From 4b226858b1cfca70c9f27dd589e91e6139577c6d Mon Sep 17 00:00:00 2001
From: Richard R <me@richardr.dev>
Date: Sat, 6 Jun 2026 11:41:47 -0600
Subject: [PATCH] perf(highlight): optimize exact token sequence matching for
 large documents

Add a fast linear scan to resolve exact token sequence highlights before
falling back to the fuzzy window search, significantly improving
performance for large documents. Update unit tests to verify that
exact matches are found efficiently without invoking expensive
comparisons.
---
 src/lib/client/highlight-token-alignment.ts   | 23 +++++++++++++++++++
 .../highlight-token-alignment.vitest.spec.ts  | 15 ++++++++++++
 2 files changed, 38 insertions(+)

diff --git a/src/lib/client/highlight-token-alignment.ts b/src/lib/client/highlight-token-alignment.ts
index 26210ad..dd560bf 100644
--- a/src/lib/client/highlight-token-alignment.ts
+++ b/src/lib/client/highlight-token-alignment.ts
@@ -36,6 +36,29 @@ export function findBestHighlightTokenMatch(
 
   if (!patternLen || !normalizedTargets.length) return responseBase;
 
+  // Most viewer highlights are exact token sequences. Resolve those in one
+  // linear pass before entering the fuzzy window search, which is expensive
+  // enough to stall the UI when the target is a full HTML/TXT/MD document.
+  const firstPatternToken = normalizedPattern[0];
+  for (let start = 0; start < normalizedTargets.length; start += 1) {
+    if (normalizedTargets[start] !== firstPatternToken) continue;
+    let matches = true;
+    for (let offset = 1; offset < normalizedPattern.length; offset += 1) {
+      if (normalizedTargets[start + offset] !== normalizedPattern[offset]) {
+        matches = false;
+        break;
+      }
+    }
+    if (matches) {
+      return {
+        start,
+        end: start + normalizedPattern.length - 1,
+        rating: 1,
+        lengthDiff: 0,
+      };
+    }
+  }
+
   const patternTokenCount = normalizedPattern.length;
   const minWindowTokens = Math.max(1, Math.floor(patternTokenCount * 0.6));
   const maxWindowTokens = Math.max(
diff --git a/tests/unit/highlight-token-alignment.vitest.spec.ts b/tests/unit/highlight-token-alignment.vitest.spec.ts
index 7ee3212..32fc7d5 100644
--- a/tests/unit/highlight-token-alignment.vitest.spec.ts
+++ b/tests/unit/highlight-token-alignment.vitest.spec.ts
@@ -19,6 +19,21 @@ const alignmentWords = (
   }));
 
 describe('shared viewer highlight token alignment', () => {
+  test('finds an exact sentence in a large document without entering fuzzy comparison', () => {
+    const targets = Array.from({ length: 20_000 }, (_, index) => `token-${index}`);
+    targets.splice(15_000, 4, 'the', 'exact', 'sentence', 'here');
+
+    expect(findBestHighlightTokenMatch(
+      ['the', 'exact', 'sentence', 'here'],
+      targets,
+    )).toEqual({
+      start: 15_000,
+      end: 15_003,
+      rating: 1,
+      lengthDiff: 0,
+    });
+  });
+
   test('matches a complete Japanese sentence using locale-aware token count', () => {
     const sentence = 'これは日本語です。';
     const patternTokens = segmentWords(sentence, 'ja').map((token) => token.text);