Emphasize that filter(x, A, B) is not strictly the same as filter(filter(x, A), B)

[MichaelChirico]

Have had to re-confirm this for myself a few times:

filter(mtcars, cyl < max(cyl), hp < max(hp)) |> dim()
# [1] 18 11

# vs
filter(filter(mtcars, cyl < max(cyl)), hp < max(hp)) |> dim()
# [1] 17 11

This in ?dplyr hints at what's going on:

If multiple expressions are included, they are combined with the & operator.

But this behavior is a bit more subtle / worth calling out IMO. This came up again recently here:

r-lib/lintr#2305 (comment)

FWIW it's also really not clear from reading the filter.data.frame implementation without being well-versed in {dplyr} internals.

[markolipka]

Seems to me that in the second case there will be a different value for 'max(hp)' because of the pre-filtering of the dataset. Therefore the B's in the issue title are not the same in the compared cases.

[MichaelChirico]

Yes, that's exactly the point.

I'm not saying the behavior is wrong, I'm saying it may not be obvious how filter() will process ... from the current documentation.

[markolipka]

If you compare the two cases with identical values, the results are identical:

> filter(mtcars, cyl < 8, hp < 335) |> dim()
[1] 18 11
> # vs
> filter(filter(mtcars, cyl < 8), hp < 335) |> dim()
[1] 18 11

I think there is nothing wrong here...

[MichaelChirico]

Not going to waste more of my time engaging when it's clear you haven't read the issue or my response carefully.