[Tyagiquamar]

Fixes #1096.

Summary

splitQueries() in the upstream SQL test harness only tracked ', " and backticks, backslash escapes outside backticks, and the two comment styles. The server lexer also has dollar quoting and honors backslash escapes inside quoted identifiers, so two constructs were mis-split:

• $$ ... $$ / $tag$ ... $tag$ heredocs (7 upstream files). A ' or " inside the heredoc toggled the splitter's quote state, so the remaining statements were merged into one, and a ; inside the heredoc could split it. Example: 03011_definitive_guide_to_cast was seen as 82 statements instead of 91, 04038_ascii_cjk_tokenizer as 28 instead of 43.
• A backslash-escaped backtick inside a quoted identifier (1 upstream file). `a\`b` closed the identifier early, leaving it open and merging the rest of the file: 04097_tuple_inside_nullable_subcolumn_descendants was seen as 15 statements instead of 18.

Root Cause

src/split-queries.ts hand-scans the SQL with three quote flags and no notion of dollar quoting. It also only entered the escape handling for ' and " (if ((inSingleQuote || inDoubleQuote) && ch === "\\")), while the server lexer's quotedString<>() treats ', " and ` identically. There is no bug in ClickHouse; this is purely the harness's statement splitter disagreeing with src/Parsers/Lexer.cpp about where a statement ends.

Technical Solution

Kept the hand-written scanner (the issue's measurement does not justify adopting the lexer, and 17 upstream files use syntax the lexer rejects) and taught it the two missing rules, mirroring the lexer:

• findHeredocEnd() implements the lexer's $ branch: an opening $, an optional tag of word characters ([A-Za-z0-9_]), and $ again; the heredoc then extends to the next occurrence of the same delimiter. An unterminated delimiter is not a heredoc, and its $ stays an ordinary character. The region is copied verbatim, so quotes, comment markers and semicolons inside it cannot affect the split.
• Backslash escapes are now recognized inside backticks as well, matching quotedString<''>()`.

No parser rewrite, no new dependency, no change to test-hint.ts or the backends.

Resulting statement counts on the files named in the issue now match the official lexer: 03011 82 -> 91, 04038 28 -> 43, 04097 15 -> 18, and 02683/02685/02686/02687/03032 split at the correct boundary. A 150-file sample of tests/queries/0_stateless produces byte-identical output before and after, so no currently allowlisted file changes behavior. Adding the affected tests to upstream-allowlist.txt is deliberately left out: per tests/clickhouse-test-runner/AGENTS.md, allowlist growth is validated separately through the harness against both CI server versions.

Test Plan

New cases in tests/clickhouse-test-runner/__tests__/split-queries.test.ts (heredoc with quotes/backslashes, tagged heredoc, statement boundaries after a heredoc, unterminated $$, $foo not being a heredoc, escaped backtick in a quoted identifier). All four heredoc/backtick cases fail on main and pass with the fix.

Commands run locally (Node 24.22.1, Windows):

cd tests/clickhouse-test-runner
npx vitest run                 # 7 files, 86 tests passed
npx tsc --noEmit               # clean (after building packages/datatype-parser and the rowbinary skill)
npm run lint                   # eslint --max-warnings=0, clean
npx prettier --check src/split-queries.ts __tests__/split-queries.test.ts   # clean

Statement counts old vs new were compared by running both implementations over the upstream files from the issue plus a 150-file sample of tests/queries/0_stateless (ClickHouse master).

No changelog entry: tests/clickhouse-test-runner is a private, unpublished workspace package with no CHANGELOG.md, and no published package changes.

Checklist

• Unit tests covering the fixed behaviour were added
• Changelog: not applicable (private test-harness package, no published API change)
• Docs: not applicable (internal harness behaviour)

[Copilot]

🟢 Approval recommended

No active review issues are reported for the internal SQL statement-splitting changes.

No open issues 🎉

What changed in this PR

Fixes #1096 by aligning the upstream SQL test harness’s statement splitter with ClickHouse’s handling of dollar-quoted strings and escaped backticks. Changes are limited to the internal harness; published client APIs are unaffected.

Changes:

• Preserve $$...$$ and $tag$...$tag$ heredocs without interpreting their contents as statement boundaries.
• Recognize backslash escapes inside backtick-quoted identifiers.
• Add regression tests for heredocs, non-heredoc dollar sequences, and escaped backticks.

File	Description
tests/​clickhouse-test-runner/​src/​split-queries.ts	Adds heredoc detection and backtick escape handling.
tests/​clickhouse-test-runner/​__tests__/​split-queries.test.ts	Adds six regression cases covering the corrected splitting behavior.

_{Review effort: Balanced}