model-writtenrun 9530 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
Oct - Dec 2021 ยท 31 entries
- Tom Lane opened the thread after a pg_dump optimization broke dumping PostgreSQL 8.0 catalogs, asking whether to revert it or remove support for ancient servers and simplify compatibility code #1.
- David G. Johnston favored dropping legacy support and suggested a ten-release horizon, while Robert Haas emphasized the practical difficulty of upgrading across many releases and building old servers with current libraries #2 #3.
- Tom Lane and Alvaro Herrera reproduced configure and lexer failures on old releases, establishing that compatibility testing itself was becoming difficult #5 #6.
- Robert Haas proposed limited maintenance to make old branches buildable; Tom Lane wanted narrowly scoped fixes and a sunset policy, while Andrew Dunstan offered existing old binaries and discussed buildfarm support #7 #11 #12 #14.
- Alvaro Herrera and Andres Freund favored limited buildfarm coverage, and Andres advocated a predictable compatibility policy instead of periodically debating individual cutoffs; Tom initially preferred the recognizable 9.0 boundary #17 #27 #28 #29 #31.
- Tom posted the sole draft patch, removing roughly a thousand lines of pre-9.0 support from pg_dump and pg_upgrade, but deferred landing until other pg_dump work was committed; he also raised a separate concern about pg_upgrade's float8 passing-convention check #34.
- Andrew Dunstan initially preferred retaining the existing 8.4 pg_upgrade boundary, so the minimum supported version remained under discussion after the draft #37.
- In December, Peter Eisentraut proposed keeping pg_dump and psql compatible with at least ten earlier major releases and maintaining enough old-branch buildability for testing, which made 9.2 the proposed PostgreSQL 15 cutoff #38.
- Andres Freund requested including pg_upgrade and questioned the long maintenance horizon, while Peter and Tom argued that compatibility promises should remain stable throughout a supported major release's lifetime #42 #44 #48.
- Tom's macOS and Fedora experiments found 9.2 readily recoverable but 9.1 substantially harder to build; he proposed including compiler-warning suppression and clean check-world output in the maintenance effort #45 #50.
- Tom narrowed the maintenance goal to minimal changes needed for clean builds and tests with selected configuration options, Robert supported that scope, and Peter accepted the expansion; maintaining separate historical minor releases was not taken on #52 #53 #54 #56 #59.
- Further investigation found lexer and catalog-layout obstacles below 9.2, and Tom settled on restoring buildability from 9.2 onward before removing older-server support from pg_dump and psql; possible libpq removals remained an open question #60 #61 #62.
- The buildability work began with Peter Eisentraut's existing configure correction, backpatched to replace unprototyped exit() calls and give main() an explicit return type 47fca011b8.
- Multibyte conversion checks were backpatched to suppress uninitialized-variable warnings in UtfToLocal() and LocalToUtf() 404d987646 70c64135d0.
- Tom's proposed backpatch of the pg_upgrade test-script cleanup landed separately from the attached compatibility-removal draft, removing shell tracing and explicitly selecting trust authentication to reduce check-world noise #50 119b9d060e a4061eeaf3.
- Compiler-option detection and warning suppression were backpatched for gcc truncation warnings and clang's unused-command-line-argument warnings 276c017f12.
- The ecpg date-conversion function definitions were made consistent with their declarations to avoid newer gcc array-parameter warnings 9ae132dfe3 b30e23fe58 87671beecf.
- PL/Perl's clang macro warnings were suppressed through configure, allowing the selected Perl-enabled testing scope to remain practical c16a78147c.
- Two pgcrypto cleanups reformatted an imath macro and supplied consistent SHA digest-array lengths to eliminate indentation and array-parameter warnings 9d73dac127 116a509e09.
- A missing initialization in RestoreGUCState() was backpatched to silence an additional PostgreSQL 9.5 compiler warning 60ca4eec8d.
- The reorder_function_arguments() warning fix added a negative argument-count check and replaced MemSet with memset across the listed old-branch commits 0af208402d a1fe82d987 265b96a96a.
- Further backpatches enlarged a pgbench formatting buffer and corrected an authentication-parser macro to return NULL instead of false fcf1f27fe8 468773837b.
- For PostgreSQL 9.2, Tom suppressed clang's unreliable sometimes-uninitialized warnings, changed the non-libxml fallback to avoid a null format argument, and added unreachable abort() calls to clarify PANIC control flow for gcc cea316881f 3f9700acb2 985c8de212.
- Tom reported successful minimal-configuration builds and check-world runs back to 9.2 on Fedora and macOS, while leaving pre-v10 parallel check-world reliability, an occasional generated-lexer warning, and older macOS PL/Perl sysroot problems outside the completed work #63.
- The pg_dump and pg_dumpall portion of the draft landed separately with the cutoff raised to 9.2, additional obsolete query branches removed, and database and role configuration dumping simplified with unnest() #34 30e7c175b8.
- After finishing that portion, Tom proposed deleting --no-synchronized-snapshots because its remaining use enabled potentially inconsistent parallel dumps from old standby servers #65.
- The pg_upgrade portion of the draft then landed with the same 9.2 cutoff and additional simplifications for tablespace locations, visibility maps and server startup #34 e469f0aaf3.
- The separate snapshot-option removal landed without an attached revision, making unsupported parallel dumps from old standby servers fail instead of allowing synchronization to be bypassed #65 2a712066d0.
- Justin Pryzby found missed compatibility branches in pg_upgrade's PL/Python and socket handling after the initial commits; Tom traced them to the draft's former 9.0 cutoff and committed a broader cleanup that also removed obsolete pg_dump logic #68 #69 c49d926833.
- The psql cleanup landed separately, removing pre-9.2 compatibility paths and warning at startup when the server is older than the supported cutoff #62 cf0cab868a.
- By the final message, the listed compatibility removals and buildability fixes had landed; Greg Stark added historical experience building much older releases, while the thread recorded no resolution of the separate libpq-removal or float8 passing-convention questions #70 #62 #34.