properly determine length of emojis when deleting

fixes issue described in https://github.com/Helium314/HeliBoard/issues/426#issuecomment-1944132643, but not the initial problem

app/src/main/java/helium314/keyboard/latin/RichInputConnection.java

@@ -1046,6 +1046,14 @@ public final class RichInputConnection implements PrivateCommandPerformer {
         return mCommittedTextBeforeComposingText.indexOf(" ") != -1;
     }
 
+    public int getCharCountToDeleteBeforeCursor() {
+        final int lastCodePoint = getCodePointBeforeCursor();
+        if (!Character.isSupplementaryCodePoint(lastCodePoint)) return 1;
+        if (!StringUtils.mightBeEmoji(lastCodePoint)) return 2;
+        final String text = mCommittedTextBeforeComposingText.toString() + mComposingText;
+        return StringUtilsKt.getFullEmojiAtEnd(text).length();
+    }
+
     public boolean hasLetterBeforeLastSpaceBeforeCursor() {
         return StringUtilsKt.hasLetterBeforeLastSpaceBeforeCursor(mCommittedTextBeforeComposingText);
     }

app/src/main/java/helium314/keyboard/latin/common/Constants.java

@@ -191,6 +191,8 @@ public final class Constants {
 
     public static final String REGEXP_PERIOD = "\\.";
     public static final String STRING_SPACE = " ";
+    public static final int CODE_ZWNJ = '\u200C';
+    public static final int CODE_ZWJ = '\u200D';
 
     public static boolean isLetterCode(final int code) {
         return code >= CODE_SPACE;

app/src/main/java/helium314/keyboard/latin/common/StringUtils.kt

@@ -53,6 +53,29 @@ fun hasLetterBeforeLastSpaceBeforeCursor(s: CharSequence): Boolean {
     return letter
 }
 
+/** get the complete emoji at end of [s], considering that emojis can be joined with ZWJ resulting in different emojis */
+fun getFullEmojiAtEnd(s: CharSequence): String {
+    val text = if (s is String) s else s.toString()
+    var offset = text.length
+    while (offset > 0) {
+        val codepoint = text.codePointBefore(offset)
+        // stop if codepoint can't be emoji
+        if (!mightBeEmoji(codepoint)) return ""
+        offset -= Character.charCount(codepoint)
+        if (offset > 0 && text[offset - 1].code == Constants.CODE_ZWJ) {
+            // continue if ZWJ
+            offset -= 1
+            continue
+        }
+        // check the whole text after offset
+        val textToCheck = text.substring(offset)
+        if (isEmoji(textToCheck)) {
+            return textToCheck
+        }
+    }
+    return ""
+}
+
 /** split the string on the first of consecutive space only, further consecutive spaces are added to the next split */
 fun String.splitOnFirstSpacesOnly(): List<String> {
     val out = mutableListOf<String>()

app/src/main/java/helium314/keyboard/latin/inputlogic/InputLogic.java

@@ -1253,12 +1253,8 @@ public final class InputLogic {
                         // TODO: Add a new StatsUtils method onBackspaceWhenNoText()
                         return;
                     }
-                    if (StringUtils.mightBeEmoji(codePointBeforeCursor)) {
+                    final int lengthToDelete = Character.isSupplementaryCodePoint(codePointBeforeCursor)
-                        // emoji length varies, so we'd need to find out length to delete correctly
+                            ? mConnection.getCharCountToDeleteBeforeCursor() : 1;
-                        // the solution is not optimal, but a reasonable workaround for issues when trying to delete emojis
-                        sendDownUpKeyEvent(KeyEvent.KEYCODE_DEL);
-                    } else {
-                        final int lengthToDelete = Character.isSupplementaryCodePoint(codePointBeforeCursor) ? 2 : 1;
                     mConnection.deleteTextBeforeCursor(lengthToDelete);
                     int totalDeletedLength = lengthToDelete;
                     if (mDeleteCount > Constants.DELETE_ACCELERATE_AT) {
@@ -1270,8 +1266,8 @@ public final class InputLogic {
                         final int codePointBeforeCursorToDeleteAgain =
                                 mConnection.getCodePointBeforeCursor();
                         if (codePointBeforeCursorToDeleteAgain != Constants.NOT_A_CODE) {
-                                final int lengthToDeleteAgain = Character.isSupplementaryCodePoint(
+                            final int lengthToDeleteAgain = Character.isSupplementaryCodePoint(codePointBeforeCursorToDeleteAgain)
-                                        codePointBeforeCursorToDeleteAgain) ? 2 : 1;
+                                    ? mConnection.getCharCountToDeleteBeforeCursor() : 1;
                             mConnection.deleteTextBeforeCursor(lengthToDeleteAgain);
                             totalDeletedLength += lengthToDeleteAgain;
                         }
@@ -1279,7 +1275,6 @@ public final class InputLogic {
                     StatsUtils.onBackspacePressed(totalDeletedLength);
                 }
             }
-            }
             if (!hasUnlearnedWordBeingDeleted) {
                 // Consider unlearning the word being deleted (if we have not done so already).
                 unlearnWordBeingDeleted(

app/src/test/java/helium314/keyboard/latin/StringUtilsTest.kt

@@ -2,6 +2,8 @@
 package helium314.keyboard.latin
 
 import helium314.keyboard.latin.common.StringUtils
+import helium314.keyboard.latin.common.getFullEmojiAtEnd
+import org.junit.Assert.assertEquals
 import org.junit.Test
 
 // todo: actually this test could/should be significantly expanded...
@@ -39,6 +41,18 @@ class StringUtilsTest {
         assert(StringUtils.isInsideDoubleQuoteOrAfterDigit("hello \"yes\", \"h"))
     }
 
+    @Test fun detectEmojisAtEnd() {
+        assertEquals("\uD83C\uDF83", getFullEmojiAtEnd("\uD83C\uDF83"))
+        assertEquals("ℹ️", getFullEmojiAtEnd("ℹ️"))
+        assertEquals("ℹ️", getFullEmojiAtEnd("ℹ️ℹ️"))
+        assertEquals("\uD83D\uDE22", getFullEmojiAtEnd("x\uD83D\uDE22"))
+        assertEquals("", getFullEmojiAtEnd("x\uD83D\uDE22 "))
+        assertEquals("\uD83C\uDFF4\u200D☠️", getFullEmojiAtEnd("ok \uD83C\uDFF4\u200D☠️"))
+        assertEquals("\uD83C\uDFF3️\u200D\uD83C\uDF08", getFullEmojiAtEnd("\uD83C\uDFF3️\u200D\uD83C\uDF08"))
+        assertEquals("\uD83C\uDFF3️\u200D\uD83C\uDF08", getFullEmojiAtEnd("\uD83C\uDFF4\u200D☠️\uD83C\uDFF3️\u200D\uD83C\uDF08"))
+        assertEquals("\uD83C\uDFF3️\u200D⚧️", getFullEmojiAtEnd("hello there🏳️‍⚧️"))
+    }
+
     // todo: add tests for emoji detection?
     //  could help towards fully fixing https://github.com/Helium314/HeliBoard/issues/22
     //  though this might be tricky, as some emojis will show as one on new Android versions, and