Some `__str__` methods of nodes raise errors and warnings

[jacobtylerwalls]

Steps to reproduce

>>> fd = astroid.nodes.FunctionDef()
>>> str(fd)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/Users/jacobwalls/astroid/astroid/nodes/node_ng.py", line 219, in __str__
    value = getattr(self, field)
            ^^^^^^^^^^^^^^^^^^^^
AttributeError: 'FunctionDef' object has no attribute 'args'

Less severely, but still not great, launching python with warnings e.g. python -Wall:

>>> cd = astroid.nodes.ClassDef()
>>> str(cd)
/Users/.../astroid/astroid/nodes/scoped_nodes/scoped_nodes.py:2041: DeprecationWarning: The 'ClassDef.doc' attribute is deprecated, use 'ClassDef.doc_node' instead.

---

We should add a unittest that generates the str() and repr() of all node types.

python -c "from astroid import __pkginfo__; print(__pkginfo__.version)" output

2.12.13

[DanielNoord]

The first issue is caused by the fact that the str relies on postinit being called I think.

I ran into this while debugging astroid many times..

[crazybolillo]

I made a quick unit test on #2198. Am I going in the right direction? Also, I guess fixing the errors themselves will be part of this issue right?

[DanielNoord]

I'd prefer using pytest semantics instead of unittests but that's more a review comment.

I think the bigger issue is that the reliance on __postinit__ makes this very hard to solve. We would need to revisit either 1) what we put in the __str__ or 2) how we instantiate nodes. Both are not optimal I think.

[jacobtylerwalls]

I labeled this as a good first issue because I was hoping it wasn't requiring knowledge of any of that, just placing a good blank placeholder in the str/repr when the info is not available.

[DanielNoord]

That would require a fix here:

astroid/astroid/nodes/node_ng.py

Line 213 in a6eb2b8

value = getattr(self, field)

I guess that would work? We don't really need a regression test for it though (I think).

[jacobtylerwalls]

LGTM. There are bunch of overridden __str__ methods that might have similar problems, so I was thinking the unit tests would be nice. But if we find no other problems in the other methods, then such a test could be overly paranoid. EDIT: but writing the test would be the fastest way to find out!