model-writtenrun 3775 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Andres Freund opened the thread in #1 with slow TAP-test timings, especially on Windows, proposing acceleration or smaller scripts that could run concurrently.
- Andres traced substantial backup-test overhead to missing -cfast and --no-sync options and supplied #2, reporting roughly halved standalone pg_basebackup test time; Robert Haas endorsed the algorithm-test change in #3, and it landed with comment edits as 839f9636b3.
- Discussion also exposed unexpectedly slow Windows checksum verification: Robert suggested differing implementations in #5, and Andres suspected the client was using PostgreSQL's slower implementation despite OpenSSL being enabled in #7, without a resolution shown here.
- Andres proposed reusing initdb output in #4, while Tom Lane recalled an earlier copying experiment with limited gains on disk-bound machines and suggested sharing running clusters in #6.
- Robert's split in #8 separated basic, integrity, and option tests but increased serial runtime through additional initdb runs; Andres supported the split's concurrency benefits and proposed creating a template during temporary installation in #9.
- Andres implemented template copying in #10, reporting substantial CPU savings, and Tom confirmed improvements on slower hardware in #11 before sharing his independent older prototype in #13.
- Windows copying and installation remained unfinished: Andres compared native copying tools in #14, raised xcopy path problems in #15, and Andrew Dunstan supplied a robocopy example with exit-status handling in #16; Andres still questioned where the legacy MSVC workflow should create templates in #17.
- After more than eighteen months, CI costs prompted Andres to revive the template patch in #18 with Meson support and native copying commands; he retained duplicated C and Perl logic because sharing it through initdb or a Perl helper introduced unwanted complexity or dependencies.
- Daniel Gustafsson reported successful testing and accepted the duplicated logic in #21; he and Andres declined broader support for extra initdb options in #22 and settled on the existing template terminology in #23.
- The revived template patch #18 landed as 252dcb3239, with Andres explaining in #24 that he added logging redirection for copying in pg_regress.
- Solaris buildfarm failures immediately exposed unsupported cp -a usage, so Thomas Munro supplied #25; Daniel endorsed its portable flags in #26, and Andres committed it unchanged as 1a4fd77db8, announcing the fix in #27.
- Nathan Bossart independently reported tests running about 30 percent faster in #28, confirming a practical benefit after the template optimization landed.
- In December, Matthias van de Meent reported that failed template initialization removed useful debugging files and supplied --no-clean additions for Make and Meson in #29; Daniel requested clearer long options in #30, which Matthias incorporated in #31.
- Daniel committed #31 unchanged as 00edb2061f and confirmed completion in the thread's final message #32; the acceleration and template fixes had landed, while Robert's test split from #8 had no demonstrated landing or further revision in this thread.