fix: skip partition-clone child triggers in dump (#472)

[tianzhou]

Summary

A FOR EACH ROW trigger on a partitioned table causes PostgreSQL to clone the trigger onto every partition child, creating one extra pg_trigger row per partition with tgparentid != 0 (and tgisinternal = false). pg_dump never emits these clones — only the top-level trigger on the partitioned parent (tgparentid = 0).

pgschema's trigger inspector query filtered only tgisinternal, so the clones slipped through and were dumped as standalone CREATE OR REPLACE TRIGGER ... ON <child> statements. When the child partitions were excluded via .pgschemaignore, the resulting desired-state file failed pgschema's own plan with relation "<child>" does not exist.

This is the trigger analog of the internal per-partition FK constraint problem fixed in #409/#460. The fix adds AND t.tgparentid = 0 to the trigger query in ir/queries, matching pg_dump and the existing FK-constraint treatment. tgparentid exists since PostgreSQL 13, so all supported versions (14–18) are covered.

Verified against a live PostgreSQL 17/18 catalog: the parent trigger has tgisinternal=f, tgparentid=0; the cloned child trigger has tgisinternal=f, tgparentid!=0.

Fixes #472

Test plan

Added dump regression testdata/dump/issue_472_partition_clone_trigger/ (partitioned ledger with a row-level trigger). Before the fix, dump emitted a duplicate trg_rollup on ledger_2026_06; after, only the parent trigger is emitted.

go test ./cmd/dump -run TestDumpCommand_Issue472PartitionCloneTrigger
PGSCHEMA_TEST_FILTER="create_trigger/" go test ./internal/diff -run TestDiffFromFiles

Both pass; go build ./... clean.

🤖 Generated with Claude Code

[greptile-apps]

Greptile Summary

This PR fixes two related trigger-inspection bugs: (1) partition-clone child triggers (tgparentid != 0) were leaking into dump output because the query only filtered tgisinternal, causing bogus CREATE TRIGGER statements on child partitions and plan failures when those partitions were ignored; (2) pg_get_triggerdef was called without a forced search_path, so WHEN-clause type casts in non-public schemas rendered schema-qualified in the current state but unqualified in the desired state, producing a perpetual recreate loop.

• Adds AND t.tgparentid = 0 to GetTriggersForSchema, directly mirroring the conparentid = 0 guard already in the FK-constraint query (PR skip constraints for foreign keys referencing ignored tables (.pgschemaignore) #409/fix: skip internal per-partition FK constraints in dump (#409) #460).
• Wraps pg_get_triggerdef in a LEFT JOIN LATERAL with set_config('search_path', quote_ident(n.nspname), true), matching the established pattern used by GetRLSPoliciesForSchema and GetConstraintsForSchema.
• Adds a dump regression test for issue Partition-clone child triggers (tgparentid != 0) are dumped even when child tables are ignored; qualified [triggers] patterns do not match #472 and an idempotency sub-test for issue main: repeat plan drift for trigger WHEN expression with same-schema enum casts #481, plus syncs the sequence_comment column that was already live in the generated Go file back into the source .sql file.

Confidence Score: 5/5

Safe to merge — both fixes are narrowly scoped catalog filter/rendering changes with no destructive side effects, and regression tests cover both scenarios.

The tgparentid = 0 guard and the set_config LATERAL pattern are direct analogs of identical fixes already merged for FK constraints and RLS policies, so the approach is well-proven. Test coverage is thorough: a dump regression for the partition-clone case and an idempotency round-trip for the search_path case. The sequence-comment sync is a housekeeping correction with no behavioral change. No logic in the query result pipeline was removed or restructured.

No files require special attention.

