model-writtenrun 5451 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
May 2021 · 1 entry
- Andrew Dunstan discovered that RPM packages exposed generic Perl test-module names and proposed a PostgreSQL-specific namespace in #1; Tom Lane supported waiting for PostgreSQL 15 in #2, while Devrim GÜNDÜZ applied the immediate packaging workaround in #3.
Aug - Oct 2021 · 6 entries
- When development reopened, Andrew proposed PgTest and Utils in #4, Tom suggested replacing the vague Node name with Cluster in #5, and Mark Dilger advocated PostgreSQL::Test to distinguish official modules from existing third-party namespaces in #9.
- Michael Paquier asked whether SSLServer.pm and RewindTest.pm should also move in #8, but Andrew explained that these private helpers lacked the packaging and extension-testing rationale of src/test/perl in #13, leaving their module names unchanged.
- Robert Haas argued for a hierarchical namespace in #21, which Andrew adopted as Postgres::Test in the first patch #22; Noah Misch pointed out that PL/Perl already used PostgreSQL:: in #23, prompting the consistent spelling in #24.
- Michael questioned PostgreSQL::Test::PostgresVersion in #25, Andrew proposed PostgreSQL::Version to clarify what was being versioned in #27, and Noah expressed reservations about occupying a potentially useful public name in #28; Andrew nevertheless included that move alongside the refreshed migration in #30.
- Erik Rijkers found broken installation references and supplied a temporary deletion that allowed check-world to pass in #31; Michael corrected the proposed remedy to install PostgreSQL/Version.pm in #32.
- Andrew committed both migration patches from #30, incorporating the installation correction and light cleanup, as b3b4d8e68a, and announced the landing in #34.
Apr 2022 · 5 entries
- Six months later, Andres Freund reported that the rename made backpatching TAP tests substantially harder in #36; Tom suggested backpatching the rename in #37, while Andres considered retaining old modules alongside the new implementation in #38.
- Noah offered a translation script as an interim workaround in #39, Tom explained that parallel implementations would not solve additions to existing tests in #43, and Michael warned about duplicate maintenance and testing burdens in #48.
- Andrew instead proposed wrappers and per-symbol aliases over the existing stable-branch modules, plus an optional constructor adjustment, in #50; Andres endorsed the approach in #51, and Andrew fixed the wrapper filename and exports in #52.
- Michael raised the risk of forgetting newly added symbols in the alias list in #53, but treated it as nonblocking in #55; Andrew proposed omitting functions absent from each older branch in #56, and Michael approved in #57.
- The compatibility portions of #52 landed for branches 10 through 14 as b235d41d96, omitting the optional constructor patch and unavailable pump_until support; Andrew announced completion in #58.
Jun 2022 · 4 entries
- Noah returned with a fix for incomplete alias lists in #60, replacing individual aliases with whole package symbol tables and reporting that converted tests passed check-world on the PostgreSQL 14 branch; Andrew endorsed the approach in #61.
- Testing the PostgreSQL 10 branch exposed incompatible constructor dispatch, so Noah revised the heuristic to inspect arguments and preserve subclass dispatch in #62, while explicitly noting that subclass permutations had not been tested.
- Noah committed the revised symbol-table patch from #62 with only comment and formatting changes as 20911775de.
- At the thread's last message, Noah reported that the landing broke buildfarm members lapwing and wrasse and said he would investigate #63; the subsequent candidate 4b0d21b06c fixes Perl 5.10.1 crashes by aliasing before package loading and removing explicit package declarations, with no attached patch or later confirmation in this thread.