Fix type annotation to allow int keys in data_regression.check()

[tsvikas]

Summary

The DataRegressionFixture.check() method's type annotation was overly restrictive, only allowing dict[str, Any] when the documentation states it accepts "any yaml serializable dict". This PR updates the type annotation to dict[str | int, Any] to allow integer keys.

Motivation

Integer keys are commonly used in test data (e.g., for indexing test cases by number), and YAML fully supports them. Users currently have to convert int keys to strings unnecessarily to satisfy type checkers, even though the code works correctly at runtime.

Changes

• Updated type annotation in src/pytest_regressions/data_regression.py:39 from dict[str, Any] to dict[str | int, Any]
• Added test_integer_keys() test case demonstrating integer key support

Type Choice Rationale

We chose str | int rather than broader types because:

• int: Commonly used for test case indexing and numbering
• str: Already supported
• Not float: Floating-point keys are problematic for dictionary lookups due to precision issues
• Not bool: Can cause confusion in YAML between true and "true"

Testing

The new test case test_integer_keys() verifies that integer keys work correctly with the data regression fixture.

Real-World Use Case

This fix enables code like:

# Previously required str(n) to satisfy type checker
data = {
    n: test_function(n)  # Now works with type checking!
    for n in numbers
}
data_regression.check(data)

[nicoddemus]

Thanks @tsvikas

The Python 3.9 job failed; we will likely drop support for 3.9 soon, but probably simpler in your case to just update the type annotations for this PR.

[tsvikas]

Done :)

[tsvikas]

Hi @nicoddemus , just a gentle nudge on this. I’ve updated the type annotations to Union for Python 3.9 compatibility as requested. Let me know if there’s anything else needed to get the CI workflows running!

[nicoddemus]

Thanks for the nudge, this went under my radar!

[nicoddemus]

Perhaps we might as well just use

Suggested change

	data_dict: dict[Union[str, int], Any],
	data_dict: dict[Any, Any],

here?

We don't have a type for "anything yaml serializable", so might as well use Any and hope for the best.

If you agree, we can also just remove the regression test, seems overkill to have a test for a specific data type.

[tsvikas]

It's up to you, ultimately. My logic: if we look at allowed YAML keys (bool, float, str, int, null), then str | int is the reasonable choice because:

• str and int are practical and commonly used
• bool can cause confusion (true vs "true", yes vs "yes")
• float has precision issues for dict lookups
• null as a key is weird, and don't speaks well with the Python world (None vs float('na'))

[nicoddemus]

Good points, lets go with your suggestion then!

[tsvikas]

The CI failure is unrelated to this PR - it's a matplotlib/pyparsing deprecation warning that only appears on Python 3.9. The test is importing matplotlib which uses pyparsing's deprecated oneOf (should be one_of). Since Python 3.9 reached EOL in October 2025, matplotlib/pyparsing likely won't fix it for py39.

I suggest that we will approve this PR, and write a separate PR that drop py39 support (remove from tox.ini envlist). alternatively, we can add a filterwarning to ignore this specific deprecation

[nicoddemus]

I suggest that we will approve this PR, and write a separate PR that drop py39 support

Agreed, I will go ahead and merge this.