validate_query returns valid: true for format() placeholders inside RAISE EXCEPTION

[eberns1995]

Summary

validate_query currently reports valid: true for a PL/pgSQL block that should fail at runtime because format() has more %s placeholders than arguments inside a RAISE EXCEPTION expression.

I hit this while evaluating dryrun v0.8.0 as a migration-safety guardrail. This is reproducible through the MCP validate_query tool once a schema is loaded.

Minimal repro

Run dryrun MCP with any loaded schema, then call validate_query with:

DO $$
BEGIN
  RAISE EXCEPTION 'bad format output: %', format('value %s %s', 'one');
END
$$;

Observed response:

{
  "valid": true,
  "errors": null,
  "warnings": null,
  "_meta": {
    "hint": "Query is valid. Use advise if you need optimization suggestions.",
    "mode": "offline",
    "pg_version": "18.3.0"
  }
}

Expected behavior

I would expect validate_query to flag this as invalid, or at least warn that PL/pgSQL expression bodies / format() placeholder arity are not validated.

PostgreSQL will reject the inner format('value %s %s', 'one') call at execution time because the format string expects two arguments but receives one. In migrations, this class of issue can hide inside DO $$ ... $$ or function bodies and only surface when the migration executes.

Environment

• dryrun version: v0.8.0
• PostgreSQL version in loaded schema metadata: 18.3.0
• Tool path: MCP validate_query

Thanks for building this. The schema linting and migration-lock checks were useful; this issue is specifically about the PL/pgSQL/body-validation boundary.

[radim]

Thank you @eberns1995 - this is genuine gap and you've pinpointed it perfectly. validate_query now does full query parse. But the limitation is that it stops as procedural-body and marks it as valid. PL/pgSQL is never parsed as it is now.

As immediate 'workaround' I will prepare a fix to emit warning about this.

The proper fix (or better to call this feature) is going to be bit more complicated:

• This is closely relevant to support for functions/procedures #1
• I might somehow integrate https://github.com/okbob/plpgsql_check - but that needs local development PostgreSQL instance being available, which is now blocked by cleanup of No MCP tool to use production database connection #24 / or I need to consider giving MCP ability to work with local docker instances

All this is on my roadmap, but at the moment can't promise ETA (definitely not for v0.9)