wip and recognize word-aligned

lib/model/emoji.dart

@@ -328,7 +328,19 @@ class EmojiStoreImpl with EmojiStore {
 /// (Rather vacuous for the moment; this structure will
 /// gain more substance in an upcoming commit.)
 enum EmojiMatchQuality {
-  exact, prefix, other;
+  /// The query matches the whole emoji name (or the literal emoji itself).
+  exact,
+
+  /// The query matches a prefix of the emoji name, but not the whole name.
+  prefix,
+
+  /// The query matches starting at the start of a word in the emoji name,
+  /// but not the start of the whole name.
+  wordAligned,
+
+  /// The query matches somewhere in the emoji name,
+  /// but not at the start of any word.
+  other;
 
   /// The best possible quality of match.
   static const best = exact;
@@ -374,26 +386,30 @@ class EmojiAutocompleteView extends AutocompleteView<EmojiAutocompleteQuery, Emo
     return match == null ? null : EmojiAutocompleteResult(match, candidate);
   }
 
-  static const _numResultRanks = 7;
+  static const _numResultRanks = 9;
 
   // Compare sort_emojis in Zulip web:shared/src/typeahead.ts .
-  // Differences in behavior include:
+  // Behavior differences we might copy:
   //  * Web ranks each name of a Unicode emoji separately.
+  //  * Web recognizes a word-aligned match starting after [ /-] as well as [_].
+  //
+  // Behavior differences that web should probably fix (of varying impact):
   //  * Among popular emoji with non-exact matches,
-  //    web doesn't prioritize by quality of match; we do.
-  //  * OTOH web only counts an emoji as "popular" for ranking if the query
-  //    is a prefix of a word in the emoji's name or alias.
-  //    TODO reconcile that; requiring word boundary may be helpful;
-  //      rejecting query that spans words seems unhelpful.
-  //  * TODO web ranks matches starting at a word boundary ahead of other
-  //      non-prefix matches; we don't yet.
-  //  * Web distinguishes prefix matches by case-sensitive vs. not; we don't.
-  //    That doesn't seem helpful for emoji search.
+  //    web doesn't prioritize prefix over word-aligned; we do.
+  //    (This affects just one case: for query "o",
+  //    we put :octopus: before :working_on_it:.)
+  //  * Web only counts an emoji as "popular" for ranking if the query
+  //    is a prefix of a single word in the name; so "thumbs_" or "working_on_i"
+  //    lose the ranking boost for :thumbs_up: and :working_on_it: respectively.
+  //  * Web starts with only case-sensitive exact matches ("perfect matches"),
+  //    and puts case-insensitive exact matches just ahead of prefix matches;
+  //    it also distinguishes prefix matches by case-sensitive vs. not.
+  //    We use case-insensitive matches throughout;
+  //    case seems unhelpful for emoji search.
   //  * Web suppresses Unicode emoji names shadowed by a realm emoji
   //    only if the latter is also a match for the query.  That mostly works,
   //    because emoji with the same name will mostly both match or both not;
   //    but it breaks if the Unicode emoji was a literal match.
-  //    TODO(web): that's a bug
   static int _rankResult(EmojiAutocompleteResult result) {
     if (result.matchQuality == EmojiMatchQuality.exact) {
       return 0;
@@ -408,10 +424,10 @@ class EmojiAutocompleteView extends AutocompleteView<EmojiAutocompleteQuery, Emo
       ReactionType.unicodeEmoji => false,
     };
     return switch (result.matchQuality) {
-      EmojiMatchQuality.exact  => throw Error(), // handled above
-      EmojiMatchQuality.prefix => isPopular ? 1 : isCustomEmoji ? 3 : 4,
-      // TODO word-boundary vs. not
-      EmojiMatchQuality.other  => isPopular ? 2 : isCustomEmoji ? 5 : 6,
+      EmojiMatchQuality.exact       => throw Error(), // handled above
+      EmojiMatchQuality.prefix      => isPopular ? 1 : isCustomEmoji ? 3 : 5,
+      EmojiMatchQuality.wordAligned => isPopular ? 2 : isCustomEmoji ? 4 : 6,
+      EmojiMatchQuality.other       =>                 isCustomEmoji ? 7 : 8,
     };
   }
 
@@ -460,29 +476,23 @@ class EmojiAutocompleteQuery extends ComposeAutocompleteQuery {
   }
 
   EmojiMatchQuality? _matchName(String emojiName) {
-    if (emojiName == _adjusted)          return EmojiMatchQuality.exact;
-    if (emojiName.startsWith(_adjusted)) return EmojiMatchQuality.prefix;
-    if (_nameMatches(emojiName))         return EmojiMatchQuality.other;
-    return null;
-  }
-
-  // Compare query_matches_string_in_order in Zulip web:shared/src/typeahead.ts .
-  bool _nameMatches(String emojiName) {
+    // Compare query_matches_string_in_order in Zulip web:shared/src/typeahead.ts .
     // TODO(#1067) this assumes emojiName is already lower-case (and no diacritics)
-    const String separator = '_';
-
-    if (!_adjusted.contains(separator)) {
-      // If the query is a single token (doesn't contain a separator),
-      // the match can be anywhere in the string.
-      return emojiName.contains(_adjusted);
+    if (emojiName == _adjusted)              return EmojiMatchQuality.exact;
+    if (emojiName.startsWith(_adjusted))     return EmojiMatchQuality.prefix;
+    const String sep = '_';
+    if (emojiName.contains(sep + _adjusted)) return EmojiMatchQuality.wordAligned;
+    if (_adjusted.contains(sep)) {
+      // If there is a separator in the query, then we
+      // require the match to start at the start of a token.
+      // (E.g. for 'ab_cd_ef', query could be 'ab_c' or 'cd_ef',
+      // but not 'b_cd_ef'.)
+      return null;
     }
-
-    // If there is a separator in the query, then we
-    // require the match to start at the start of a token.
-    // (E.g. for 'ab_cd_ef', query could be 'ab_c' or 'cd_ef',
-    // but not 'b_cd_ef'.)
-    assert(!emojiName.startsWith(_adjusted)); // checked before calling this method
-    return emojiName.contains(separator + _adjusted);
+    // If the query is a single token (doesn't contain a separator),
+    // the match can be anywhere in the string.
+    if (emojiName.contains(_adjusted))       return EmojiMatchQuality.other;
+    return null;
   }
 
   @override