Subqueries using count(*) and + inferred as boolean instead of string

[karlhorky]

Describe the bug

Usage of summed count(*) subqueries (using + operator) has yielded string in versions before @ts-safeql/[email protected], with 3.6.1 this instead yields boolean:

// 💥 Query has incorrect type annotation.
//	Expected: { employee_count: string; }
//	Actual: { employee_count: boolean; }[]
await sql<{ employee_count: string }[]>`
  SELECT
    (
      SELECT count(*) FROM caregivers
    ) + (
      SELECT count(*) FROM caregiver_assistants
    ) AS employee_count
`;

Workaround

Cast the expression to text in the query:

// ✅
await sql<{ employee_count: string }[]>`
  SELECT
    (
      (
        SELECT count(*) FROM caregivers
      ) + (
        SELECT count(*) FROM caregiver_assistants
      )
    )::text AS employee_count
`;

To Reproduce

See above

Expected behavior

Inference of string, because pg_typeof() reports bigint, which is a string with Postgres.js

Screenshots

--

Desktop (please complete the following information):

• OS: macOS Sequoia 15.2 (24C101)
• PostgreSQL version 14.13
• Version 3.6.1

Additional context

Originally introduced in this PR:

• fix: handle AExpr and NullTest nodes in SQL parsing #354

[Newbie012]

Oh this is bad... pushing a fix

[karlhorky]

@Newbie012 Thanks for #359!

I can confirm that @ts-safeql/[email protected] correctly infers string now:

// ✅
await sql<{ employee_count: string }[]>`
  SELECT
    (
      SELECT count(*) FROM caregivers
    ) + (
      SELECT count(*) FROM caregiver_assistants
    ) AS employee_count
`;

[karlhorky]

However, some new regressions:

• Regression: inference of subquery aliased field name and type (number instead of boolean) #362
• Regression: Incorrect inference on field names in CTE #363

And one previously-fixed regression has appeared again:

• Regression: Some JsonbAgg types returning unknown #353