codeql
codeql copied to clipboard
False positive – "Statement has no effect" for Python type hint ellipsis
Description of the false positive
The ellipsis ... (three dots) is commonly used in Python to omit bodies for type hinting declarations. Since technically a lone ... is an expression statement, LGTM/CodeQL marks it as a statement without effect.
For example, we could have something like this:
@overload
def some_func(a: float) -> str:
... # body is omitted since this is just for type hinting
@overload
def some_func(a: int) -> int:
...
def some_func(a):
return a if isinstance(a, int) else str(a)
In this case, both @overload definitions exist only for type hinting. Only the final definition is actually executable at runtime. Emitting a warning for the type hint bodies is misleading; they have no effect on purpose.
Roughly, it should be fine to suppress the warning for def blocks if:
- the entire body is
... - it is in a type-declaration context such as
@overloadorProtocol
See PEP 8 on "Function annotations …" for reference.
Code samples or links to source code
https://github.com/maxfischer2781/asyncstdlib/blob/b4b3a557bf7d0c411495d9fd7bd8e8f500ab6b97/asyncstdlib/builtins.py#L328-L335
https://github.com/maxfischer2781/asyncstdlib/blob/57f372b3ea8367b9cb448647664e582e783e4e75/asyncstdlib/_lrucache.py#L77-L79
URL to the alert on GitHub code scanning (optional)
https://github.com/maxfischer2781/asyncstdlib/security/code-scanning/33
https://github.com/maxfischer2781/asyncstdlib/security/code-scanning/98
Also seeing lot (several pages) of these spurious false positives: e.g. https://github.com/bokeh/bokeh/security/code-scanning/339
Besides pep8 it's worth noting the Mypy docs explicitly demonstrate this usage for @overload specifically: https://mypy.readthedocs.io/en/stable/more_types.html#function-overloading
I've just updated the description since I noticed that not only @overload is a frequent situation for this false positive but also Protocol. The docs also explicitly list an example using ellipsis:
class Proto(Protocol):
def meth(self) -> int:
...
Is there any workaround for this issue? I'm facing the same problem (example), and it's making PRs much harder to review.