model-writtenrun 3508 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Jacob Champion proposed restoring json_errdetail() for frontend utilities by attaching error-message allocations to JsonLexContext, extracting work needed by OAuth and incremental JSON parsing while leaving full libpq compatibility outside this patch's scope #1.
- Andrew Dunstan supported the design and suggested separating Daniel Gustafsson's destroyStringInfo() helper, while Michael Paquier also requested that split and proposed a macro for repetitive token formatting #2 #3.
- Jacob split the helper into its own patch, converted additional callers, and introduced token_error(); Michael then supplied a revision placing the helper first, with small documentation adjustments and the JSON changes depending on it #4 #5.
- Daniel requested clearer ownership documentation, identified a whitespace mistake, questioned the macro's benefit, and proposed replacing the read-only StringInfo assertion with an error; Michael raised concerns about introducing more exit paths #6 #7.
- Tom Lane emphasized that libpq must never call exit(), Álvaro Herrera explained that the separate OAuth work already used PQExpBuffer for frontend allocation, and Daniel confirmed that this thread deliberately targeted frontend utilities using StringInfo #8 #9 #10.
- The read-only StringInfo check remained an assertion after Tom and Andrew argued against an elog-based path that could exit, and Daniel agreed to commit the series; Michael noted that his detailed review had concentrated on the helper #12 #13 #14 #15.
- Daniel committed the StringInfo helper from #5 as b783186515, correcting only the accidental whitespace change, and reported that the JSON patch would follow #17.
- Daniel then committed the JSON patch from #5 as d6607016c7, retaining the macro and making light comment and initialization-placement edits; he also reported an SSL buildfarm failure that he could not reproduce #18.
- Tom traced the SSL failure to an occupied TCP port, and Daniel confirmed that the failure cleared on its own, so no corrective patch or revert was needed #19 #20.
- Both patches had landed by the thread's closing thanks from Jacob and Michael, with no remaining patch under review here and the broader OAuth/libpq work left to its separate thread #21 #22 #10.