0 Fork 0

T

d74d6ea249 Build the calendar month grid in UTC (#10476 )

* Build the calendar month grid in UTC

The grid walked a cursor at local midnight and stepped it with setDate(getDate() + 1). Chile starts DST at 24:00 on the first Saturday of September, so local midnight of 6 September 2026 does not exist, and Qt's V4 engine resolves a missing local time backward -- onto 5 September 23:00, the same calendar date, which the next iteration reads again. September 2026 draws 30 31 1 2 3 4 5 and then 5 6 7 8 9 10 11, and every later date sits one weekday early. Cuba, Lebanon, the Azores and Egypt change over at 00:00 too and break the same way in their own transition month; where the gap lands on a leading padding cell instead, the whole six-week grid shifts by a day with no duplicate in it.

Node resolves the identical gap forward, so the suite has never seen this: running the shipped monthGrid under Node across all 598 tzdata zones for 2020-2035 reports nothing, while the same code under the QML engine breaks 402 grids. A sequence of civil dates never needed a wall clock, so it walks UTC, where every day is 24 hours long, and the test pins that by building the grid with a Date whose local constructor and accessors throw.

stepMonth reads a month back off the same kind of date. No zone puts a midnight gap on a first of the month, so it is not broken today, but it moves with the grid rather than waiting for a tzdata release to make it a second bug.

Co-Authored-By: Codex XHigh <noreply@openai.com>

* Name the calendar heading's month from local noon

The heading built a local midnight on the first of the month, and that midnight does not always exist: Cuba started DST at 00:00 on 1 April 2012, so in Havana the QML engine resolves it back onto 31 March and the heading reads March 2012 above an April grid. Noon exists on every first of the month there has been, and keeping it local is what makes Qt.formatDate print the right month in every zone.

The Apia case comes from #13186: Samoa skipped 30 December 2011 entirely, so unlike a one-hour midnight gap, Node's V8 cannot resolve its way past it, and the test fails on the old grid under Node instead of only in the shell.

Co-Authored-By: Codex Medium <noreply@openai.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: Codex XHigh <noreply@openai.com>
Co-authored-by: Omarchybot <317366263+omarchybot@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

2026-10-11 06:14:23 +02:00