fix: Remove deprecation message from upstream polars in some cases with %in%

[ppanko]

Presumably fixes (#258); not sure if this change has implications elsewhere in the codebase.

[etiennebacher]

I don't think this would change any behavior. The message is about a deprecated usage that will change in a future version of polars, it was already doing the required change under the hood.

I don't have a good way to test the absence of a message since it appears that this is not really considered as a standard message by R, i.e. those tests wouldn't work as expected before this PR:

expect_message(
  iris_pd |> 
    head() |> 
    filter(Species %in% c("setosa", "virginica"))
)

# Error (but shouldn't):
# ! `filter(head(iris_pd), Species %in% c("setosa", "virginica"))` did not throw the expected message.

expect_silent(
  iris_pd |> 
    head() |> 
    filter(Species %in% c("setosa", "virginica"))
)
# doesn't error while it should

So I just added an extra test to ensure that the $implode() doesn't change anything.

Thanks for opening the issue and the PR!

[ppanko]

As far as I can tell, the deprecation message seems to go through the OS-level stderr. A workaround for testing whether %in% triggers the message could be to capture the output from a separate R session. For example:

testthat::test_that("`is_in` deprecation warning is not captured", {
 scriptToRun <-
   'sink <- dplyr::filter(tidypolars::as_polars_df(head(iris)), Species %in% "setosa")'
 result <- system2(
   R.home("bin/Rscript"), c("-e", shQuote(scriptToRun)), stderr = TRUE
 )
 testthat::expect_true(!any(grepl("issues\\/22149", result)))
})

[etiennebacher]

Thanks for the workaround, but that seems quite hacky and wouldn't work on Windows as far as I can see so I'm not convinced it should be part of the test suite. I had seen the deprecation message before, I just couldn't have it reliably so let it aside until you reported it.

[eitsupi]

Just wondering, doesn't snapshot testing work?

I think r-polars tests messages generated on the Rust side with snapshot tests.

[etiennebacher]

True, forgot about that, I guess I was mostly annoyed by the fact this doesn't happen on Windows (and potentially macOS, I can't test on this platform). I'll add a test later

[etiennebacher]

I just tested and the message also doesn't appear in snapshot tests (before the fix).

[ppanko]

polars' handle_result bridge doesn't seem to handle Rust-side warnings, so the message bypasses R and cannot be intercepted by R-side sinks.

Whether or not the message appears is dependent on how the R console/process captures the subprocesses' stderr. On my Windows machine, the message does not appear in RGui or RStudio, but does appear if using Rscript or Emacs+ESS.

[eitsupi]

Oh, thanks for the detail.
Maybe supporting Rust side warnings is a interesting new feature for savvy.