[PurHur]

Summary

• After mutable DateTime::add/sub with a known DateInterval, refresh compileTimeDateTimeTimestamp + dateTimeLocalInstants (or invalidate on runtime LLVM paths) so format() / getTimestamp() no longer return the construct instant.
• Closes Regression: AOT DateTime::add/sub with DateInterval is a silent no-op for format/getTimestamp (re-#33781/#30760, ext/date/php_date.c) #33911 (re-AOT: DateTime{,Immutable}::add/sub with runtime DateInterval aborts compile (re-#30760, ext/date/php_date.c) #33781 / Stdlib: DateTime::add()/sub() AOT segfault / silent no-op — VM/JIT match Zend (ext/date/php_date.c) #30760). Peer of Regression: AOT DateTime::diff() fails after two locals — New_ resultOp unnamed so construct stamps never bind (re-#27309) #33906 stamp publishing: stamps were present, mutation wrote __dt_timestamp, but reads preferred the stale stamp.

Test plan

• ./script/aot-smoke.sh → 8 passed, 0 failed
• ./script/phpunit.sh --filter Issue33911DateTimeAddSubStampAotTest → OK (3 tests, 19 assertions)
• Zend vs AOT repro:

  add=2020-02-15 / sub=2020-02-15 / cfd=2020-01-02 02:00
  after_ts=1581724800 / after_fmt=2020-02-15
  

• #33906 two-local diff still prints 9
• php-src: ext/date/php_date.c — zim_DateTime_add / zim_DateTime_sub

Made with Cursor