https://issues.apache.org/jira/browse/ARROW-14659

if removed the test fails?

Fails how?

With the mutate line removed (line 128)

e.g.

 compare_dplyr_binding(
    .input %>%
      mutate(
        y = if_else(int > 5, fct, factor("a"))
      ) %>%
      collect(),
    tbl
  )

── Failure (test-dplyr-funcs-conditional.R:119:3): if_else and ifelse ───────────────────────────
`object` (`actual`) not equal to `expected` (`expected`).

`levels(actual$y)[1:4]`:   "a"             "g" "h" "i"
`levels(expected$y)[1:7]`: "a" "b" "c" "d" "g" "h" "i"

  `actual$y[4:10]`: NA 1 NA 2 3 4 5
`expected$y[4:10]`: NA 1 NA 5 6 7 8
Backtrace:
 1. compare_dplyr_binding(...) test-dplyr-funcs-conditional.R:119:2
 2. expect_equal(via_table, expected, ...) helper-expectation.R:129:4
 3. testthat::expect_equal(...) helper-expectation.R:42:4```

There is a similar test here, in which I solved the same failure by using:

transmute(across(where(is.factor), ~ factor(.x, levels = c(...))))

https://github.com/apache/arrow/pull/11272/files?authenticity_token=86nJx4XEwecmmySQatP4CD6%2BYi62xBNQWT4vtOE0iIzJT6wH0ZJM3imNvGEBkfiPii6v0VXSUmag%2B5%2F8ycn0CQ%3D%3D&file-filters%5B%5D=.R#diff-c63a873ac0b560d2ca7229ac1df2628df8396b0cc51fc247b71a2499c492fe3aR317-R320

I think you can use that approach here too!

Thanks @ianmcook, I replaced the as.character mutate with your approach above and added a comment to reference ARROW-14649. And apologies, I should have added a co-author to that commit.

Benchmark runs are scheduled for baseline = 4913352 and contender = b83e6b0. b83e6b0 is a master commit associated with this PR. Results will be available as each benchmark for each run completes.
Conbench compare runs links:
[Finished ⬇️0.0% ⬆️0.0%] ec2-t3-xlarge-us-east-2
[Failed] ursa-i9-9960x
[Finished ⬇️0.8% ⬆️0.13%] ursa-thinkcentre-m75q
Supported benchmarks:
ursa-i9-9960x: langs = Python, R, JavaScript
ursa-thinkcentre-m75q: langs = C++, Java
ec2-t3-xlarge-us-east-2: cloud = True