model-writtenrun 9602 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Tom Lane proposed measuring each regression script's runtime to identify disproportionately slow tests within parallel groups, attaching instrumentation for scheduled and individual tests in #1.
- Peter Geoghegan, Andres Freund, and Daniel Gustafsson supported collecting timings in #2, #3, and #4; Tom Lane landed the attachment unchanged as 72d71e0356 and announced it in #5.
- Peter Eisentraut reported that timings obscured pass/fail results and requested optional output in #6, while Tom Lane wanted timings enabled by default for buildfarm data and proposed more whitespace in #7; Michael Paquier also favored an optional switch in #9.
- Peter Eisentraut proposed separate logging or buildfarm configuration, then supplied an inline formatting change that removed parentheses and used an eight-character numeric field in #10; he separately requested alignment between successful and failed results in #11.
- Tom Lane accepted the spacing approach but raised localization and additional failure-text complications for alignment in #12; he rejected enabling prove --timer without compatibility checks in #14 and defended retaining failed-test timings as useful timeout evidence in #15, responding to John Naylor's suggestion in #13.
- Tom Lane applied Peter Eisentraut's inline spacing change as 93b5cc039e, announcing in #18 that they would try the wider field before making further changes.
- Christoph Berg found mixed successful and failed results still confusing and submitted a separate alignment patch in #19, padding individual-test output to FAILED and scheduled-test output to failed (ignored), with a narrower scheduled runtime field.
- Peter Eisentraut identified translation problems with separate padding and opposed accommodating the rarely used failed (ignored) case in #20; Christoph Berg acknowledged the translation issue in #21, while Alvaro Herrera noted that pg_regress had no translations in #22.
- Christoph Berg simplified the patch in #24 to pad the translated ok string to FAILED's width in both execution paths, dropping special alignment for ignored failures and retaining the existing runtime field width.
- Peter Eisentraut committed the revised attachment unchanged as 148cf5f462 and confirmed completion in the thread's final message, #25; runtime reporting remained enabled by default, with the spacing and alignment improvements landed.