If I define a class within a class, like this:

class A:
    class B:
        pass
    
    def foo(self):
        pass

The class browser shows that A contains foo, but it doesn't show B at all.

Confirmed on trunk and py3k.

The class browser relies on the pyclbr module to scan the code. This
module doesn't support classes within classes. Both pyclbr and IDLE's
class browser need to be modified.

The attached patch adds support for nested objects in pyclbr and also in
ClassBrowser.

I have yet to find an optimal way to test this on test_pyclbr (I did
some things here, none very nice), so you will notice tests are missing
on this patch.

I've removed the previous patch from here and attached an updated one on
issue6691.

Now the patch attached here only contains changes related to
idlelib.ClassBrowser.

Class Browser is now a module browser as it includes functions not in classes. I guess pyclbr does this. Enhancing this for current versions would require adding a private copy of enhanced pyclbr to idlelib.

I believe the patch for #6691 will also add detection of nested functions.  If  so, the IDLE patch must also make use of that new information.

If the patch for #6691 is only applied to 3.7, I want to put a temporary copy of the revised pyclbr in 3.6 idlelib, perhaps as _pyclbr, and change the 3.6 import accordingly.

I created issue 30881 to add docstrings to browser.py to make creating user tests easier.

I added a more complete set of unit tests to the PR.

Nick, this over-due improvement to IDLE depends on the over-due improvement to pyclbr in 3.7.  I would like to backport the IDLE improvement by putting a copy of 3.7 pyclbr in 3.6 idlelib.  Do you agree that PEP434 allows this?

Would it make any difference if the module were renamed? and made to only work with IDLE?

I think a bundled copy as idlelib._pyclbr in the 3.6 branch would be within the intent of PEP 434.

I am revising the codecontext patch to use comprehensions and the sort key option.  I will rename the new function 'collect_objects' (GPolo) / '_traverse_node' (CSabella) as 'list_children' and have it return the name-object list, ordered by line number, needed by the tree widget.

[The input is a dictionary of children of a node, rather than a node, because pyclbr returns the dictionary of the children of the root module node of the tree, rather than a root node itself.  Sorting is not actually needed for 3.6/7 because the dicts are already ordered, but pyclbr does not guarantee this by using OrderedDict.  Proposing these two changes is a separate potential issue.]

New changeset 058de11360ea6816a6e978c7be0bcbea99a3f7da by Terry Jan Reedy (Cheryl Sabella) in branch 'master':
bpo-1612262: IDLE: Class Browser shows nested functions, classes (#2573)
https://github.com/python/cpython/commit/058de11360ea6816a6e978c7be0bcbea99a3f7da

New changeset fa1cae5832cbcfafedc4b1879c2abc85452f4edd by Terry Jan Reedy in branch '3.6':
 [3.6] bpo-1612262: IDLE: Class Browser shows nested functions, classes (GH-2573)  (#3702)
https://github.com/python/cpython/commit/fa1cae5832cbcfafedc4b1879c2abc85452f4edd

I settled on 'transform_children' for the new function. #31461 is the master issue for improving the browsers.