model-writtenrun 2399 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
Oct 2024 - Apr 2025 ยท 27 entries
- Amul Sul proposed SQL-standard NOT ENFORCED support for CHECK and foreign-key constraints in #1, choosing to omit, drop and recreate foreign-key triggers while leaving ALTER CHECK enforcement for later work.
- Joel Jacobson tested the initial series and supported preserving constraint identity while recreating triggers in #2, and Andrew Dunstan endorsed avoiding trigger enable/disable interactions in #3.
- Peter Eisentraut's initial review in #7 led to corrected LIKE behavior, unsupported-attribute checks, catalog documentation and trigger-OID initialization in #8 and #9.
- Jian He's validation tests in #10 and Peter's response in #11 led Amul to make unenforced constraints invalid and reject VALIDATE on them in #12; Peter then requested consistent positive field names, information_schema support and better inheritance handling in #13.
- Alvaro Herrera established that a child may enforce an unenforced parent's CHECK constraint but cannot weaken an enforced parent in #19, and Amul implemented those rules in #21.
- Jian's parser and ADD COLUMN reports in #18 and #22 drove fixes through #25, which Jian accepted in #26; Triveni N also reported testing inheritance, partitioning, foreign keys and backup tools in #28.
- Peter committed the CHECK component of #25 as ca87c415e2, with documentation and test edits announced in #29.
- Peter next committed the unchanged recursion-refactoring component of #25 as 7a947ed25b and requested the remaining foreign-key series in #31.
- Amul resumed the remainder in #33, adding partition-attachment refactoring and an optional patch to merge differently enforced constraints, then added an OID-argument prerequisite in #34 and removed the separately landed NOT VALID partitioned-foreign-key dependency in #38.
- Peter argued in #35 and #37 that switching to ENFORCED must validate existing rows, while Ashutosh Bapat proposed preserving the previous validation state across enforcement toggles in #39 and #44.
- Jian found that ALTER DEFERRABLE unintentionally enabled enforcement in #41, prompting Amul's new AlterConstraintStmt representation and explicit attribute-selection flags in #50.
- Amul implemented independent enforcement and validation states in #55, including a patch relaxing the already committed invariant that unenforced constraints must be invalid.
- Alvaro and Isaac Morland challenged the safety and meaning of valid-but-unenforced constraints in #56, #57 and #59; after consulting standards archives, Peter concluded in #63 that NOT ENFORCED means disabled and proposed three meaningful states.
- Amul renamed the new node to ATAlterConstraint in #69, and Alvaro posted a standalone rewrite in #70 before committing a further restructuring as 80d7f99049, jointly authored with Amul.
- Amul rebased the remaining feature work in #73 and added a prerequisite restoring ATExecAlterChildConstr(), which Alvaro's committed restructuring had removed.
- Alexandra Wang reviewed and ran make check on every component in #76, with corrections and integration changes following in #77.
- Peter committed three components of #77 as 64224a834c, 1d26c2d2c4 and dabccf4513, restoring child recursion, splitting partition attachment and moving inherited-constraint removal earlier.
- Alvaro objected to the recursion structure in #78, and Peter discussed possibly reverting the restored helper in #79 while also rejecting the retained four-state patch and requesting documentation and test improvements.
- Amul's substantial rework in #80 introduced separate attribute handlers and a shared catalog-update helper, removed the four-state patch, and made enabling an unenforced constraint validate its data; Alvaro accepted the simpler layout in #81.
- After comment polishing in #82, Peter committed its three prerequisites as a3280e2a49, 639238b978 and ef7a5af77d, renaming the catalog helper, restoring the recursion changed guard and adjusting formatting.
- The new split in 639238b978 absorbed the restored child-helper structure; discussion in #83, #84 and #86 established that an older comment about independently altered child constraints was stale.
- Ashutosh reviewed the remaining regression tests in #87, requesting meaningful nonempty validation cases and stronger trigger checks, while Peter extracted the convalidated display change from #82 into 5d5f415816.
- Amul posted the remaining series with test improvements in #88, while Peter independently posted a rebase in #89 and questioned the complexity of automatically merging constraints with different enforcement settings.
- After Amul explained the duplicate-constraint alternative in #90 and Alvaro objected to it in #91, Amul added an explicit enforcement-mismatch error in #92, keeping automatic merging as a separable optional patch.
- Ashutosh suggested masking generated trigger OIDs in #93, and Amul replaced count-only checks with explicit trigger listings in #94.
- Peter committed the trigger-discovery and foreign-key feature portions of #94 as eec0040c4b, omitting automatic merging, fixing expected trigger names and adding an optimizer filter for unenforced foreign keys, as explained in #95.
- At the thread's end in #96, Amul accepted the optimizer choice and thanked Peter, although he could not reproduce the reported expected-output failure; the core CHECK and foreign-key features had landed, automatic differing-enforceability merging was left out, and ALTER CHECK enforcement remained outside this series.