Why is this an issue class B, def __init__, self, z?

Also for the other issues you found, please describe them clearer and post a traceback if there are exceptions. It's easier to fix them this way. Thanks!

Sorry for the late reply.

1. The issue is that defined_names doesn't list z. Although, __init__ is considered a parent.
   this is a problem when trying to use jedi for building hierarchical tree of symbols in document for LSP. To get z in symbols list I need to get all symbols and then build the tree from flat list for bottom up. Which is inconvenient.

2. About the traceback:

Traceback (most recent call last):
  File "file.py", line 31, in <module>
    print(all_names[1].defined_names())
  File "/home/m-panarin/.virtualenvs/pyls/lib/python3.7/site-packages/jedi/api/classes.py", line 376, in __repr__
    self.description,
  File "/home/m-panarin/.virtualenvs/pyls/lib/python3.7/site-packages/jedi/api/classes.py", line 551, in description
    return typ + ' ' + self._name.to_string()
AttributeError: 'TreeNameDefinition' object has no attribute 'to_string'

Inferior Python exited abnormally with code 1 at Sun Oct  6 14:00:28

You can reproduce it with the snippet above if you uncomment the last line.

I seem to have fixed the traceback. The other issue still exists.

Thanks again for the reproduction steps. That really helps a ton. I just tried to reproduce 3 other issues that I couldn't, which is a bit frustrating.

I wrote a test and then realized that this is actually about a self name. This should probably be changed, but it's quite a bit harder than I initially thought. Therefore not a high priority.

def test_defined_param_names(names):
    code = dedent("""
        class B:
            def __init__(self):
                self.z = 5
        """)

    # Will contain 4 definitions: class B, def __init__, self, z
    all_names = names(
        code,
        definitions=True,
        all_scopes=True,
        references=False,
    )
    assert all_names[-1].name == "z"
    assert all_names[-1].parent().name == "__init__"

    assert all_names[1].name == "__init__"
    assert sorted([x.name for x in all_names[1].defined_names()]) == sorted(["self", "z"])

@mpanarin Does it makes sense for you if other side effects are also listed? IMO that's the only thing that makes sense. Say someone your code writes sys.path = []. That would also be a definition, right?

@mpanarin Does it makes sense for you if other side effects are also listed? IMO that's the only thing that makes sense. Say someone your code writes sys.path = []. That would also be a definition, right?

This does make sens, yes. Is there any possibility to easily narrow it down on the client side after?
Something like side-effect parameter to differentiate would be great, but definitely not an obligatory thing.

For now there's probably nothing like this. If anything self modifications would also be side effects.

Decided that it might make sense to make this possible in the future, but it's really a new type of definition or as I said a side effect and should be listed as such. Therefore a new feature, not a bug.

I added is_side_effect to definition. This means that you have to use names(references=True) and then filter with definition.is_side_effect(). IMO this is the best way to do this for now.

Will be part of the 0.17.0 release.