fix(core): Track quote state when sanitizing SQL literals

packages/core/src/integrations/postgresjs.ts

@@ -9,7 +9,7 @@ import { SPAN_STATUS_ERROR } from '../tracing';
 import { hasSpanStreamingEnabled } from '../tracing/spans/hasSpanStreamingEnabled';
 import { startSpanManual } from '../tracing/trace';
 import type { Span, SpanAttributes } from '../types/span';
-import { getSqlQuerySummary } from '../utils/sql';
+import { getSqlQuerySummary, sanitizeSqlQuery } from '../utils/sql';
 import { debug } from '../utils/debug-logger';
 import { isObjectLike } from '../utils/is';
 import { getActiveSpan } from '../utils/spanUtils';
@@ -242,7 +242,7 @@ function _wrapSingleQueryHandle(
     }
 
     const fullQuery = _reconstructQuery(query.strings);
-    const sanitizedSqlQuery = _sanitizeSqlQuery(fullQuery);
+    const sanitizedSqlQuery = sanitizeSqlQuery(fullQuery);
 
     const client = getClient();
     const querySummary = getSqlQuerySummary(sanitizedSqlQuery);
@@ -366,58 +366,6 @@ export function _reconstructQuery(strings: string[] | undefined): string | undef
   return strings.reduce((acc, str, i) => (i === 0 ? str : `${acc}$${i}${str}`), '');
 }
 
-let integerLiteralRE: RegExp | undefined;
-
-/**
- * Sanitize SQL query as per the OTEL semantic conventions
- * https://opentelemetry.io/docs/specs/semconv/database/database-spans/#sanitization-of-dbquerytext
- *
- * PostgreSQL $n placeholders are preserved per OTEL spec - they're parameterized queries,
- * not sensitive literals. Only actual values (strings, numbers, booleans) are sanitized.
- *
- * @internal Exported for testing only
- */
-export function _sanitizeSqlQuery(sqlQuery: string | undefined): string {
-  if (!sqlQuery) {
-    return 'Unknown SQL Query';
-  }
-
-  // Lazy init: constructing this at module scope would evaluate the lookbehind
-  // on import and crash Safari <16.4 browser bundles that reach this file via
-  // the core barrel. Building it on first call keeps the cost off the import path.
-  if (!integerLiteralRE) {
-    integerLiteralRE = new RegExp('(?<!\\$)-?\\b\\d+\\b', 'g');
-  }
-
-  return (
-    sqlQuery
-      // Remove comments first (they may contain newlines and extra spaces)
-      .replace(/--.*$/gm, '') // Single line comments (multiline mode)
-      .replace(/\/\*[\s\S]*?\*\//g, '') // Multi-line comments
-      .replace(/;\s*$/, '') // Remove trailing semicolons
-      // Collapse whitespace to a single space (after removing comments)
-      .replace(/\s+/g, ' ')
-      .trim() // Remove extra spaces and trim
-      // Sanitize hex/binary literals before string literals
-      .replace(/\bX'[0-9A-Fa-f]*'/gi, '?') // Hex string literals
-      .replace(/\bB'[01]*'/gi, '?') // Binary string literals
-      // Sanitize string literals (handles escaped quotes)
-      .replace(/'(?:[^']|'')*'/g, '?')
-      // Sanitize hex numbers
-      .replace(/\b0x[0-9A-Fa-f]+/gi, '?')
-      // Sanitize boolean literals
-      .replace(/\b(?:TRUE|FALSE)\b/gi, '?')
-      // Sanitize numeric literals (preserve $n placeholders via negative lookbehind)
-      .replace(/-?\b\d+\.?\d*[eE][+-]?\d+\b/g, '?') // Scientific notation
-      .replace(/-?\b\d+\.\d+\b/g, '?') // Decimals
-      .replace(/-?\.\d+\b/g, '?') // Decimals starting with dot
-      .replace(integerLiteralRE, '?') // Integers (NOT $n placeholders)
-      // Collapse IN clauses for cardinality (both ? and $n variants)
-      .replace(/\bIN\b\s*\(\s*\?(?:\s*,\s*\?)*\s*\)/gi, 'IN (?)')
-      .replace(/\bIN\b\s*\(\s*\$\d+(?:\s*,\s*\$\d+)*\s*\)/gi, 'IN ($?)')
-  );
-}
-
 /**
  * Returns connection context attributes.
  *

