model-writtenrun 5812 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Michael Paquier proposed making tar compression use an explicit method rather than infer it from the compression level, adding gzip tests and separate method selection as groundwork for future LZ4 support in #1.
- Robert Haas asked that the interface accommodate forthcoming server-side compression and questioned whether gzip integrity tests should also verify restored backups in #2; Michael defended the smaller tests as useful initial coverage in #3.
- Georgios Kokolatos endorsed the refactor but found a potentially uninitialized streamer and an incorrect test skip count in #4, both corrected in #5.
- Robert argued that users need both client-side and server-side compression choices and that the algorithm should be the primary setting in #9 and #12, so Michael split the agreed refactor and tests from the disputed interface in #10.
- Michael landed the production-code portion of #10 as 000f3adfdc, adding explicit gzip branch handling in tar_write(), and landed its lightly formatted tests separately as 50e144193c; he confirmed that the options remained unresolved in #13.
- Following Robert's renewed concerns about server-side integration in #14, Michael proposed --client-compress and --compression-level in #15 and posted that remaining work with the old compression switches removed in #17.
- Magnus Hagander proposed putting compression location, method, and optionally level into one option in #18, while Robert objected to losing backward compatibility in #20.
- Álvaro Herrera consolidated the single-option proposal with preservation of existing numeric and short-option forms in #25; David G. Johnston supported compatibility and discussed extensible method options in #26 and #27, and Robert accepted Álvaro's design without an implicit default algorithm in #28.
- Robert proposed implementing client-side METHOD[:LEVEL] parsing first and adding location selection in the server-compression thread in #30, leading Michael to post the backward-compatible implementation and additional tests in #31.
- Robert found misleading method-prefix parsing and requested splitting at the colon in #32, while Álvaro corrected the help syntax in #33; Michael incorporated those changes and method-specific level validation in #35.
- After Robert suggested eliminating duplicate method comparisons in #36, Michael applied that cleanup and committed #35 with light edits as 5c649fe153, announcing completion in #37.
- Robert then pointed out that WalCompressionMethod also described data-directory compression in #38, prompting Michael's separate proposal to rename it DataCompressionMethod and move it into common.h in #39.
- The thread ended with the refactor, tests, and client-side option extension committed, but the enum cleanup still unresolved: Robert noted duplication with CompressionAlgorithm in #40, Michael preferred delaying unification until pg_dump supported LZ4 in #41, and Robert explained an alternative without insisting on it in #42.