model-writtenrun 1997 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
Jun - Aug 2025 · 15 entries
- Peter Eisentraut opened the thread by questioning why PostgreSQL 18 needed --with-data when dumping data was already the default, then highlighted the surprising rejection of --no-data with --with-data while --schema-only could be overridden by --with-data #1 #3.
- Robert Haas proposed removing redundant options, while Fujii Masao reported that testing had exposed the same surprising option combinations; Jeff Davis supported changing the interface provided the functionality remained available #4 #5 #8.
- Nathan Bossart and Robert Haas questioned retaining --statistics-only as redundant, but Corey Huinker defended its convenience for copying production statistics into QA databases, and Fujii Masao and Greg Sabino Mullane also favored retaining it #6 #11 #15 #21 #23.
- Jeff Davis resisted introducing order-sensitive option handling because existing pg_dump combinations already raise errors, while Daniel Gustafsson argued against adding another naming convention and Jeff proposed shortening --with-statistics to --statistics #20 #25 #26 #27.
- The discussion reopened whether statistics should be dumped by default: Corey Huinker favored enabling them everywhere, Nathan Bossart found the earlier consensus inconclusive, and Tom Lane supported the simplification this could enable #32 #33 #35.
- Jeff Davis proposed a narrower compromise that preserved existing defaults, removed redundant data and schema switches, retained exclusion switches, and rejected conflicting -only/--with-* combinations; Robert Haas and Greg Sabino Mullane continued to favor statistics being omitted from ordinary dumps by default #44 #45 #48.
- Fujii Masao outlined smaller option sets with different switches for dumping and restoring, but Jeff Davis preferred consistent interfaces and retaining both statistics switches to permit a future default change; by late July Jeff judged that enabling statistics by default lacked consensus #49 #55 #60.
- Jeff Davis submitted conflict validation in #58, and Álvaro Herrera's request to keep option names outside translatable strings produced the lightly revised submission in #61 #59.
- Álvaro Herrera proposed replacing the growing switch combinations with --include=schema,data,statistics; Jeff Davis raised alias and repeated-option questions, Álvaro favored allowing only one --include, and Corey Huinker remained concerned about differing defaults between tools #62 #63 #64 #65.
- Jeff Davis committed the conflict-validation patch from #61 unchanged apart from rebasing as 60121890f7, with a backport to PostgreSQL 18.
- Building on the committed validation, Jeff Davis submitted #66 to remove --with-data and --with-schema and rename --with-statistics to --statistics, retaining both --statistics and --no-statistics for interface consistency and possible future default changes.
- Jeff Davis required a concrete patch before pursuing --include further, while Corey Huinker argued that its verbosity would be a regression with unchanged defaults, leaving that redesign outside the submitted work #67 #68.
- Nathan Bossart requested alphabetical documentation ordering, and Fujii Masao approved the simplification while suggesting a separate cleanup of pg_dump's option handling #69 #70.
- Jeff Davis committed #66 as a3e8dc1438 with only the documentation ordering correction, backported it to PostgreSQL 18, and announced that the Open Item was closed #71.
- The thread ended with both submitted changes landed: statistics defaults remained unchanged for lack of consensus, --include remained unimplemented without a patch or agreed details, redundant statistics switches were deliberately retained, and Fujii Masao's option-handling cleanup was deferred to PostgreSQL 19 #71 #72.