fix(syntax): do not let a regex literal unbalance a block

The brace scanner knew about strings and comments but had no case for regex
literals. A quote inside one opened a phantom string that swallowed every brace
until the next quote; a lone `{` or `}` inside one miscounted block depth. Both
failed the component with "Unbalanced braces" pointing at the block's first line.

`/-/g` parsed fine, which is why this went unnoticed -- it needs a quote or a
brace inside the pattern to bite.

Regex-vs-division is decided by scanning back to the last significant
character, erring towards division: mistaking division for a regex would
swallow code to the next `/` and lose any braces between. A regex cannot span a
newline, so an unterminated one on the line is treated as "not a regex", which
is what keeps a bare URL in view text intact.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

packages/syntax/src/tokenizer.ts

@@ -48,6 +48,85 @@ export const isIdentPart = (c: string) => /[A-Za-z0-9_]/.test(c);
  * reimplementing it — a second hand-rolled scanner is how apostrophes in
  * prose used to swallow braces.
  */
+/**
+ * Identifiers that can precede a `/` without ending an expression, so the `/`
+ * opens a regex rather than dividing.
+ */
+const REGEX_PRECEDING_KEYWORDS = new Set([
+  "return",
+  "typeof",
+  "instanceof",
+  "in",
+  "of",
+  "new",
+  "delete",
+  "void",
+  "do",
+  "else",
+  "yield",
+  "await",
+  "case",
+]);
+
+/**
+ * Decide whether the `/` at `i` opens a regex literal or is a division sign.
+ *
+ * Scans backwards for the last significant character. `a / b` divides; `(/a/)`,
+ * `= /a/` and `return /a/` do not. Erring towards division is the safe
+ * direction -- mistaking division for a regex would swallow everything to the
+ * next `/` and lose any braces in between.
+ */
+function opensRegex(src: string, i: number): boolean {
+  let j = i - 1;
+  while (j >= 0 && (src[j] === " " || src[j] === "\t" || src[j] === "\r" || src[j] === "\n")) j--;
+  if (j < 0) return true;
+  const prev = src[j]!;
+  if (/[A-Za-z0-9_$]/.test(prev)) {
+    // An identifier ends an expression, so `/` divides -- unless it is a
+    // keyword that cannot end one, like `return`.
+    let k = j;
+    while (k >= 0 && /[A-Za-z0-9_$]/.test(src[k]!)) k--;
+    return REGEX_PRECEDING_KEYWORDS.has(src.slice(k + 1, j + 1));
+  }
+  // `)` and `]` close an expression, `.` continues one, and a quote ends a
+  // literal; anything else leaves us in a position where a regex may start.
+  return (
+    prev !== ")" && prev !== "]" && prev !== "." && prev !== '"' && prev !== "'" && prev !== "`"
+  );
+}
+
+/**
+ * Scan a regex literal starting at `i`, returning the index just past its
+ * closing `/` and flags, or null when this is not in fact a regex.
+ *
+ * A regex literal cannot span a newline, so an unterminated one is treated as
+ * "not a regex" rather than swallowing the rest of the file. That is what keeps
+ * a bare URL in view text (`https://example.com/a//b`) intact.
+ */
+function skipRegex(src: string, i: number): number | null {
+  let j = i + 1;
+  let inClass = false;
+  while (j < src.length) {
+    const c = src[j]!;
+    if (c === "\n") return null;
+    if (c === "\\") {
+      j += 2;
+      continue;
+    }
+    if (inClass) {
+      if (c === "]") inClass = false;
+    } else if (c === "[") {
+      inClass = true;
+    } else if (c === "/") {
+      j++;
+      while (j < src.length && /[a-z]/.test(src[j]!)) j++;
+      return j;
+    }
+    j++;
+  }
+  return null;
+}
+
 export function skipLiteralOrComment(src: string, i: number, atLineStart: boolean): number | null {
   const c = src[i];
   if (c === "/" && src[i + 1] === "*") {
@@ -70,6 +149,13 @@ export function skipLiteralOrComment(src: string, i: number, atLineStart: boolea
     }
     return src.length;
   }
+  // A regex literal is neither a string nor a brace pair, but it can contain
+  // both. Without this, a quote inside one opened a phantom string that
+  // swallowed every brace to the next quote, and a lone `{`/`}` miscounted
+  // block depth.
+  if (c === "/" && opensRegex(src, i)) {
+    return skipRegex(src, i);
+  }
   return null;
 }
 

packages/syntax/test/syntax.test.ts

@@ -235,3 +235,54 @@ test("comments may contain apostrophes without unbalancing a block", () => {
   // The URL in view text must survive: `//` is only a comment at line start.
   expect(JSON.stringify(ast.view)).toContain("https://example.com/a//b");
 });
+
+// A regex literal is not a string and not a pair of braces. The brace scanner
+// knew about quotes and comments but had no case for regexes, so a quote inside
+// one opened a phantom string that swallowed every brace until the next quote,
+// and a lone `{` or `}` inside one miscounted depth. Both failed the whole
+// component -- with a green build in the reported case, because the damage
+// landed in generated output rather than at parse time.
+test("regex literals do not unbalance a block", () => {
+  const mk = (body: string) =>
+    `page P {
+  load server {
+${body}
+    return { x };
+  }
+
+  view { <div>{x}</div> }
+}
+`;
+  // A quote inside a regex used to open a string that ran to the next quote.
+  expect(parse(mk(`    const x = /it's/.test("its");`)).name).toBe("P");
+  expect(parse(mk(`    const x = /"/.test("q");`)).name).toBe("P");
+  // A brace inside a regex used to be counted as block depth.
+  expect(parse(mk(`    const x = /\{/.test("{");`)).name).toBe("P");
+  expect(parse(mk(`    const x = /}/.test("}");`)).name).toBe("P");
+  // A brace quantifier is balanced, but must not be counted either.
+  expect(parse(mk(`    const x = /^a{2,3}$/.test("aa");`)).name).toBe("P");
+  // A `/` inside a character class does not close the regex.
+  expect(parse(mk(`    const x = /[/'"{]/.test("/");`)).name).toBe("P");
+  // The case that already worked must keep working.
+  expect(parse(mk(`    const x = "a-b".replace(/-/g, " ");`)).name).toBe("P");
+});
+
+// Division must not be mistaken for a regex, or the scanner would swallow code
+// from the `/` to the next one and lose any braces in between.
+test("division is not treated as a regex literal", () => {
+  const source = `page P {
+  load server {
+    const half = 10 / 2;
+    const ratio = (a + b) / 2;
+    const each = items[0] / total;
+    if (half > 1) {
+      return { half };
+    }
+    return { half: 0 };
+  }
+
+  view { <div>{half}</div> }
+}
+`;
+  expect(parse(source).name).toBe("P");
+});