openpyxl type disagreements

[multimeric]

I originally posted this on the openpyxl GitLab but it seems that they don't maintain the type annotations.

The first issue is that worksheet[start:end] seems to return a tuple[tuple[Cell]], but the type annotations imply that it returns a tuple[cell].

You can see this issue in the following code:

from typing import reveal_type
from openpyxl import Workbook
from openpyxl.worksheet.worksheet import Worksheet

wb = Workbook()
ws = wb.active
if isinstance(ws, Worksheet):
    for cell in ws["A1":"A12"]:
        reveal_type(cell)

Type checkers believe one thing:

$ pyright tuple_excel.py
tuple_excel.py:9:21 - information: Type of "cell" is "Cell"

Python disagrees:

$ python tuple_excel.py
Runtime type is 'tuple'

Versions:

openpyxl==3.1.2
types-openpyxl==3.1.0.20240402
Python 3.12.2

---

The second issue relates to a single string index:

from typing import reveal_type
from openpyxl import Workbook
from openpyxl.worksheet.worksheet import Worksheet

wb = Workbook()
ws = wb.active
if isinstance(ws, Worksheet):
    reveal_type(ws["A"])

$ python tuple_excel.py
Runtime type is 'tuple'

$ pyright tuple_excel.py
tuple_excel.py:8:17 - information: Type of "ws["A"]" is "Any"

[srittau]

Cc @Avasam

[Avasam]

Unfortunately, when it comes to using a str as a key, there is simply no way to know statically whether you are targeting an individual cell, row, column or full range. So the type would be Cell | tuple[Cell, ...] | tuple[tuple[Cell, ...], ...], which is too distruptive so we fallback to Any (see python/mypy#1693)

The case for slicing can be improved, but unfortunately only to remove false-positives, in favor of false-negatives. (which we prefer).