fix: create SQL functions after the later-batch functions they call (#596)

[tianzhou]

Summary

Follow-up to #615 (merged), found while applying the schema attached to #596. Since #615 the schema plans, but pgschema apply then failed:

ERROR: function apni_ordered_nom_synonymy(bigint) does not exist (SQLSTATE 42883)

New functions are emitted in several batches (before function-dependent tables, after them, after the last table batch, after views, after aggregates). topologicallySortFunctions orders functions only within a batch. A LANGUAGE sql function that touches no new table itself but calls a function deferred to a later batch stayed in the first batch, so it was created before its callee — and SQL-language bodies are validated at creation time.

Fix: the transitive "a caller of a held-back function waits too" closure that splitFunctionsCallingAggregates already had is extracted into holdBackCallers and also applied to the table batches and the view batch. Only SQL-language functions move; plpgsql resolves calls at run time, and moving those could push trigger functions past their triggers.

Fixes #596

Test plan

• Folded into testdata/diff/dependency/issue_530_function_table_function_chain: first_is_flagged() references no table but calls x_is_flagged, which is created after table x. Fails before the fix at apply, passes after.
• Ran dependency/, create_function/, create_aggregate/ for both TestDiffFromFiles and TestPlanAndApply.
• Rebased on main (includes fix: resolve managed-schema qualifiers in inlined SQL function bodies (#596) #615): the reporter's 9.3k-line schema applies via pgschema apply and re-plans to No changes detected.

PGSCHEMA_TEST_FILTER="dependency/issue_530" go test -v ./cmd -run TestPlanAndApply
PGSCHEMA_TEST_FILTER="dependency/" go test -v ./internal/diff -run TestDiffFromFiles

🤖 Generated with Claude Code

[greptile-apps]

Confidence Score: 2/5

The PR is not safe to merge because several realistic dependency chains can still create SQL functions or function-dependent tables before their required routines exist.

Findings

1.  Pairwise passes miss callers ▶
2.  Functions move past dependent tables ▶
3.  Recreated-view callers remain early ▶

Summary

This PR extracts transitive SQL-function caller holdback logic and applies it across table and view creation batches so callers follow functions whose creation is deferred.

• Adds caller propagation between the three table-relative function batches.
• Adds caller propagation for functions dependent on newly created views.
• Extends the dependency fixture with a SQL caller of a table-dependent function.
• The current partition sequence still permits callers and dependent tables to be emitted before their prerequisites.

Diagram

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[Independent functions] --> B[Tables depending on functions]
    B --> C[Table-dependent functions]
    C --> D[Final table batch]
    D --> E[Functions after all tables]
    E --> F[Views and view-dependent functions]

    H[Pairwise caller holdback] -. can move a prerequisite past B .-> C
    I[Later partition changes] -. not rechecked .-> E
    J[Recreated-view functions] -. extracted after closure .-> G[Modify phase]

Loading

_{Reviews (1) · Last reviewed commit: "fix: create SQL functions after the late..."}

[Copilot]

Copilot review overview

🟡 Changes recommended

Cross-batch scheduling and imprecise call detection can still emit functions after objects that require them.

Get a fresh assessment by requesting another Copilot review.

Review effort: Balanced
Findings: 3

Open (3)

• Caller promotion misses functions deferred by later partitions · New
• Dependent objects remain in stale batches after caller promotion · New
• Name-only regex incorrectly detects cross-batch function calls · New

What changed in this PR

Improves PostgreSQL function creation ordering when SQL functions call functions deferred to later dependency batches.

Changes:

• Extracts transitive caller deferral into holdBackCallers.
• Applies caller deferral across table and view batches.
• Extends dependency regression fixtures and expected plans.

File	Description
testdata/​diff/​dependency/​issue_530_function_table_function_chain/​plan.txt	Updates human-readable expected plan.
testdata/​diff/​dependency/​issue_530_function_table_function_chain/​plan.sql	Updates expected plan SQL.
testdata/​diff/​dependency/​issue_530_function_table_function_chain/​plan.json	Updates expected JSON plan.
testdata/​diff/​dependency/​issue_530_function_table_function_chain/​new.sql	Adds an indirect SQL-function dependency case.
testdata/​diff/​dependency/​issue_530_function_table_function_chain/​diff.sql	Updates expected migration SQL.
internal/​diff/​diff.go	Adds transitive caller deferral across creation batches.

---

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

[Copilot]

Copilot review overview

🟡 Changes recommended

The newly added view-batch scheduling path still lacks regression coverage.

Get a fresh assessment by requesting another Copilot review.

Review effort: Balanced
Findings: 1

Open (1)

• Add regression coverage for view-dependent function promotion ordering · New

Resolved since last review (3)

• Name-only regex incorrectly detects cross-batch function calls
• Dependent objects remain in stale batches after caller promotion
• Caller promotion misses functions deferred by later partitions

[Copilot]

Copilot review overview

🟢 Approval recommended

The focused ordering fix has targeted unit and integration coverage with no unresolved findings.

Review effort: Balanced
Findings: None

Resolved since last review (1)

• Add regression coverage for view-dependent function promotion ordering