model-writtenrun 3611 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Peter Eisentraut opened the discussion by showing that CREATE TABLE rejected differing inherited compression settings although ALTER TABLE allowed them, and proposed inheriting defaults, allowing explicit child overrides, leaving attached children unchanged, and requiring overrides for conflicting parents #1; Ashutosh Bapat agreed, and Peter clarified that unresolved parent conflicts would still produce an error #2 #3.
- Ashutosh Bapat found that child storage declarations were ignored because inheritance examined ColumnDef::storage while parsing populated storage_name, and favored removing the former field instead of duplicating conversion logic #4.
- Ashutosh Bapat submitted separate compression and storage patches implementing that approach, noting that storage defaults come from the column type while unspecified compression follows the configuration setting #5; he then added the missing expected output for builds without LZ4 after CI failures #6.
- Peter Eisentraut reviewed the behavior, code, and tests favorably but asked whether pg_dump preserved the resulting structures #7; Ashutosh Bapat reported successful pg_upgrade testing and reasoned that ALTER TABLE ONLY preserved independent child settings #8.
- Peter Eisentraut combined both patches from #6 into 0413a55699, changing conflict-marker checks to pointer comparisons, adding invalid-storage error handling, simplifying compression-string ownership, and polishing documentation and diagnostics, then announced the commit #9.
- After the commit, Tom Lane questioned whether pg_upgrade testing could detect lost settings #10 and demonstrated that ordinary dump/restore failed to recreate a child with conflicting parent storage settings and silently changed another child's storage setting #11.
- Ashutosh Bapat explained that binary-upgrade dumps attach inheritance separately, masking the failure, proposed emitting storage and compression clauses during table creation, and identified a pre-existing restore failure when parent settings diverged after child creation #12.
- Peter Eisentraut reverted the entire change in 74563f6b90 because pg_dump could not reproduce all newly permitted structures, then reopened the commitfest entry and requested better dump/restore coverage #13.
- Ashutosh Bapat started a separate dump/restore testing thread and found that restoring an attached child could replace its unspecified compression with its parent's setting #14; he subsequently identified COMPRESSION DEFAULT and explicit storage clauses for inherited tables as ways to preserve those defaults #15.
- Ashutosh Bapat proposed resolving conflicting parent settings by choosing default compression or type-default storage when the child supplied no override, eliminating the conflict error rather than requiring an explicit override #16.
- Peter Eisentraut declined further changes that development cycle because of the uncovered pitfalls, inadequate tests, and absence of a replacement patch, suggesting reconsideration for PostgreSQL 18 #17; Ashutosh Bapat accepted deferral and asked about marking the entry returned with feedback, leaving the original series reverted and the redesign unimplemented in this thread #18.