[R] R builds failing with error 'Invalid: Timestamps already have a timezone: 'UTC'. Cannot localize to 'UTC''

[thisisnic]

Describe the bug, including details regarding any error messages, version, and platform.

Lots of R builds (e.g. here) are failing with the following error:

Invalid: Timestamps already have a timezone: 'UTC'. Cannot localize to 'UTC'.

which comes from this line in the C++ library:

arrow/cpp/src/arrow/compute/kernels/scalar_temporal_unary.cc

Line 124 in b60bb96

return Status::Invalid("Timestamps already have a timezone: '", timezone,

This happened since we updated the vendored version of the date library, so perhaps could be related to that?

Component(s)

C++

[thisisnic]

This fails on the nightly builds, but I can't replicate this when locally, or when running it on PRs (see #3561).

[jorisvandenbossche]

This happened since we updated the vendored version of the date library, so perhaps could be related to that?

Was that done in a specific PR? (to understand what changed)

[jorisvandenbossche]

I see that this is #35612. Now, that was actually passing on main, so it might also have been triggered by #35449 (which is failing with this on main, and was actually also already failing on the PR ..)

[thisisnic]

Yep, it was in #35612

[jorisvandenbossche]

My assumption is that the fix of #35449 caused one of the test dates to become UTC, causing the failure. If that's the case, we will need to fix the test I think

[thisisnic]

You're right; it is failing on the CI on that PR, e.g. https://github.com/apache/arrow/actions/runs/4900485911/jobs/8751167526

[jorisvandenbossche]

Here:

arrow/r/tests/testthat/test-dplyr-funcs-datetime.R

Line 292 in 99ef3aa

formats <- c(formats, "%a", "%A", "%b", "%B", "%OS", "%I%p", "%r", "%T%z")

This adds the format "%T%z", and that will now actually result in a tz-aware timestamp, while before it didn't parse the timezone correctly (that's the fix in #35449).

And "strptime" dplyr binding ends up calling assume_timezone, which will not work if the timestamp already has a timezone.