Wrong reported location for undefined references in f-strings

[khyox]

Repro:

"""mypy test: location for undefined references in f-strings"""

from typing import List

my_list: List[str] = ['Anna', 'Jonathan', 'Sandra', 'Joseph']
print(f'My list has {len(my_list)} elements.')

print('The list has %i elements.' % len(my_wrong_list))
print(f'The list has {len(another_wrong_list)} elements.')
print(f'The list has {missing_integer} elements.')
print('The list has {0:d} elements.'.format(len(yet_another_wrong_list)))

I get (mypy 0.520-dev):

> mypy --python-version 3.6 --show-column-numbers --check-untyped-defs test_mypy.py
test_mypy.py:1:5: error: Name 'another_wrong_list' is not defined
test_mypy.py:8:40: error: Name 'my_wrong_list' is not defined
test_mypy.py:10:6: error: Name 'missing_integer' is not defined
test_mypy.py:11:48: error: Name 'yet_another_wrong_list' is not defined

I'd expect:

mypy --python-version 3.6 --show-column-numbers --check-untyped-defs test_mypy.py
test_mypy.py:8:40: error: Name 'my_wrong_list' is not defined
test_mypy.py:9:26: error: Name 'another_wrong_list' is not defined
test_mypy.py:10:22: error: Name 'missing_integer' is not defined
test_mypy.py:11:48: error: Name 'yet_another_wrong_list' is not defined

So, for references in built-in functions in f-strings, both line and column are wrong (line is always reported as being the 1st, so these messages become the first ones reported by mypy) and for plain references in f-strings, just the reported column is wrong. Hope this helps.

[ethanhs]

@ambv you were working on something related to this, no? It was a CPython issue as I recall? Or was that a different issue related to f-strings?

[ambv]

Yes, this is bpo-30465 and the fix is in python/cpython#1800 but I won't have time before July to finish it.

[FichteFoll]

The referenced patch/PR has been merged into master and the 3.6 branch, so hopefully this will be resolved with the next CPython release.

[ethanhs]

The actual fix for mypy will happen when typed_ast is updated, which should be when the next Python version is released, but may be sooner.

[ethanhs]

Also, this is an issue with typed_ast, so I am closing this.

[gvanrossum]

If it's fixed in master (or 3.6) upstream it can be fixed in typed_ast, there's no need to wait for a CPython release. But there needs to be a separate PR for typed_ast (it doesn't automatically get copied from CPython nor does it somehow invoke the CPython version).

I opened an issue against typed_ast (python/typed_ast#56) to copy the fix there (but I'm not personally going to do the busy-work of creating the PR).