model-writtenrun 7259 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Justin Pryzby proposed zstd compression for pg_dump for PostgreSQL 16, accompanied by long-distance matching for dumps and base backups and a temporary patch that exercised zstd as the CI default #1.
- Michael Paquier questioned the benefit of long-distance matching, and Justin explained the compression-versus-memory tradeoff and why the existing basebackup proposal was included in this series #2 #3.
- Euler Taveira supplied the missing cpluspluscheck exclusion for the new header; Justin incorporated it while fixing build issues, refactoring zstd routines, updating documentation, and bundling prerequisite LZ4 fixes #5 #6.
- Jacob Champion found that ordinary zstd compressed his sample better than gzip with much less CPU time, but long-distance matching brought little benefit; Justin supplied contrasting datasets and declined to expose window-size tuning because larger windows could require decompressor configuration #7 #8 #10.
- Jacob requested parameter names in unsupported-option errors, and Justin incorporated that improvement alongside buffer-allocation, initialization, cleanup, and test-environment fixes while rebasing over the LZ4 fixes #9 #11.
- Jacob's Windows tests produced correct restored output, and further testing with Meson exposed missing dependencies and prompted detailed feedback on suffix detection, compression capability checks, namespace use, buffer handling, and errors; Justin addressed much of that feedback in the next revision #12 #13 #14.
- Tomas Vondra explained why library availability could not replace pg_dump-specific capability checks and argued for preserving explicit compressed-suffix handling; Justin restored supports_compression() while adapting the patch to the revised compression I/O API #15 #17.
- Jacob confirmed the revised Meson build worked but highlighted uncertainty about shared archive state under Windows threads, and Tomas recommended disabling zstd workers until their interaction with parallel dumps was understood #18 #21.
- Justin split threaded workers into a future-feature attachment, retained a warning when workers were requested, expanded temporary CI testing, and added .lz4/.zst suffix checks for consistency with gzip #24.
- Tomas and Justin agreed to defer worker threads until PostgreSQL 17 because testing was limited and their benefit remained uncertain; they also established that long-distance matching required no output-format change #28 #29 #30.
- Tomas requested that long_distance use bool, and Justin submitted the final series with that change plus comment, indentation, assertion, and warning-message cleanup #31 #32.
- Tomas committed the main zstd component from the final submission as 84adc8e20f, with only comment and formatting edits and a revised commit message #32 #33.
- Tomas then committed the matching long-distance component as 2820adf775, changing its commit message and declaring the PostgreSQL 16 work complete; although the work originated in thread #42349, the committed version came from this thread's final submission #32 #34.
- Threaded workers remained deferred and the CI-only default change was not landed; Justin was willing to wait for experience with PostgreSQL 16, and Tom Lane closed the discussion by asking that the commitfest entry be marked committed #35 #36.