packages/core/src/server-exports.ts

@@ -29,14 +29,17 @@ export type {
 /* oxlint-enable typescript/no-deprecated */
 export {
   instrumentPostgresJsSql,
-  _sanitizeSqlQuery as _INTERNAL_sanitizeSqlQuery,
   _reconstructQuery as _INTERNAL_reconstructPostgresQuery,
   _buildConnectionContext as _INTERNAL_buildPostgresConnectionContext,
   _getConnectionAttributes as _INTERNAL_getConnectionAttributes,
   _getOperationName as _INTERNAL_getPostgresOperationName,
 } from './integrations/postgresjs';
 export type { PostgresConnectionContext } from './integrations/postgresjs';
-export { getSqlQuerySummary as _INTERNAL_getSqlQuerySummary } from './utils/sql';
+export {
+  getSqlQuerySummary as _INTERNAL_getSqlQuerySummary,
+  sanitizeSqlQuery as _INTERNAL_sanitizeSqlQuery,
+} from './utils/sql';
+export type { SqlDialect } from './utils/sql';
 
 export { patchHttpModuleClient } from './integrations/http/client-patch';
 export { getHttpClientSubscriptions } from './integrations/http/client-subscriptions';

packages/core/src/utils/sql.ts

@@ -141,3 +141,155 @@ function truncate(summary: string): string {
   const lastSpace = truncated.lastIndexOf(' ');
   return lastSpace > 0 ? truncated.substring(0, lastSpace) : truncated;
 }
+
+let integerLiteralRE: RegExp | undefined;
+
+/**
+ * SQL dialect variants that matter for finding the end of a string literal:
+ * - `standard` (PostgreSQL, SQLite): `"` quotes identifiers and `''` is the only in-string escape.
+ * - `mysql`: `"` quotes a string literal unless `ANSI_QUOTES` is set, and `\` escapes the next
+ *   character unless `NO_BACKSLASH_ESCAPES` is set. Both default to off, and mysql/mysql2 escape
+ *   inlined values with backslashes, so this is the mode their statements arrive in.
+ */
+export type SqlDialect = 'standard' | 'mysql';
+
+/**
+ * Returns the index just past the run's closing `delimiter`, or the end of the query if the run is
+ * never closed — an unterminated literal must swallow the remainder rather than let it through.
+ *
+ * A doubled delimiter (`''`) escapes itself in every dialect; backslash escapes are dialect- and
+ * context-dependent, so the caller decides.
+ */
+function findQuotedRunEnd(sql: string, start: number, delimiter: string, backslashEscapes: boolean): number {
+  for (let i = start + 1; i < sql.length; i++) {
+    const char = sql[i];
+    if (backslashEscapes && char === '\\') {
+      i++;
+    } else if (char === delimiter) {
+      if (sql[i + 1] !== delimiter) {
+        return i + 1;
+      }
+      i++;
+    }
+  }
+  return sql.length;
+}
+
+/**
+ * Replaces every string literal with `?` and drops every comment, in one pass.
+ *
+ * Doing this by scanning rather than by regex is what keeps quote state and comment state from
+ * being decided independently: a regex for `'...'` cannot see that the quote it stopped at was
+ * backslash-escaped, and a regex for `--...` cannot see that the `--` sits inside a literal. Both
+ * mistakes end with user data surviving into `db.query.text` and `db.query.summary`.
+ *
+ * Quoted identifiers are preserved — they are the table and column names the query summary is
+ * built from.
+ */
+function stripLiteralsAndComments(sql: string, dialect: SqlDialect): string {
+  const isMysql = dialect === 'mysql';
+  let out = '';
+  let i = 0;
+
+  while (i < sql.length) {
+    const char = sql[i]!;
+    const next = sql[i + 1];
+
+    if ((char === '-' && next === '-') || (isMysql && char === '#')) {
+      const lineEnd = sql.indexOf('\n', i);
+      i = lineEnd === -1 ? sql.length : lineEnd;
+      continue;
+    }
+
+    if (char === '/' && next === '*') {
+      const commentEnd = sql.indexOf('*/', i + 2);
