improve detection of emojis when deleting

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

@@ -62,21 +62,32 @@ fun getFullEmojiAtEnd(s: CharSequence): String {
     while (offset > 0) {
         val codepoint = text.codePointBefore(offset)
         // stop if codepoint can't be emoji
-        if (!mightBeEmoji(codepoint)) return ""
+        if (!mightBeEmoji(codepoint))
+            return text.substring(offset)
         offset -= Character.charCount(codepoint)
-        // todo: if codepoint in 0x1F3FB..0x1F3FF -> combine with other emojis in front, but only if they actually combine
-        //  why isn't this done with zwj like everything else? skin tones can be emojis by themselves...
         if (offset > 0 && text[offset - 1].code == KeyCode.ZWJ) {
+            // todo: this appends ZWJ in weird cases like text, ZWJ, emoji
+            //  and detects single ZWJ as emoji (at least irrelevant for current use of getFullEmojiAtEnd)
             offset -= 1
             continue
         }
+
+        if (codepoint in 0x1F3FB..0x1F3FF) {
+            // Skin tones are not added with ZWJ, but just appended. This is not nice as they can be emojis on their own,
+            // but that's how it is done. Assume that an emoji before the skin tone will get merged (usually correct in practice)
+            val codepointBefore = text.codePointBefore(offset)
+            if (isEmoji(codepointBefore)) {
+                offset -= Character.charCount(codepointBefore)
+                continue
+            }
+        }
         // check the whole text after offset
         val textToCheck = text.substring(offset)
         if (isEmoji(textToCheck)) {
             return textToCheck
         }
     }
-    return ""
+    return text.substring(offset)
 }
 
 /** split the string on the first of consecutive space only, further consecutive spaces are added to the next split */

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

@@ -42,6 +42,8 @@ class StringUtilsTest {
     }
 
     @Test fun detectEmojisAtEnd() {
+        assertEquals("", getFullEmojiAtEnd("\uD83C\uDF83 "))
+        assertEquals("", getFullEmojiAtEnd("a"))
         assertEquals("\uD83C\uDF83", getFullEmojiAtEnd("\uD83C\uDF83"))
         assertEquals("ℹ️", getFullEmojiAtEnd("ℹ️"))
         assertEquals("ℹ️", getFullEmojiAtEnd("ℹ️ℹ️"))
@@ -51,6 +53,15 @@ class StringUtilsTest {
         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🏳️‍⚧️"))
+        assertEquals("\uD83D\uDD75\uD83C\uDFFC", getFullEmojiAtEnd(" 🕵🏼"))
+        assertEquals("\uD83D\uDD75\uD83C\uDFFC", getFullEmojiAtEnd("🕵🏼"))
+        assertEquals("\uD83C\uDFFC", getFullEmojiAtEnd(" \uD83C\uDFFC"))
+        // fails, but unlikely enough that we leave it unfixed
+        //assertEquals("\uD83C\uDFFC", getFullEmojiAtEnd("\uD83C\uDF84\uD83C\uDFFC"))
+        // below also fail, because ZWJ handling is not suitable for some unusual cases
+        //assertEquals("", getFullEmojiAtEnd("\u200D"))
+        //assertEquals("", getFullEmojiAtEnd("a\u200D"))
+        //assertEquals("\uD83D\uDE22", getFullEmojiAtEnd(" \u200D\uD83D\uDE22"))
     }
 
     // todo: add tests for emoji detection?