model-writtenrun 5621 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Georgios Kokolatos proposed LZ4 support for pg_dump and pg_restore in #1, first unifying archive output through cfp and separating compression method from level, then adding LZ4 streaming and file wrappers; he credited Rachel Heaton with testing, debugging, and test expansion.
- Michael Paquier asked to separate the existing gzip coverage improvements and replace method-specific conditionals with callbacks and separate implementation files in #2, while Justin Pryzby raised the possibility of LZ4 high-compression interfaces in #3.
- Justin Pryzby explained that the LZ4 feature targeted the next development cycle in #5, and Rachel Heaton posted Georgios's expanded four-patch series in #6, separating tests and dead tar-compression cleanup while fixing explicit --compress=none behavior; Georgios clarified in #7 that the callback redesign remained deferred.
- Michael Paquier identified the tests and tar cleanup as independently committable in #8, and Justin Pryzby confirmed the tar code's unsupported status in #9 before reporting old-LZ4 compatibility problems, rejected -Z0 input, excessive conditionals, and opportunities to share compression specifications in #10 and #11.
- Robert Haas advocated reusing common compression parsing and specifications in #14, #17, and #21, and Michael Paquier agreed after resolving his concern about client/server-specific parsing in #23.
- Michael Paquier's review in #19 found a missing GZIP_PROGRAM export and nonportable gzip -k usage, which Georgios corrected in #20 while explaining that the tests needed decompressed SQL output rather than merely an archive-validity check.
- Michael Paquier extracted the tar/compression-option test from #20 into edcedcc2c7, adding a forced-error option for builds without zlib, and requested additional blob-path cleanup and test documentation in #22.
- Georgios addressed those requests in #24, and Michael Paquier committed its tar-cleanup component unchanged as 8ac4c25a05, confirming the removal of the unused gzip paths in #25.
- The remaining discussion separated library availability from external utility availability, and Georgios's reduced series in #26 represented compression commands as a program plus arguments while retaining the outstanding refactoring and LZ4 patches.
- Michael Paquier supplied the focused test revision in #27, introducing compile_option to skip unsupported command runs and consolidating directory tests; Georgios accepted it in #28, and it landed with comment and formatting edits as 98fe74218d.
- On February 23, 2023, Tomas Vondra committed the archive-output portion of #26 as 03d02f54a6, retaining the cfp conversion while adapting it to shared compression specifications and centralized support checks.
- Tomas Vondra then committed e9960732a9, substantially reworking the abstraction and archive-header portions of #26 into callback APIs and separate gzip/uncompressed implementations, with the archive header storing the compression algorithm instead of its level.
- The LZ4 component of #26 landed as 0da243fed0, adapted to the new callbacks in compress_lz4.{c,h} and incorporating compatibility fixes; the supplied thread contains the earlier implementation and design reviews but no attachment of these final 2023 revisions.
- Tom Lane reopened the testing discussion in October 2025 in #31, explaining that the tests introduced by 98fe74218d compressed blobs.toc but missed toc.dat, leaving zero-length gzip reads and the getc callback untested.
- At the thread's end, Michael Paquier agreed in #32 that both TOC files should be manually compressed and that equivalent coverage should extend to LZ4 and Zstandard; he reported the underlying bug already fixed elsewhere, but the supplied thread ends without an attached or committed coverage extension.