fix(parse): keep getFeatureData's global regex a literal (#65)

* fix(parse): keep getFeatureData's global regex a literal

0.15.0 built getFeatureData's regex with new RegExp(featuresRe, "gs").
A transpiler that polyfills named capture groups only wraps regex
literals. CRA's production Babel does this, so in flowchart.fun's
production bundle the copy came back as a plain native RegExp, and
match.groups was undefined. The loop then hit
`if (!match.groups) continue;` on an empty match at lastIndex 0 and
never ended, which froze the page on its first parse.

The global regex is now its own literal, featuresReGlobal, next to
featuresRe. A test keeps the two sources identical. A match without
groups now breaks out of the loop instead of continuing, because an
empty match never advances lastIndex.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* chore(lint): ban building a RegExp from a non-literal

Consumers that transpile named capture groups, like CRA's production
Babel, only wrap regex literals. A RegExp built from another regex
loses match.groups, which is how 0.15.0 froze flowchart.fun's
production build. The lookbehind rule guards against the same kind of
consumer break. Run against 0.15.0's getFeatureData.ts, this rule
flags line 5. It flags nothing in the current tree.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

Author: rob-gordon

graph-selector/.eslintrc.cjs

@@ -20,6 +20,15 @@ module.exports = {
         caughtErrorsIgnorePattern: "^_",
       },
     ],
+    "no-restricted-syntax": [
+      "error",
+      {
+        selector:
+          ":matches(NewExpression, CallExpression)[callee.name='RegExp'] > :not(Literal, TemplateLiteral).arguments:first-child",
+        message:
+          "Write a regex literal instead. Consumers that transpile named groups (CRA's Babel) wrap literals only, so a RegExp built from another regex loses match.groups.",
+      },
+    ],
   },
   env: {
     browser: true,

graph-selector/src/getFeatureData.test.ts

@@ -1,8 +1,14 @@
 import { describe, expect, test } from "vitest";
 
 import { getFeatureData } from "./getFeatureData";
+import { featuresRe, featuresReGlobal } from "./regexps";
 
 describe("getFeatureData", () => {
+  test("its global regex is featuresRe with the g flag", () => {
+    expect(featuresReGlobal.source).toBe(featuresRe.source);
+    expect(featuresReGlobal.flags).toBe("gs");
+  });
+
   test("it parses IDs", () => {
     const { id, line } = getFeatureData("#hello");
     expect(id).toBe("hello");

graph-selector/src/getFeatureData.ts

@@ -1,9 +1,6 @@
-import { featuresRe } from "./regexps";
+import { featuresReGlobal as re } from "./regexps";
 import { Data, Descriptor } from "./types";
 
-// compiled once; lastIndex is reset per call so no state leaks between calls
-const re = new RegExp(featuresRe, "gs");
-
 /**
  * Given a line, it returns the feaures from the line
  */
@@ -17,7 +14,8 @@ export function getFeatureData(_line: string) {
   let attributes = "";
 
   while ((match = re.exec(line)) != null) {
-    if (!match.groups) continue;
+    // an empty match never advances lastIndex, so continuing here would loop forever
+    if (!match.groups) break;
     if (!match.groups.replace) break;
 
     // if (match.groups.pointer) pointers.push(match.groups.pointer);

graph-selector/src/regexps.ts

@@ -15,3 +15,11 @@ export const getFeaturesIndex = (line: string) => {
  */
 export const featuresRe =
   /(?<replace>(?<id>#[\w-]+)?(?<classes>(\.[a-zA-Z]{1}[\w-]*)*)?(?<attributes>(\[[^\]=]+(?<attributeValue>=(".*"|'.*'|.*))?\])*))/s;
+
+/**
+ * featuresRe with the g flag, for getFeatureData's exec loop. Keep it a literal: a transpiler
+ * that polyfills named groups (CRA's Babel does) wraps literals only, and a copy made with
+ * new RegExp(featuresRe, "gs") comes back without match.groups.
+ */
+export const featuresReGlobal =
+  /(?<replace>(?<id>#[\w-]+)?(?<classes>(\.[a-zA-Z]{1}[\w-]*)*)?(?<attributes>(\[[^\]=]+(?<attributeValue>=(".*"|'.*'|.*))?\])*))/gs;