[TTS] Fix crash calling KitKat+ API without check.

chrome/android/java/src/org/chromium/chrome/browser/contextualsearch/ContextualSearchContext.java

 // Copyright 2017 The Chromium Authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
 package org.chromium.chrome.browser.contextualsearch;
 
 import android.annotation.SuppressLint;
+import android.os.Build;
 import android.text.TextUtils;
 
 import org.chromium.base.annotations.CalledByNative;
 
 import javax.annotation.Nullable;
 
 /**
  * Provides a context in which to search, and links to the native ContextualSearchContext.
  * Includes the selection, selection offsets, surrounding page content, etc.
  * Requires an override of #onSelectionChanged to call when a non-empty selection is established
@@ skipping 250 matching lines @@
     /**
      * Finds the offset of the start of the word that includes the given initial offset.
      * The character at the initial offset is not examined, but the one before it is, and scanning
      * continues on to earlier characters until a non-word character is found.  The offset just
      * before the non-word character is returned.  If the initial offset is a space immediately
      * following a word then the start offset of that word is returned.
      * @param text The text to scan.
      * @param initial The initial offset to scan before.
      * @return The start of the word that contains the given initial offset, within {@code text}.
      */
     private int findWordStartOffset(String text, int initial) {

[Theresa, 2017/06/02 21:51:42]

This is just used to generate a ranker signal for whether the user tapped on the
edge of a word the middle, right?

[Donn Denman, 2017/06/02 22:00:42]

Correct, so it's non-critical for the feature.

         // Scan before, aborting if we hit any ideographic letter.
         for (int offset = initial - 1; offset >= 0; offset--) {
-            if (isIdeographicAtIndex(text, offset)) return INVALID_OFFSET;
+            if (isUnreliableWordBreakAtIndex(text, offset)) return INVALID_OFFSET;
 
             if (isWordBreakAtIndex(text, offset)) {
                 // The start of the word is after this word break.
                 return offset + 1;
             }
         }
         return INVALID_OFFSET;
     }
 
     /**
      * Finds the offset of the end of the word that includes the given initial offset.
      * NOTE: this is the index of the character just past the last character of the word,
      * so a 3 character word "who" has start index 0 and end index 3.
      * The character at the initial offset is examined and each one after that too until a non-word
      * character is encountered, and that offset will be returned.
      * @param text The text to scan.
      * @param initial The initial offset to scan from.
      * @return The end of the word that contains the given initial offset, within {@code text}.
      */
     private int findWordEndOffset(String text, int initial) {
         // Scan after, aborting if we hit any CJKN letter.
         for (int offset = initial; offset < text.length(); offset++) {
-            if (isIdeographicAtIndex(text, offset)) return INVALID_OFFSET;
+            if (isUnreliableWordBreakAtIndex(text, offset)) return INVALID_OFFSET;
 
             if (isWordBreakAtIndex(text, offset)) {
                 // The end of the word is the offset of this word break.
                 return offset;
             }
         }
         return INVALID_OFFSET;
     }
 
     /**
-     * @return Whether the character at the given index in the text is "Ideographic" (as in CJKV
-     *         languages), which means there may not be reliable word breaks.
+     * @return Whether the character at the given index in the text might be in an alphabet that has
+     *         unreliable word breaks, such as CKJV languages.  Returns {@code true} on older
+     *         platforms where we can't know for sure.
      */
     @SuppressLint("NewApi")
-    private boolean isIdeographicAtIndex(String text, int index) {
-        return Character.isIdeographic(text.charAt(index));
+    private boolean isUnreliableWordBreakAtIndex(String text, int index) {
+        if (Build.VERSION.SDK_INT >= Build.VERSION_CODES.KITKAT) {
+            return Character.isIdeographic(text.charAt(index));
+        } else {
+            return true; // Assume the worst if we can't check.
+        }
     }
 
     /**
      * @return Whether the character at the given index is a word-break.
      */
     private boolean isWordBreakAtIndex(String text, int index) {
         return !Character.isLetterOrDigit(text.charAt(index))
                 && text.codePointAt(index) != SOFT_HYPHEN_CHAR;
     }
 
@@ skipping 10 matching lines @@
     // ============================================================================================
     // Native methods.
     // ============================================================================================
     private native long nativeInit();
     private native void nativeDestroy(long nativeContextualSearchContext);
     private native void nativeSetResolveProperties(
             long nativeContextualSearchContext, String homeCountry, boolean maySendBasePageUrl);
     private native void nativeAdjustSelection(
             long nativeContextualSearchContext, int startAdjust, int endAdjust);
 }