fix(lint): locate findings in linear time on large compositions (#4373)

Since the source-location mapping landed, building the whitespace-collapsed copy of a file grew a
string with += and called endsWith on it per character. V8 flattens that string on every call, so a
1.2 MB index.html cost ~31 s of CPU in every lint run, including the one preview runs before it
listens.

packages/lint/src/sourceLocations.test.ts

@@ -201,3 +201,12 @@ it("uses classic-script Acorn semantics for new.target and hashbang", async () =
   const hashbang = await lintHyperframeHtml("<script>#!/usr/bin/env node\nconst a=1;</script>");
   expect(hashbang.findings.some((f) => f.code === "invalid_inline_script_syntax")).toBe(false);
 });
+
+it("locates a snippet at the end of a large source in linear time", () => {
+  const filler = "<div>\n  <p>a  b</p>\n</div>\n".repeat(12_000);
+  const html = `${filler}<style>\n  .late {\n    position: fixed;\n  }\n</style>`;
+  const start = performance.now();
+  const location = sourceLocationFor(html, finding({ snippet: ".late { position: fixed; }" }));
+  expect(location).toEqual({ line: 36_002, column: 3 });
+  expect(performance.now() - start).toBeLessThan(1_000);
+});

packages/lint/src/sourceLocations.ts

@@ -5,17 +5,18 @@ import { parseHtmlStructure, readDecodedAttr } from "./utils";
 type Location = Pick<HyperframeLintFinding, "line" | "column">;
 
 function normalized(source: string): { text: string; offsets: number[] } {
-  let text = "";
+  // An array, not +=: V8 flattens a += string on every endsWith, which made this quadratic.
+  const chars: string[] = [];
   const offsets: number[] = [];
+  let lastWasSpace = false;
   for (let i = 0; i < source.length; i++) {
-    const char = source[i]!;
-    if (/\s/.test(char)) {
-      if (text.endsWith(" ")) continue;
-      text += " ";
-    } else text += char;
+    const isSpace = /\s/.test(source[i]!);
+    if (isSpace && lastWasSpace) continue;
+    chars.push(isSpace ? " " : source[i]!);
     offsets.push(i);
+    lastWasSpace = isSpace;
   }
-  return { text, offsets };
+  return { text: chars.join(""), offsets };
 }
 
 /** Resolve against original bytes, not the comment-stripped/template-unwrapped rule input.