From e05af818c6cb270ac321a39bf7b8f2fb4c42a08e Mon Sep 17 00:00:00 2001
From: Koji Ishii <kojii@google.com>
Date: Sat, 12 Aug 2023 18:16:46 +0900
Subject: [PATCH] parseBoundaries

---
 javascript/src/html_processor.ts | 33 ++++++++++----------------------
 javascript/src/parser.ts         | 25 ++++++++++++++++++++----
 2 files changed, 31 insertions(+), 27 deletions(-)

diff --git a/javascript/src/html_processor.ts b/javascript/src/html_processor.ts
index de8865a8..f2ced89a 100644
--- a/javascript/src/html_processor.ts
+++ b/javascript/src/html_processor.ts
@@ -319,31 +319,18 @@ export class HTMLProcessor {
     // No changes if whitespace-only.
     if (/^\s*$/.test(text)) return;
 
-    // Split the text into a list of phrases.
-    const phrases = this.parser_.parse(text);
-    assert(phrases.length > 0);
-    assert(
-      phrases.reduce((sum, phrase) => sum + phrase.length, 0) === text.length
-    );
+    // Compute the phrase boundaries.
+    const boundaries = this.parser_.parseBoundaries(text);
     // No changes if single phrase.
-    if (phrases.length <= 1) return;
-
-    // Compute the boundary indices from the list of phrase strings.
-    const boundaries = [];
-    let char_index = 0;
-    for (const phrase of phrases) {
-      assert(phrase.length > 0);
-      char_index += phrase.length;
-      boundaries.push(char_index);
-    }
-
-    // The break opportunity at the end of a block is not needed. Instead of
-    // removing it, turn it to a sentinel for `splitTextNodesAtBoundaries` by
-    // making it larger than the text length.
+    if (boundaries.length <= 0) return;
+    // The boundaries should be between 1 and `text.length - 1` in the
+    // ascending order.
     assert(boundaries[0] > 0);
-    assert(boundaries[boundaries.length - 1] === text.length);
-    ++boundaries[boundaries.length - 1];
-    assert(boundaries.length > 1);
+    assert(boundaries.every((x, i) => i === 0 || x > boundaries[i - 1]));
+    assert(boundaries[boundaries.length - 1] < text.length);
+
+    // Add a sentinel to help iterating.
+    boundaries.push(text.length + 1);
 
     this.splitTextNodes(textNodes, boundaries);
     this.applyBlockStyle(paragraph.element);
diff --git a/javascript/src/parser.ts b/javascript/src/parser.ts
index 4825e5f6..e8e4e537 100644
--- a/javascript/src/parser.ts
+++ b/javascript/src/parser.ts
@@ -59,9 +59,27 @@ export class Parser {
    * @param sentence An input sentence.
    * @returns The retrieved chunks.
    */
-  parse(sentence: string) {
+  parse(sentence: string): string[] {
     if (sentence === '') return [];
-    const result = [sentence[0]];
+    const boundaries = this.parseBoundaries(sentence);
+    const result = [];
+    let start = 0;
+    for (const boundary of boundaries) {
+      result.push(sentence.slice(start, boundary));
+      start = boundary;
+    }
+    result.push(sentence.slice(start));
+    return result;
+  }
+
+  /**
+   * Parses the input sentence and returns a list of boundaries.
+   *
+   * @param sentence An input sentence.
+   * @returns The list of boundaries.
+   */
+  parseBoundaries(sentence: string): number[] {
+    const result = [];
     const baseScore =
       -0.5 *
       sum([...this.model.values()].map(group => [...group.values()]).flat());
@@ -81,8 +99,7 @@ export class Parser {
       score += this.model.get('TW2')?.get(sentence.slice(i - 2, i + 1)) || 0;
       score += this.model.get('TW3')?.get(sentence.slice(i - 1, i + 2)) || 0;
       score += this.model.get('TW4')?.get(sentence.slice(i, i + 3)) || 0;
-      if (score > 0) result.push('');
-      result[result.length - 1] += sentence[i];
+      if (score > 0) result.push(i);
     }
     return result;
   }