Using column selectors in rules

[EgeKaraismailogluQC]

I ran into a weird behavior where an already-validated data frame failed validation. The failing rule uses polars.selectors.string() to check if all string columns satisfy some condition. The issue goes away if I define the list of columns some other way. I sadly could not reproduce the issue, hence I ask it here as a question instead of a proper issue. Is there some implementation detail in dataframely that could explain this behavior, like adding latent columns at runtime? Should I avoid using selectors in the future?

[Oliver Borchert (borchero)]

Hey EgeKaraismailogluQC! I would be surprised if this was an issue in dataframely, but honestly hard to tell without MWE 😅

[Andreas Albert (AndreasAlbertQC)]

Bit of a longshot, but the only mechanism I could imagine for selectors to be funky is that they are evaluated on an intermediate dataframe that has more columns than your input dataframe. One example would be here, where we evaluate simple rule expressions after having evaluated group rule expressions. I don't think any of those should return string columns, so your specific example is not immediately easy to explain

[EgeKaraismailogluQC]

I was also thinking in this direction but did not come too far without knowledge of the internals. Thanks!