fix(editor): make statement splitting dollar-quote aware for PostgreSQL (#1559)

[datlechin]

Summary

Fixes #1559. Running a PostgreSQL script with a DO $$ ... $$; block or a dollar-quoted function body failed with ERROR: unterminated dollar-quoted string at or near "$$".

Root cause

The editor's statement splitter, SQLStatementScanner, splits a buffer on semicolons while skipping over strings and comments, but it had no concept of PostgreSQL dollar-quoting. A semicolon inside DO $$ BEGIN ...; ... END $$; was treated as a statement terminator, so a broken fragment reached the server.

A second splitter, SQLFileParser (used only for SQL file import), already handled dollar-quoting correctly and was gated by dialect. The editor path never got that logic. QueryClassifier.isMultiStatement delegates to the editor splitter, so the MCP, AI, and safe-mode gates miscounted these scripts too.

What changed

• New SqlDollarQuote holds the dollar-quote lexing (opener detection, close matching, identifier rules) as one shared, pure NSString helper.
• SQLFileParser now delegates to it. Behavior is unchanged; its existing tests still pass.
• SQLStatementScanner takes a dialect and treats a dollar-quoted region as opaque when the dialect is PostgreSQL family. A token-start guard keeps $1 positional parameters and a$$ identifiers from being read as openers. Other dialects are unaffected.
• The dialect is threaded from every execution call site (Execute All, Execute Current Statement, EXPLAIN, ClickHouse) and into QueryClassifier.isMultiStatement, so the AI/MCP/safe-mode gates stop miscounting a DO block as multi-statement.

Tests

• New SQLStatementScanner cases: DO $$ ... $$;, tagged $func$, nested different tags, $1 not an opener, a$$ identifier not an opener, dialect gating, and cursor inside a dollar body.
• All existing SQLFileParser, QueryClassifier, and scanner tests pass.
• swiftlint --strict clean.

Not in scope

The two splitters still differ on nested block comments and E'...' escapes, and autocomplete is not dialect-aware. These are separate from this issue and left for a follow-up.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 5a1b2445a1

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Classify PostgreSQL DO blocks before unblocking them

For PostgreSQL, this now makes DO $$ BEGIN DROP TABLE t; END $$; count as a single statement, but classifyTier still only checks top-level prefixes and returns .safe for DO. The AI/MCP execute_query paths rely on this multi-statement guard plus the tier check before running with pre-cleared capabilities, so a destructive operation hidden inside a dollar-quoted DO block can bypass the dedicated confirmation tool and safe-mode write/destructive classification. Please either classify DO blocks conservatively for PostgreSQL or keep a separate non-UI guard for these executable blocks.

Useful? React with 👍 / 👎.