[R] Flaky BOM removal test

[asfimport]

The test introduced in ARROW-14644 appears to be flaky.
See example failed runs:
https://github.com/apache/arrow/runs/4466790381?check_suite_focus=true#step:8:21277
https://github.com/apache/arrow/runs/4463832536?check_suite_focus=true#step:9:22039

── Failure (test-dataset-csv.R:297:3): open_dataset() deals with BOMs (byte-order-marks) correctly ──
`object` (`actual`) not equal to `expected` (`expected`).

actual vs expected
                a b
- actual[1, ]   3 4
+ expected[1, ] 1 2
- actual[2, ]   1 2
+ expected[2, ] 3 4

  `actual$a`: 3 1
`expected$a`: 1 3

  `actual$b`: 4 2
`expected$b`: 2 4

Reporter: Antoine Pitrou / @pitrou
Assignee: Jonathan Keane / @jonkeane

Related issues:

• [C++] open_dataset doesn't ignore BOM in csv file (is caused by)

PRs and other links:

• GitHub Pull Request #11917

_{Note: This issue was originally created as ARROW-15041. Please see the migration documentation for further details.}

[asfimport]

Antoine Pitrou / @pitrou:
@wjones127

[asfimport]

Dragoș Moldovan-Grünfeld / @dragosmg:
This seems to have something to do with the order in which the 2 files are created and subsequently brought together with open_dataset(). Creating file 2 first doesn't trigger a test failure.

  temp_dir <- make_temp_dir()
  writeLines("\xef\xbb\xbfa,b\n1,2\n", con = file.path(temp_dir, "file1.csv"))
  writeLines("\xef\xbb\xbfa,b\n3,4\n", con = file.path(temp_dir, "file2.csv"))

  expect_equal(
    open_dataset(temp_dir, format = "csv") %>% collect(),
    tibble(a = c(1, 3), b = c(2, 4))
  )

[asfimport]

Jonathan Keane / @jonkeane:
Issue resolved by pull request 11917
#11917