model-writtenrun 7264 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Daniel Gustafsson opened the thread by reviving a proposal from the Meson discussion, attaching selectable TAP output alongside the traditional pg_regress format in #1.
- Daniel corrected ignored-test accounting and ECPG tag handling in #2, while Andres Freund questioned using TAP bailouts broadly in #3.
- After Andres reported segmentation faults in #5, Daniel fixed logfile handling and added a separate experiment reducing default verbosity with a --verbose option in #6.
- Daniel combined TAP and verbosity support in #7, completing output-path conversion and adding bailout handling for unrecoverable errors.
- Peter Eisentraut questioned maintaining two formats in #8, and Tom Lane insisted that timing and parallel-test information remain available in #9.
- Andres showed how TAP could preserve readable timings and parallel groups in #17, while Tom supported retaining stable schedule-order test numbering in #18.
- Daniel consequently replaced selectable formats with unconditional TAP output in #21, preserving useful test information while removing verbose setup chatter and targeting the original TAP specification.
- Andres requested aligned test numbers and indentation for parallel groups in #22; Daniel implemented those changes in #23 and corrected a compiler warning in #25.
- Andrew Dunstan confirmed that the buildfarm checks exit status rather than parsing the old output in #24, reducing compatibility concerns about the format change.
- Once Meson had landed, Andres requested TAP integration and flagged a missing format attribute in #26; Nikolay Shaplov then tested the output with prove and requested simpler shared formatting in #27.
- Daniel removed unnecessary logarithmic alignment calculations, introduced a shared status printer, and supplied Meson TAP integration in #28.
- Andres requested stderr failure diagnostics and centralized protocol prefixes in #29, and Ilmari proposed note/diag helpers in #30; Daniel introduced the typed TAP emitter in #31.
- Andres's local testing exposed malformed ECPG tag output and missing newlines in #32, leading Daniel to fix those cases, add a failure summary, and fold Meson integration into the main patch in #33.
- The same review identified that a failed child process could still produce a successful test result in #32; Daniel explicitly separated that issue from output formatting in #33.
- Nikolay requested clearer bailout semantics and less duplicated formatting in #34, Andres suggested bail_noatexit in #36, and Daniel adopted that name and a single status printf in #37.
- After further discussion of recursion and parallel-test indentation, Nikolay declared the patch ready in #43; Daniel then repaired va_list handling and polished comments in #44.
- Following a pause, Vignesh supplied a Meson rebase in #46, and Daniel resumed with #50, adapting the patch to upstream removal of ignored-test support and making small cleanups.
- The process-status bug was fixed separately in thread #47010, where #528697 landed unchanged as 337903a16f, making nonzero child exits fail both scheduled and individual tests; Daniel rebased this thread's TAP patch over it in #51.
- Peter's late review in #52 requested cleaner translation handling, conversion of remaining error paths, TAP-compatible Makefile banners, and clearer output; Daniel addressed these and centralized newline handling in #53, followed by a compiler-warning fix in #54.
- Peter found incorrect Makefile escaping, leftover disabled code, punctuation issues, and awkward Meson indentation in #55; Daniel fixed these and replaced whitespace grouping with '+' for parallel tests and '-' for sequential tests in #56.
- Daniel updated the documented output example in #57, then committed that version with only additional comments and documentation as 558fff0adf.
- Daniel confirmed the final TAP landing in #58, completing this thread's work after the separate process-status fix; the optional traditional format and verbose mode had been abandoned, while further readability adjustments were left to feedback from wider use as discussed in #56.