model-writtenrun 5578 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
Oct - Nov 2021 · 7 entries
- Justin Pryzby reported that PostgreSQL 14 broke psql descriptions using database.schema.table, and Tom Lane confirmed that even a correctly spelled current database name failed #1 #3.
- Investigation showed that older versions silently discarded arbitrary leading name components, prompting Tom Lane and Peter Geoghegan to favor compatibility while Robert Haas and Stephen Frost argued for accepting only the current database prefix and rejecting garbage #4 #7 #12 #13 #17.
- Justin Pryzby proposed separating restoration of old behavior from stricter validation, and Mark Dilger supplied a preliminary restoration experiment, but the discussion moved toward a combined fix #20 #23 #28.
- Justin Pryzby's explanation that database-qualified names were copied from server logs supplied a practical reason to preserve three-part names, while Robert Haas favored literal database-name matching rather than matching database patterns #32 #33 #30.
- Mark Dilger extended the analysis to pg_dump and pg_dumpall, where silently ignoring a database prefix could select the wrong data, and Robert Haas requested explicit errors for cross-database references #35 #36.
- Mark Dilger's first indexed patch implemented client-side strcmp validation, rejected excessive name components and unquoted database metacharacters, retained pg_amcheck database patterns, and left reindexdb outside the change because it uses a different literal-name interface #37.
- The patch was repeatedly rebased through winter while the compatibility dispute remained unresolved; although Mark Dilger reported addressing documentation feedback, the corresponding attachment still lacked those fixes #39 #43 #44 #46 #48 #49 #50.
Mar - Apr 2022 · 10 entries
- Robert Haas revived the stalled decision in March, and Justin Pryzby and David G. Johnston endorsed accepting the current database prefix while rejecting invalid qualifications, emphasizing consistency with regclass and existing user workflows #51 #53 #54.
- After another rebase, Robert Haas supplied the complete patch plus a cleanup that removed special rejection of database metacharacters, and Mark Dilger folded that cleanup and the missing documentation into the next version #56 #57 #58.
- Robert Haas's detailed review led Mark Dilger to propagate psql errors correctly, distinguish disconnected sessions, remove unnecessary buffer refactoring, and retain quote-aware extraction of literal database names #59 #60.
- Concurrent support for dotted configuration names complicated integration, and Tom Lane and Greg Stark questioned the caller-side validation API; Robert Haas defended it because the frontend tools require different error handling, and no replacement API was adopted #61 #62 #64 #65 #67 #68.
- Mark Dilger's next submission accommodated the newer callers and logging interfaces, allowed an omitted dot-count output, and added degenerate-pattern tests while retaining the existing design #69.
- Although Robert Haas initially favored landing only in the development branch, David G. Johnston, Justin Pryzby, Tom Lane, and Vik Fearing supported treating the change as a PostgreSQL 14 bug fix and backpatching it #70 #71 #72 #73 #74.
- Mark Dilger supplied development and stable-branch versions with expanded quoted-name tests that exposed another parsing problem, fixing database-buffer selection in the final indexed attachment #75.
- Robert Haas committed the final attachment unchanged as d2d3547979 and confirmed that the development and PostgreSQL 14 patches had landed #75 #76.
- Thomas Munro then reported Windows test failures where wildcard arguments expanded into filenames, Robert Haas diagnosed the immediate cause, and Munro pointed to IPC::Run escaping differences; the supplied thread ends with that portability problem still under investigation #77 #78 #79 #80.
- The subsequent corrective commit a66e722cc1 removed five affected pg_dump tests while preserving the functional fix, leaving a systematic solution to Windows argument expansion for future work.