model-writtenrun 4357 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
Nov 2022 - Feb 2023 · 20 entries
- The thread revived a stalled pg_dump LZ4 effort after Justin Pryzby noted the long gap and unresolved issues in #1, and Georgios Kokolatos confirmed continued work in #3.
- Georgios posted a five-part series in #6, separating Meson test setup, shared compression-option parsing, pg_dump preparation, a callback-based compression API, and LZ4 support.
- Michael Paquier took the Meson prerequisite from #6, extended it to missing pg_verifybackup tool settings, and committed 00ae5d6f58, explaining the additional fix in #7.
- Michael retained validation in the common compression routines and consolidated parsing for both pg_receivewal and pg_basebackup in #7; Georgios incorporated that revision in #9, whose parser component became d18655cc03 with added grammar documentation.
- Michael requested separating compression-specification adoption from file-API preparation in #10, producing #11, and made explicitly requested unsupported compression fail rather than silently fall back in #12.
- Georgios restored the accidentally lost default compression for custom and directory archives in #13, which became 5e73a60488 with fuller option documentation and numeric TOC compression output retained.
- The default-compression regression exposed a testing gap: successful restore did not prove data were compressed, so Michael's glob checks and Georgios's pg_restore output checks developed through #15, #17, and #18 into a7885c9bb2.
- After Georgios resubmitted the three remaining components in #20, Justin found portability, error-handling, and compression-level problems in #21, while Michael identified Windows callback-name collisions in #22; the fixes were incorporated in #23 and #28.
- Justin's January review in #30 questioned API naming, structure passing, archive enum storage, and gzip-specific assumptions, then supplied full-series revisions with separate fixups and an LZ4-default CI experiment in #32 and #33.
- Michael argued against changing input-only specifications to pointers without a concrete benefit in #34 and #38, while Georgios continued his own branch with the rebase in #39.
- Tomas Vondra asked for the actual LZ4 implementation to be review-ready alongside its prerequisites in #40; Georgios responded with #41, briefly merged the preparation and API patches in #43, then restored a smaller split and separate uncompressed implementation in #47.
- Justin reported that the already-landed compression-specification change prevented pg_restore from listing compressed archives without compression support in #53; Michael tested the minimal warning restoration in #57, which became 783d8abc3b.
- Georgios initially proposed removing the header warning in #59, but retained it after Michael's feedback in #60 and revised the broader series in #61.
- Further compatibility and old-dump testing appeared in #65 and #67, and Shi Yu's callback-comment and binary-mode review in #68 was incorporated into the rebased remainder #69.
- Tomas prepared commit descriptions in #70, Georgios corrected credits and final comments in #72, and Tomas announced landing the preparation, generic API, and LZ4 components in #73 and #74.
- While preparing zstd support, Justin opened a post-landing cleanup round in #78 and #79, identifying documentation errors, a malformed TAP condition, confusing default-compression code, and an unreachable gzip error check.
- Justin also demonstrated poor compression from treating each narrow row independently and supplied streaming and LZ4-frame prototypes in #81, establishing the basis for the later format rework.
- Georgios reviewed Justin's wording and TAP fixes in #83; the documentation hunks became 6095069b40, while the TAP correction, also carried in #86, became 34ce114374.
- Justin found that empty gzip-compressed large objects produced invalid dumps and proposed earlier initialization in #84; Georgios made the fix standalone and added an empty-object regression test in #86.
- The earlier default-compression discussion in #78 and #82 eventually produced bedc1f0564, which moved default algorithm selection before parsing so pg_dump constructed only one compression specification.
Mar - May 2023 · 16 entries
- The gzip review retained existing gzip naming for consistency, and discussion in #94 and #95 settled on replacing the unreachable fatal check with an assertion; Justin consolidated those changes and function-order cleanup in #96.
- Georgios clarified in #93 that -Z0 already preserved backward-compatible behavior and posted additional test coverage, rather than a behavior change.
- Justin argued in #97 that the LZ4 frame change should happen before release to avoid future archive-format compatibility burdens; he explained existing frame support in #99, and Michael confirmed its use elsewhere in PostgreSQL in #100.
- Alexander Lakhin demonstrated a real signedness error behind a compiler warning in #101, prompting Georgios's narrow fix #102; Peter Eisentraut and Tomas broadened the discussion to inconsistent library return types and API error contracts in #104 and #106.
- Georgios reworked the type fix into consistent API results, shared buffer constants, and improved comments in #113, then adopted boolean callbacks and corrected signed library handling in #118.
- Tomas prepared the final three-part cleanup in #120, and Justin reviewed it and successfully adapted zstd to the API in #121; its components landed as d3b57755e6, f081a48f9a, and d0160ca11e, with only a boolean initializer correction and commit-message editing.
- Tomas explicitly deferred broader error-reporting and comment suggestions in #122 so they would not block the completed type, buffer, and documentation fixes.
- Georgios endorsed the final empty-data patch in #124, and #96 became 00d9dcf5be after rebasing onto the intervening buffer cleanup, as announced in #129.
- Georgios developed Justin's frame prototype into shared LZ4 state and initialization in #125, then incorporated Tomas's naming review into #130; this became 0070b66fef with comment cleanup and consistent state-variable names described in #131.
- Alexander's Valgrind report in #132 exposed missing null termination in LZ4Stream_gets, and Georgios supplied #133; Andrew Dunstan's review in #134 replaced whole-buffer clearing with explicit trailing termination in #135.
- While reviewing that fix, Michael requested a consistent callback contract and discovered corrupt directory-format dumps in #139; Tom Lane reproduced the corruption in #140 and confirmed that custom format worked in #142.
- Georgios traced the corruption to a missing input-pointer advance and supplied multichunk tests in #144, while Tomas independently posted the minimal fix in #145 and suggested direct pointer updates and buffer-size documentation in #146.
- Georgios incorporated that feedback in #147, Michael confirmed the repair in #148, and the patch became 1a05c1d252 with only a header-comment adjustment.
- Georgios separately removed the disputed assertion from the null-termination patch in #150, which landed unchanged as 3c18d90f89.
- The thread records no separate landing for the -Z0 test attachment #93, and broader error-reporting improvements were left for later work in #122.
- At the final message #154, Tomas reported that both remaining pg_dump compression open issues were fixed for beta1 and that he would monitor the buildfarm.