+      i = commentEnd === -1 ? sql.length : commentEnd + 2;
+      continue;
+    }
+
+    // Quoted identifiers: backticks in MySQL, double quotes everywhere else
+    if (char === '`' || (char === '"' && !isMysql)) {
+      const runEnd = findQuotedRunEnd(sql, i, char, false);
+      out += sql.slice(i, runEnd);
+      i = runEnd;
+      continue;
+    }
+
+    if (char === "'" || (char === '"' && isMysql)) {
+      // A prefix like `X'1A'`, `B'01'` or PostgreSQL's `E'a\nb'` is part of the literal, so it has
+      // to collapse into the same `?` instead of being left behind as a bare identifier.
+      const prefix = char === "'" ? getLiteralPrefix(out, isMysql) : undefined;
+      out = prefix ? out.slice(0, -1) : out;
+      i = findQuotedRunEnd(sql, i, char, isMysql || prefix === 'E');
+      out += '?';
+      continue;
+    }
+
+    out += char;
+    i++;
+  }
+
+  return out;
+}
+
+/**
+ * Returns the literal-prefix character immediately before a `'`, if there is one: `X`/`B` for
+ * hex/binary literals, or `E` for a PostgreSQL escape string (which honors backslash escapes).
+ */
+function getLiteralPrefix(out: string, isMysql: boolean): 'X' | 'B' | 'E' | undefined {
+  // A prefix only counts when it stands alone — the `X` in `MAX'...'` belongs to the identifier
+  if (/[\w$]/.test(out.slice(-2, -1))) {
+    return undefined;
+  }
+
+  const prefix = out.slice(-1).toUpperCase();
+  if (prefix === 'X' || prefix === 'B') {
+    return prefix;
+  }
+  return prefix === 'E' && !isMysql ? 'E' : undefined;
+}
+
+/**
+ * Sanitize SQL query as per the OTEL semantic conventions
+ * https://opentelemetry.io/docs/specs/semconv/database/database-spans/#sanitization-of-dbquerytext
+ *
+ * PostgreSQL $n placeholders are preserved per OTEL spec - they're parameterized queries,
+ * not sensitive literals. Only actual values (strings, numbers, booleans) are sanitized.
+ *
+ * Pass `dialect` when the statement comes from a driver whose literals are not standard-quoted;
+ * see {@link SqlDialect}.
+ */
+export function sanitizeSqlQuery(sqlQuery: string | undefined, dialect: SqlDialect = 'standard'): string {
+  if (!sqlQuery) {
+    return 'Unknown SQL Query';
+  }
+
+  // Lazy init: constructing this at module scope would evaluate the lookbehind
+  // on import and crash Safari <16.4 browser bundles that reach this file via
+  // the core barrel. Building it on first call keeps the cost off the import path.
+  if (!integerLiteralRE) {
+    integerLiteralRE = new RegExp('(?<!\\$)-?\\b\\d+\\b', 'g');
+  }
+
+  return (
+    // Strip comments and string literals first: everything below is a regex that cannot tell
+    // whether it is looking at SQL syntax or at a user-supplied value.
+    stripLiteralsAndComments(sqlQuery, dialect)
+      .replace(/;\s*$/, '') // Remove trailing semicolons
+      // Collapse whitespace to a single space (after removing comments)
+      .replace(/\s+/g, ' ')
+      .trim() // Remove extra spaces and trim
+      // Sanitize hex numbers
+      .replace(/\b0x[0-9A-Fa-f]+/gi, '?')
+      // Sanitize boolean literals
+      .replace(/\b(?:TRUE|FALSE)\b/gi, '?')
+      // Sanitize numeric literals (preserve $n placeholders via negative lookbehind)
+      .replace(/-?\b\d+\.?\d*[eE][+-]?\d+\b/g, '?') // Scientific notation
+      .replace(/-?\b\d+\.\d+\b/g, '?') // Decimals
+      .replace(/-?\.\d+\b/g, '?') // Decimals starting with dot
+      .replace(integerLiteralRE, '?') // Integers (NOT $n placeholders)
+      // Collapse IN clauses for cardinality (both ? and $n variants)
+      .replace(/\bIN\b\s*\(\s*\?(?:\s*,\s*\?)*\s*\)/gi, 'IN (?)')
+      .replace(/\bIN\b\s*\(\s*\$\d+(?:\s*,\s*\$\d+)*\s*\)/gi, 'IN ($?)')
+  );
+}