Important Files Changed

Filename	Overview
ir/queries/queries.sql	Two fixes: adds AND t.tgparentid = 0 to exclude partition-clone child triggers, and wraps pg_get_triggerdef in a LATERAL with set_config for deterministic search_path. Also syncs a pre-existing sequence_comment column that was already live in the generated .sql.go file. All patterns match existing constraint/RLS-policy precedents.
ir/queries/queries.sql.go	Regenerated sqlc output reflecting the trigger query changes; the GetSequencesForSchema section was already correct before this PR (sequence_comment column was present), so only the trigger portion changes in the diff.
cmd/dump/dump_integration_test.go	Adds TestDumpCommand_Issue472PartitionCloneTrigger integration test following the standard runExactMatchTest pattern; short-mode skip is correctly guarded.
cmd/issue_449_integration_test.go	Adds a third idempotency sub-test (trigger_when_same_schema_enum_cast) covering issue #481 — trigger WHEN clauses with same-schema enum casts that previously caused perpetual recreations. Test is self-contained and follows the existing applyThenReplan pattern.
testdata/dump/issue_472_partition_clone_trigger/manifest.json	New regression test manifest for issue #472 — description and notes accurately document the partition-clone trigger scenario.
testdata/dump/issue_472_partition_clone_trigger/pgdump.sql	Reference pg_dump output; SET transaction_timeout = 0 is correctly commented out to preserve compatibility with PostgreSQL 14–16, following the documented convention.
testdata/dump/issue_472_partition_clone_trigger/pgschema.sql	Expected pgschema dump output — only the parent trigger trg_rollup ON ledger appears; no duplicate on ledger_2026_06, confirming the fix.
testdata/dump/issue_472_partition_clone_trigger/raw.sql	Minimal reproduction schema: partitioned ledger with one child partition and one FOR EACH ROW trigger — exactly the setup needed to reproduce the cloned-trigger bug.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[GetTriggersForSchema query] --> B{t.tgisinternal?}
    B -- true --> C[Skip - internal trigger]
    B -- false --> D{t.tgparentid = 0?}
    D -- "false: non-zero parent" --> E[Skip - partition-clone child added by this PR]
    D -- "true: top-level trigger" --> F[LATERAL: set_config search_path to trigger schema added by this PR]
    F --> G[pg_get_triggerdef with deterministic schema context]
    G --> H[Emit trigger row]

Loading

%%{init: {'theme': 'base', 'themeVariables': {"darkMode": true, "background": "#0d1117", "primaryColor": "#21262d", "primaryTextColor": "#e6edf3", "primaryBorderColor": "#8b949e", "lineColor": "#8b949e", "textColor": "#e6edf3", "edgeLabelBackground": "#161b22", "actorBkg": "#21262d", "actorBorder": "#8b949e", "actorTextColor": "#e6edf3", "actorLineColor": "#8b949e", "signalColor": "#8b949e", "signalTextColor": "#e6edf3", "noteBkgColor": "#373320", "noteBorderColor": "#d4a72c", "noteTextColor": "#f0e6c0", "labelBoxBkgColor": "#21262d", "labelBoxBorderColor": "#8b949e", "labelTextColor": "#e6edf3", "loopTextColor": "#e6edf3", "activationBkgColor": "#30363d", "activationBorderColor": "#8b949e"}}}%%
flowchart TD
    A[GetTriggersForSchema query] --> B{t.tgisinternal?}
    B -- true --> C[Skip - internal trigger]
    B -- false --> D{t.tgparentid = 0?}
    D -- "false: non-zero parent" --> E[Skip - partition-clone child added by this PR]
    D -- "true: top-level trigger" --> F[LATERAL: set_config search_path to trigger schema added by this PR]
    F --> G[pg_get_triggerdef with deterministic schema context]
    G --> H[Emit trigger row]

Loading

_{Reviews (1): Last reviewed commit: "fix: skip partition-clone child triggers..." | Re-trigger Greptile}

[Copilot]

Pull request overview

This PR updates pgschema’s trigger inspection to align with pg_dump by skipping partition-cloned child triggers (pg_trigger.tgparentid != 0), preventing bogus CREATE OR REPLACE TRIGGER ... ON <partition_child> output and downstream failures when partitions are ignored. It also changes how trigger definitions are rendered to be deterministic by forcing search_path before calling pg_get_triggerdef.

Changes:

• Filter trigger inspection to only dump top-level (non-cloned) triggers via t.tgparentid = 0.
• Render trigger definitions under a forced search_path to avoid schema-qualification drift in WHEN clauses (idempotent plan/apply).
• Add a dump regression fixture + integration test for issue #472, plus an integration coverage case for trigger WHEN enum casts.

Reviewed changes

Copilot reviewed 7 out of 8 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
testdata/dump/issue_472_partition_clone_trigger/raw.sql	Adds a minimal repro schema for partition-cloned triggers.
testdata/dump/issue_472_partition_clone_trigger/pgschema.sql	Expected pgschema output for the new regression case.
testdata/dump/issue_472_partition_clone_trigger/pgdump.sql	pg_dump-produced input used to seed the regression database.
testdata/dump/issue_472_partition_clone_trigger/manifest.json	Documents the regression fixture purpose and links issue #472.
ir/queries/queries.sql	Updates trigger query to (a) force deterministic rendering and (b) skip tgparentid != 0 clones. Also includes a sequence query adjustment.
ir/queries/queries.sql.go	Regenerates sqlc output to match the updated SQL queries.
cmd/issue_449_integration_test.go	Extends repeat-plan idempotency coverage to include trigger WHEN enum casts (issue #481).
cmd/dump/dump_integration_test.go	Adds an integration test runner for the new issue #472 dump fixture.

Files not reviewed (1)

• ir/queries/queries.sql.go: Generated file

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Pull request overview

Copilot reviewed 6 out of 7 changed files in this pull request and generated no new comments.

Files not reviewed (1)

• ir/queries/queries.sql.go: Generated file