GH-50399: [C++][Gandiva] Use timegm in DaysSince helper in Gandiva date_time_test

[amoeba]

Rationale for this change

DateTimeTestProjector.TestDateTime fails for me locally and it looks like the issue is that it's comparing datetimes calculated with two different methods:

1. Epoch(): Uses timegm as of GH-49952: [C++][Gandiva] Use timegm in date_time_test utilities #49953
2. DaysSince() was still using mktime

What changes are included in this PR?

Updated DaysSince() helper to use timegm to match other helpers so comparisons can be made without local time one handling issues.

Are these changes tested?

Yes. The test that fails locally now passes locally with this.

Are there any user-facing changes?

No.

• GitHub Issue: [C++][Gandiva] date_time_test fails locally due to UTC and local time comparison #50399

[amoeba]

@Reranko05 @lriggs would either of you want to look at this PR?

[Copilot]

Pull request overview

This PR fixes inconsistent timezone handling in Gandiva’s datetime test utilities by switching the DaysSince() helper from mktime() (local time) to timegm() (UTC), aligning it with Epoch() and other helpers so date/time comparisons are stable across platforms and DST settings.

Changes:

• Updated DaysSince() to use timegm() instead of mktime().
• Updated the fatal log message to match the new helper (timegm() failed).

[github-actions]

⚠️ GitHub issue #50399 has been automatically assigned in GitHub to PR creator.

[Reranko05]

LGTM. Thanks!

[amoeba]

Thanks for looking @Reranko05 and @lriggs.

Let's get at least one committer/PMC review and then I'll merge.

[adamreeve]

Looks good to me thanks @amoeba. I also saw this test failure when testing the 25.0.0 release, and can confirm that with this change the DateTimeTestProjector tests all pass for me.

[amoeba]

Thanks @adamreeve!

[conbench-apache-arrow]

After merging your PR, Conbench analyzed the 4 benchmarking runs that have been run so far on merge-commit 6096d41.

There were no benchmark performance regressions. 🎉

The full Conbench report has more details.

[conbench-apache-arrow]

After merging your PR, Conbench analyzed the 4 benchmarking runs that have been run so far on merge-commit 6096d41.

There were no benchmark performance regressions. 🎉

The full Conbench report has more details.