fix: batch per-row inspector lookups into the schema queries (#644, #654)

[tianzhou]

Summary

dump and plan were slow over high-latency connections because the inspector did one round trip per privilege row (role name), per constraint column (position in the key), per owned sequence (identity flag) and per view (column list). Each value now comes back from the batched query it belongs to, so introspection is a fixed ~30 queries regardless of schema size.

	before	after
200-table reproduction from #644 (dump, one inspection)	4,028 queries	30 queries
employee fixture with grants to two roles	164 queries	30 queries

Counts from the --debug query tree (#655). plan inspects twice, so the user-facing saving doubles. Dump output is byte-identical before and after on both databases.

Query changes (ir/queries/queries.sql)

• GetPrivilegesForSchema, GetColumnPrivilegesForSchema: the aclexplode rows move into a CTE and LEFT JOIN pg_roles on the grantee; grantee_name is PUBLIC for oid 0 and empty when the role no longer exists, which keeps today's skip behavior.
• GetConstraintsForSchema: constraint_position = array_position(conkey, attnum) for PK / UNIQUE / FK, 0 otherwise. Same value information_schema.key_column_usage.ordinal_position gave.
• GetSequencesForSchema: owned_by_identity from the owning column's attidentity.
• GetViewsForSchema: column_names aggregated from pg_attribute in attnum order.
• GetIndexesForSchema: COALESCE on parent_index / parent_index_schema. The checked-in generated code had drifted from what sqlc v1.30.0 produces for these two columns; with the coalesce, regeneration yields the nullable type the inspector already uses.

Correctness fix

The old position lookup keyed on (schema, constraint name, column) without the table. Foreign key names are only unique per table, so two tables with a same-named FK on the same columns could get each other's column order:

CREATE TABLE p (a int, b int, PRIMARY KEY (a, b), UNIQUE (b, a));
CREATE TABLE c1 (a int, b int, CONSTRAINT fk_p FOREIGN KEY (a, b) REFERENCES p (a, b));
CREATE TABLE c2 (a int, b int, CONSTRAINT fk_p FOREIGN KEY (b, a) REFERENCES p (b, a));

c2 was dumped as FOREIGN KEY (a, b) REFERENCES p (a, b), pointing at the primary key instead of the (b, a) unique constraint. New fixture testdata/dump/issue_644_fk_column_order fails before this change and passes after.

Fixes #644
Fixes #654 (closed as duplicate)

Test plan

• go test ./cmd/dump (full dump fixture suite, including the new TestDumpCommand_Issue644FKColumnOrder; red before the fix)
• go test ./internal/diff -run TestDiffFromFiles
• go test ./ir/...
• PGSCHEMA_TEST_FILTER=<category>/ go test ./cmd -run TestPlanAndApply for privilege/, create_sequence/, create_view/, create_materialized_view/, create_table/
• go vet ./ir/...
• Manual: old vs new binary dump of employee and the 200-table reproduction are identical; query counts as in the table above

🤖 Generated with Claude Code

[Copilot]

🟢 Approval recommended

The queries and inspector consumers are consistent, the ordering fix has regression coverage, and no blocking issues remain.

0 open findings

What changed in this PR

Reduces database round trips during dump and plan by batching inspector metadata lookups, addressing slow remote introspection.

Changes:

• Fetches privilege names, constraint positions, sequence identity flags, and view columns in existing queries.
• Updates generated bindings and removes per-row inspector queries.
• Fixes same-named foreign key column ordering with regression coverage.

File	Description
testdata/​dump/​issue_644_fk_column_order/​raw.sql	Reproduces conflicting foreign key column orders.
testdata/​dump/​issue_644_fk_column_order/​pgschema.sql	Records expected pgschema output.
testdata/​dump/​issue_644_fk_column_order/​pgdump.sql	Provides reference PostgreSQL output.
testdata/​dump/​issue_644_fk_column_order/​manifest.json	Documents the regression fixture.
ir/​queries/​queries.sql.go	Updates generated query bindings.
ir/​queries/​queries.sql	Batches metadata lookups.
ir/​inspector.go	Consumes batched metadata.
cmd/​dump/​dump_integration_test.go	Registers the ordering regression test.

Files not reviewed (1)

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

🧠 Review effort: Balanced

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[greptile-apps]

Confidence Score: 5/5

[High impact] The PR appears safe to merge; no actionable issues were found.

Summary

This PR moves repeated inspector lookups into the existing schema queries.

• Privilege rows now include the grantee name in the schema query.
• Foreign keys keep their own column order, even when names repeat.
• Identity-backed sequences stay managed through their columns.
• View rows now include their ordered column names.
• Missing parent-index names now come back as empty strings.

No actionable issues found. Tests were not run during this review.

_{Reviews (1) · Last reviewed commit: "fix: batch per-row inspector lookups int..." · Reviewed by Greptile}