model-writtenrun 9639 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
Oct - Nov 2018 · 5 entries
- Chapman Flack opened the discussion with a wiki comparison of PostgreSQL and SQL/XML standards in #1, and Alvaro Herrera connected it to Markus Winand's existing XMLTABLE fixes in #2.
- The initial discussion considered replacing libxml2 with an XQuery-capable library, but dependency and compatibility concerns remained unresolved; Pavel Stehule offered incremental XMLTABLE fixes in #11, while Chapman explained the semantic incompatibilities between XPath 1.0 and XQuery in #16.
- Pavel's initial patches #18 and #22 added boolean and numeric XPath results, including numeric conversion of booleans, while Chapman's Oracle comparisons in #23 clarified the intended target-type-dependent behavior.
- Pavel substantially reworked conversion in #24 around libxml's native XPath routines, also addressing descendant-text handling inspired by Markus's earlier work.
- Markus demonstrated missing XML markup and incorrect document-node output in #25, supplied an inline node-classification fix, and Pavel incorporated it with Oracle-checked regression cases in #26.
Dec 2018 - Mar 2019 · 19 entries
- After a gap, Chapman tested the patch and highlighted both regression-output differences and misleading claims about XPath compatibility in #29, starting a broader documentation effort.
- Chapman argued that rejecting BY VALUE was especially surprising because PostgreSQL already used value semantics in #33, and Pavel agreed to accept it while retaining BY REF compatibility in #35.
- Arthur Zakirov tested the result-processing patch and supplied a no-libxml build fix, corrected comments, and a documentation example in #36; Pavel folded those changes into #37.
- Chapman added alternate expected output and the separate passing-mechanism patch in #40, then identified missing escaping for XPath strings assigned to XML columns in #42, which Pavel fixed in #43.
- Documentation investigation exposed stale claims about optional COLUMNS clauses, references to earlier XMLTABLE output columns, and expression evaluation; the discussion in #47, #48, #52, and #57 established that deferred defaults mattered particularly for volatile functions.
- A parallel documentation-style discussion began when Chapman noticed contradictory link guidance in #46; Peter Eisentraut explained that PDF output already supplied URL footnotes in #71, leading Chapman to request removal of the obsolete rule in #72.
- Chapman assembled the behavior fixes, passing syntax, and new compatibility documentation in #69, added a missing passing-syntax test in #73, and reported successful comprehensive testing in #74.
- Pavel reviewed the passing syntax and documentation and reported passing tests in #79, while acknowledging that the result-processing component was his own work.
- After the commitfest rollover, Chapman proposed broadening the work to accept documents with DTDs as XML CONTENT and eventually support named parameters in #81; only the CONTENT change received an attached implementation.
- The expanded submission #83 rebased the existing work, corrected expected output and documentation, and added a parser fallback that retried CONTENT as DOCUMENT when a DTD triggered a specific libxml error; a larger parser rewrite was explicitly deferred.
- Peter committed the README.links cleanup as b060e6c1f5 and announced it in #84.
- Ramanarayana tested the patches in #85, and the apparent extra plus sign in XML output was resolved as psql's newline display through #86 and #87.
- Alvaro committed the passing-mechanism component of #83 as eaaa5986ad for PostgreSQL 12, making documentation wording and markup edits described in #90.
- Alvaro then committed the result-processing component as 251cf2e27b, renaming and relocating the allocation helper and combining regression queries as explained in #92; the commit avoided backpatching because established behavior would change.
- Buildfarm crashes immediately followed the result-processing landing in #93, prompting Pavel's defensive error-handling patch #95 and then a reproduced old-libxml workaround that avoided copying document nodes in #97.
- Alvaro considered xmlFreeNodeList in #101, but Pavel identified differences in its cleanup behavior and proposed explicitly selecting xmlFreeDoc for document nodes in #103; Alvaro implemented that selection in #105, which landed with light edits as 2e616dee9e.
- Pavel subsequently questioned the missing volatile qualifier on the cleanup pointer in #108, Alvaro acknowledged removing it in #109, and Tom Lane corrected the declaration in b2b819019f.
- Chapman also proposed clearer internal-versus-external link instructions in #102, and Peter reported pushing that additional documentation update in #110.
- The broader XML documentation continued in thread #40246 and thread #40296; the supplied attribution places 12d46ac392 with the later revision #422313, and the subsequent older-branch adaptation 603a28b449 with #430919.
May - Jun 2019 · 1 entry
- At the thread's close, the incremental behavior fixes had landed, but this thread recorded no landing for its CONTENT/DOCTYPE patch or named-parameter proposal; Alvaro and Chapman returned to unresolved XQuery and regular-expression conformance in #111 and #112, with Chapman still favoring external XQuery libraries.