like the idea of making a test directory. I'll send a separate PR for that.

[JelleZijlstra]

I tried this and I think it's not worth it. I like to simply run python test_typing_extensions.py to run the test suite; that would no longer work with a package. I'll move the test files from this PR into exec() calls instead.

[AlexWaygood]

Hmm, this isn't going to work for type parameters, which exist in neither the global nor local namespaces. Possibly the fix I made to the typing.py internals in python/cpython@1e3e7ce should also have been applied to inspect.py?

[JelleZijlstra]

Yes. In my PEP 649 implementation branch (python/cpython#119891) I handle this by getting obj's __type_params__, but this may also be worth backporting separately.

[AlexWaygood]

Filed python/cpython#120270

[AlexWaygood]

any idea why inspect.get_annotations doesn't use inspect.unwrap() here? Though I suppose we won't be able to use inspect.unwrap in the CPython version anyway after moving the implementation out of inspect.py, since we won't want the new annotations module to depend on inspect

[JelleZijlstra]

Not sure. Looks like unwrap() doesn't handle functools.partial specially, though unlike this code it deals with cycles.

[AlexWaygood]

It's pretty odd; no CPython tests fail if I make this change:

diff --git a/Lib/inspect.py b/Lib/inspect.py
index 2b7f8bec482..4f9c43d9fc7 100644
--- a/Lib/inspect.py
+++ b/Lib/inspect.py
@@ -235,13 +235,11 @@ def get_annotations(obj, *, globals=None, locals=None, eval_str=False):
             if module:
                 obj_globals = getattr(module, '__dict__', None)
         obj_locals = dict(vars(obj))
-        unwrap = obj
     elif isinstance(obj, types.ModuleType):
         # module
         ann = getattr(obj, '__annotations__', None)
         obj_globals = getattr(obj, '__dict__')
         obj_locals = None
-        unwrap = None
     elif callable(obj):
         # this includes types.Function, types.BuiltinFunctionType,
         # types.BuiltinMethodType, functools.partial, functools.singledispatch,
@@ -249,7 +247,6 @@ def get_annotations(obj, *, globals=None, locals=None, eval_str=False):
         ann = getattr(obj, '__annotations__', None)
         obj_globals = getattr(obj, '__globals__', None)
         obj_locals = None
-        unwrap = obj
     else:
         raise TypeError(f"{obj!r} is not a module, class, or callable.")
 
@@ -265,17 +262,8 @@ def get_annotations(obj, *, globals=None, locals=None, eval_str=False):
     if not eval_str:
         return dict(ann)
 
-    if unwrap is not None:
-        while True:
-            if hasattr(unwrap, '__wrapped__'):
-                unwrap = unwrap.__wrapped__
-                continue
-            if isinstance(unwrap, functools.partial):
-                unwrap = unwrap.func
-                continue
-            break
-        if hasattr(unwrap, "__globals__"):
-            obj_globals = unwrap.__globals__
+    if not isinstance(obj, types.ModuleType):
+        obj_globals = getattr(unwrap(obj), "__globals__", None) or obj_globals

You could consider that simplification here since we already import inspect in typing_extensions, though it obviously can't be applied to CPython

[JelleZijlstra]

That sounds like we're missing test coverage for functools.partial. I think we should mirror the CPython implementation as much as possible.

[AlexWaygood]

(https://docs.python.org/3/howto/enum.html#comparisons)

[AlexWaygood]

LGTM. #423 (comment) can be fixed in a followup once we've sorted it out at cpython