model-writtenrun 9233 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Ryan Lambert opened the thread by investigating why xmloption = CONTENT rejected valid XML containing DOCTYPE and proposing a different libxml parsing API; Chapman Flack pointed him to existing parser and documentation patches awaiting review #1 #2.
- Tom Lane questioned the existing patch's dependence on libxml internals and suggested stripping DOCTYPE, while Chapman explained that DOCTYPE must remain available to the document parser and proposed detecting it before choosing the parsing mode #4 #6.
- Tom proposed trying both parsing modes, but Chapman argued that a short pre-scan would avoid reparsing large inputs and make error selection clearer; Tom emphasized that dump/restore failures warranted a simple, safely backpatchable fix #7 #8 #11 #14.
- Chapman submitted an explicit DOCTYPE pre-scan, a bounded-length optimization in parse_xml_decl(), regression tests, and the outstanding documentation patch together #16.
- Chapman identified detailed processing-instruction validation as a possible simplification, and Tom removed that validation and reorganized parsing control flow before committing the parser portion as 8d1dadb25b, accepting possible differences in malformed-input diagnostics #17 #18.
- Tom also committed 4870dce37f, an unattached companion change forcing xmloption = content during restoration so the expanded parser acceptance would reliably fix dump/reload failures.
- After Chapman clarified that the parser change had already reached master and supported stable branches, Ryan tested master, PostgreSQL 11, and PostgreSQL 9.6 and confirmed that it fixed his reported problem; this testing occurred after the parser commits #23 #26.
- The remaining work was documentation, and Ryan reviewed an intermediate revision, reporting successful checks and recommending clearer library naming, fewer parenthesized sentences, and consistent node-set terminology #27 #31.
- Chapman explained why the documentation should discuss library-dependent limitations, and Alvaro Herrera recommended the name libxml2 and stressed that observed document ordering was not guaranteed; Chapman incorporated those points and Ryan's editorial feedback in the documentation remainder #33 #36 #38.
- Chapman then noticed that the existing FOR ORDINALITY description incorrectly promised input-document order and revised it to describe result node-set order, with an explicit XPath 1.0 caveat #39 #40.
- Ryan applied and reviewed the revised documentation, checking terminology and the ordering explanation, before Tom committed it with light copy-editing as 12d46ac392; Chapman confirmed that this completed the commitfest work #41 #42 #43.
- Alvaro proposed backpatching documentation corrections to PostgreSQL 10 and 11, and Chapman agreed to preserve useful compatibility explanations while adjusting claims about behavior whose fixes had not been backpatched #44 #45.
- After a four-month gap, Chapman supplied branch-specific documentation patches describing older XMLTABLE conversion limitations and supported syntax, reporting a successful PostgreSQL 11 documentation build but no PostgreSQL 10 build verification #46.
- Alvaro deferred the patches past the imminent minor releases because travel left insufficient review time, then committed them as 603a28b449 and its associated backport with light markup and wording corrections, declaring the work complete in the thread's final message #47 #49.