GH-48832: [R] Fix crash with zero-length POSIXct tzone attribute

[thisisnic]

Rationale for this change

In R 4.5.2+, as.POSIXct(x = NULL) creates a zero-length POSIXct with INTSXP (integer) type, not REALSXP (double). The GetVectorType() function in r_to_arrow.cpp only checked for POSIXct in the REALSXP branch, so it misclassified zero-length POSIXct as INT32.

What changes are included in this PR?

Map any POSIXct object to correct class by adding the check to the INTSXP branch of GetVectorType()

Are these changes tested?

Yes

Are there any user-facing changes?

No

• GitHub Issue: [R] arrow::write_parquet error with zero-length datetimes in R 4.5.2 #48832

[thisisnic]

FYI, I used Claude to generate this PR. I am not super familiar with the .cpp code it modified, but it looks convincing, and the tests replicate the original error.

[jonkeane]

Oh, huh I wonder if this type is new? Or maybe I "just" forgot about it when I made these changes?

[thisisnic]

🤖 says that it's not new, but historically, POSIXct was always REALSXP (backed by a double) until R 4.5.2, but now for zero-length POSIXct objects that's no longer the case, so it needs checking in the INTSXP path.

Or, with more detail:

So in R 4.5.0, R changed as.POSIXct({}) (i.e., zero-length POSIXct) to be internally stored as integer (INTSXP) rather than double (REALSXP). This is a deliberate change for consistency - zero-length date-time objects don't need the full double precision since they have no actual values to store.

[jonkeane]

@github-actions crossbow submit test-r-versions

[github-actions]

Revision: 8e29ee5

Submitted crossbow builds: ursacomputing/crossbow @ actions-c0634dd2bf

Task	Status
test-r-versions

[thisisnic]

══ Warnings ════════════════════════════════════════════════════════════════════
── Warning ('test-gcs.R:122:1'): (code run outside of `test_that()`) ───────────
error in running command
Backtrace:
    ▆
 1. ├─testthat::skip_if_not(system("storage-testbench -h") == 0, message = "googleapis-storage-testbench is not installed.") at test-gcs.R:122:1
 2. │ └─base::isTRUE(condition)
 3. └─base::system("storage-testbench -h")

══ Failed tests ════════════════════════════════════════════════════════════════
── Error ('test-s3.R:49:5'): read/write Feather on S3 ──────────────────────────
Error: IOError: When initiating multiple part upload for key '1774901280.20806/test.feather' in bucket 'arrow-datasets': AWS Error ACCESS_DENIED during CreateMultipartUpload operation: User: arn:aws:iam::855673865593:user/crossbow is not authorized to perform: s3:PutObject on resource: "arn:aws:s3:::arrow-datasets/1774901280.20806/test.feather" because no identity-based policy allows the s3:PutObject action (Request ID: YHWYK1PAJ4PQ4RTW)
Backtrace:
    ▆
 1. └─arrow::write_feather(example_data, bucket_uri(now, "test.feather")) at test-s3.R:49:5
 2.   └─arrow:::make_output_stream(sink)
 3.     └─fs_and_path$fs$OpenOutputStream(fs_and_path$path)
 4.       └─arrow:::fs___FileSystem__OpenOutputStream(self, clean_path_rel(path))
── Error ('test-s3.R:55:5'): read/write Parquet on S3 ──────────────────────────
Error: IOError: When initiating multiple part upload for key '1774901280.20806/test.parquet' in bucket 'arrow-datasets': AWS Error ACCESS_DENIED during CreateMultipartUpload operation: User: arn:aws:iam::855673865593:user/crossbow is not authorized to perform: s3:PutObject on resource: "arn:aws:s3:::arrow-datasets/1774901280.20806/test.parquet" because no identity-based policy allows the s3:PutObject action (Request ID: YHWZWQHNDRM7CJ24)
Backtrace:
    ▆
 1. └─arrow::write_parquet(example_data, bucket_uri(now, "test.parquet")) at test-s3.R:55:5
 2.   └─arrow:::make_output_stream(sink)
 3.     └─fs_and_path$fs$OpenOutputStream(fs_and_path$path)
 4.       └─arrow:::fs___FileSystem__OpenOutputStream(self, clean_path_rel(path))
── Error ('test-s3.R:60:5'): RandomAccessFile$ReadMetadata() works for S3FileSystem ──
Error: IOError: Path does not exist 'arrow-datasets/1774901280.20806/test.parquet'. Detail: [errno 2] No such file or directory
Backtrace:
    ▆
 1. └─bucket$OpenInputFile(paste0(now, "/", "test.parquet")) at test-s3.R:60:5
 2.   └─arrow:::fs___FileSystem__OpenInputFile(self, clean_path_rel(path))
── Error ('test-s3.R:44:1'): (code run outside of `test_that()`) ───────────────
Error: IOError: Path does not exist 'arrow-datasets/1774901280.20806/'. Detail: [errno 2] No such file or directory
Backtrace:
    ▆
 1. └─bucket$DeleteDir(now) at test-s3.R:44:1
 2.   └─arrow:::fs___FileSystem__DeleteDir(self, clean_path_rel(path))

[ FAIL 4 | WARN 1 | SKIP 30 | PASS 8080 ]
Error:
! Test failures.
Warning messages:
1: In for (i in seq_len(n)) { :
  closing unused connection 4 (/tmp/Rtmp0rl4Xk/file8f9832b0767c/part-0.csv)
2: In for (i in seq_len(n)) { :
  closing unused connection 5 (/tmp/Rtmp0rl4Xk/file8f986ad76486/part-0.tsv)
3: In for (i in seq_len(n)) { :
  closing unused connection 4 (/tmp/Rtmp0rl4Xk/file8f986bbb0db0/part-0.csv)
Execution halted

1 error ✖ | 0 warnings ✔ | 2 notes ✖

Failures look unrelated; I need to update some bucket settings. Probably worth getting that done and retriggering CI here though to make sure this isn't a new thing in 4.5.2.

[jonkeane]

I need to update some bucket settings. Probably worth getting that done and retriggering CI here though to make sure this isn't a new thing in 4.5.2.

Thanks for this! I was going to make an issue when I triggered these but then got distracted. It sounds like you know what's up there — I could guess but I haven't dug too much into it

[jonkeane]

Oops, forgot to actually approve this.

[thisisnic]

@github-actions crossbow submit test-r-versions

[github-actions]

Revision: 8e29ee5

Submitted crossbow builds: ursacomputing/crossbow @ actions-8e361a7bbb

Task	Status
test-r-versions

[thisisnic]

Needs rebasing...will try again

[thisisnic]

@github-actions crossbow submit test-r-versions

[github-actions]

Revision: e1f0d16

Submitted crossbow builds: ursacomputing/crossbow @ actions-a2f34c245b

Task	Status
test-r-versions

[conbench-apache-arrow]

After merging your PR, Conbench analyzed the 3 benchmarking runs that have been run so far on merge-commit d035788.

There were 4 benchmark results with an error:

• Commit Run on amd64-m5-4xlarge-linux at 2026-04-02 17:03:17Z
  • tpch (R) with engine=arrow, format=native, language=R, memory_map=False, query_id=TPCH-15, scale_factor=10
  • csv-read (Python) with compression=gzip, dataset=fanniemae_2016Q4, output_format=arrow_table, streaming=file
• and 2 more (see the report linked below)

There were no benchmark performance regressions. 🎉

The full Conbench report has more details. It also includes information about 2 possible false positives for unstable benchmarks that are known to sometimes produce them.