Fix: SQL Server max_rows not enforced across UNION/INTERSECT/EXCEPT

[tianzhou]

Summary

• SQLRowLimiter.applyTopToQuery injected TOP into only the query's first SELECT, so for set-operator queries (UNION [ALL], INTERSECT, EXCEPT) it capped that first branch's rows, not the combined output — the remaining branches' rows all passed through uncapped.
• When a set operator is detected, the whole statement is now wrapped in a subquery and TOP is applied to the outer result set instead, matching how the other connectors already enforce max_rows on wrapped queries.

Changes

• src/utils/sql-row-limiter.ts: add hasSetOperator, and use it in applyTopToQuery to wrap set-operator queries (SELECT TOP N * FROM (...) AS subq) instead of injecting TOP into the first SELECT.
• src/utils/__tests__/sql-row-limiter.test.ts: add coverage for the UNION ALL repro from the issue, plain UNION, INTERSECT/EXCEPT, semicolon preservation, and a false-positive check for union appearing inside a string literal.

Closes #387

Test plan

• pnpm test:unit — all 932 tests pass
• New tests reproduce the issue's exact probe query and confirm the fix
• Not re-verified against a live SQL Server container (unit coverage of the row-limiter logic was judged sufficient)

[Copilot]

Pull request overview

This PR fixes SQL Server max_rows enforcement for set-operator queries by avoiding TOP injection into only the first SELECT branch and instead applying a cap to the combined result set.

Changes:

• Add set-operator detection (UNION [ALL], INTERSECT, EXCEPT) and wrap these statements so TOP applies to the overall output.
• Add unit tests covering the UNION ALL repro from #387 and additional set-operator/semicolon/string-literal cases.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 3 comments.

File	Description
src/utils/sql-row-limiter.ts	Adds set-operator detection and changes SQL Server TOP application strategy to cap combined set-operator results.
src/utils/__tests__/sql-row-limiter.test.ts	Expands unit tests to cover SQL Server set-operator cases and edge conditions.

[Copilot]

Pull request overview

Copilot reviewed 4 out of 4 changed files in this pull request and generated no new comments.

Suppressed comments (2)

src/utils/sql-row-limiter.ts:124

• hasSetOperator can false-positive on identifiers like SQL Server variables (e.g. @union) because \bunion\b matches when preceded by @. That would incorrectly wrap non-set-operator SELECTs in a derived table and can change behavior or break otherwise-valid statements. Consider requiring a valid operator boundary (start/whitespace/)) before treating UNION/INTERSECT/EXCEPT as set operators.

    const blankedSQL = blankCommentsAndStrings(sql, "sqlserver");
    let found = false;
    this.scanTopLevel(blankedSQL, /\(|\)|\bunion\b|\bintersect\b|\bexcept\b/gi, () => {
      found = true;
    });

src/utils/sql-parser.ts:257

• blankCommentsAndStrings builds the output via repeated result += ... concatenations, which can become O(n^2) for long SQL strings. Since applyTopToQuery now calls this on every SQL Server SELECT, switching to an array builder (like stripCommentsAndStrings) avoids quadratic behavior.

export function blankCommentsAndStrings(sql: string, dialect?: ConnectorType): string {
  const scanToken = getScanner(dialect);
  let result = "";
  let i = 0;