ARROW-8217: [R] Unskip previously failing test on Win32 in test-dataset.R from ARROW-7979

[wesm]

Neal found that the test passed in a debug build, let's see if the test still fails on master

[github-actions]

https://issues.apache.org/jira/browse/ARROW-8217

[wesm]

Good grief. I'll start bisecting by hand on my fork to see if I can determine which patch fixed this. In the meantime we can merge this, I'm going to remove the commented-out code

[wesm]

OK, @nealrichardson @fsaintjacques I think the mystery is solved. This appears to have been fixed by b9b9ece (see the RTools runs at https://github.com/wesm/arrow/branches/all?query=bisect -- ignore the red X's and click inside to see whether the RTools run succeeded or failed). So apparently there was some memory corruption happening that only provoked a segfault in the 32-bit build. Not sure

Good question is whether valgrind would have detected this on Linux builds of the R library or similar.

[nealrichardson]

😅 all's well that ends well? Merge this?

[wesm]

Well, what's weird is that the failing test should not have been affected by the bug that was fixed in b9b9ece, so there is still some mystery here of understanding where the bad memory access or undefined behavior originated. If it comes up again we can revert to this discussion and produce a build prior to b9b9ece so that we can reproduce the segfault again and get a gdb backtrace

[wesm]

+1, anyway, merging this