model-writtenrun 7729 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Tom Lane reported that macro expansion let some uses of \w as a range endpoint slip through despite the documented prohibition, and considered changing stable branches too invasive in #1.
- Joel Jacobson compared PostgreSQL, Perl, and JavaScript behavior in #2, and Alvaro Herrera explained their treatment of adjacent hyphens as literals in #3; Tom initially considered adopting that interpretation in #4.
- Tom rejected relaxing range syntax because XQuery and SQL use character classes after a hyphen for set subtraction, and posted a two-patch series in #6 that replaced macro expansion, enabled complemented escapes inside brackets, and separately proposed making \D and \W always match newlines.
- Joel tested approximately 4.47 million regex/subject pairs and analyzed the changed \W matches in #7, then examined \D differences in #8, supporting the proposed newline behavior.
- Alvaro recommended exposing the word character class based on GNU precedent in #9; Tom independently reached the same conclusion, accepted the newline change, and proposed removing the obsolete re_syntax.n documentation in #10, acknowledging Alvaro's supporting evidence in #11.
- The thread ended with Tom attaching the revised series in #12, exposing the named word class while retaining the original compiler architecture and unchanged newline patch; neither patch had yet landed at that message.
- The following day, 2a0af7fe46 landed the first patch from #12 with regression-test cleanup, and 7dc13a0f08 landed the second with expanded documentation explaining the new newline behavior and how to preserve the old behavior.
- Finally, 301ed8812e implemented the deletion proposed in #10, removing re_syntax.n because the maintained user-facing documentation already lived in func.sgml.