fix recent messages by handling valueless and escaped irc tags

The recent-messages API now serializes empty IRCv3 tags as bare keys
(e.g. `emotes`) instead of `emotes=`, which threw a RangeError in the
tag parser and silently dropped every recent message. Parse tags per the
IRCv3 spec: treat bare and empty values as equivalent, split on the first
`=` so values may contain `=`, and unescape values in a single pass
(\: \s \ \r \n). Also guard user extraction against prefixes without
`!`, and isolate per-message parse failures in _handleIRCData so one bad
message can no longer abort an entire batch.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

lib/models/irc.dart

@@ -959,22 +959,25 @@ class IRCMessage {
     // Obtain the index to break off the tags.
     final tagAndIrcMessageDivider = whole.indexOf(' ');
 
-    // Get the tags substring and escape characters.
-    // IRC messages escape spaces with \s.
-    final tags = whole
-        .substring(1, tagAndIrcMessageDivider)
-        .replaceAll('\\s', ' ');
-
-    // Next, parse and map the tags.
+    // Parse the IRCv3 tags into a map. The tags section is everything between
+    // the leading '@' and the first space. We split on the *raw* string: a
+    // literal ';' is always a separator because escaped semicolons inside a
+    // value are encoded as `\:` (unescaped per-value below).
     final mappedTags = <String, String>{};
+    final rawTags = whole.substring(1, tagAndIrcMessageDivider);
 
-    // Loop through each tag and store their key value pairs into the map.
-    for (final tag in tags.split(';')) {
-      // Skip if the tag has no value.
-      if (tag.endsWith('=')) continue;
+    for (final tag in rawTags.split(';')) {
+      // Split on the first '=' only — IRCv3 tag values may contain '='.
+      final equalsIndex = tag.indexOf('=');
 
-      final tagSplit = tag.split('=');
-      mappedTags[tagSplit[0]] = tagSplit[1];
+      // Skip valueless tags. Per the IRCv3 spec a missing value (bare `key`)
+      // and an empty value (`key=`) are equivalent — live Twitch sends the
+      // latter, the recent-messages API the former.
+      if (equalsIndex == -1 || equalsIndex == tag.length - 1) continue;
+
+      mappedTags[tag.substring(0, equalsIndex)] = _unescapeTagValue(
+        tag.substring(equalsIndex + 1),
+      );
     }
 
     // Now we'll parse the message itself.
@@ -990,9 +993,19 @@ class IRCMessage {
 
     // If the username exists, set it.
     // tmi.twitch.tv means the message was sent by Twitch rather than a user, so will be irrelevant.
-    final String? user = splitMessage[0] == _twitchIrcServer
-        ? mappedTags['login']
-        : splitMessage[0].substring(0, splitMessage[0].indexOf('!'));
+    final prefix = splitMessage[0];
+    final String? user;
+    if (prefix == _twitchIrcServer) {
+      user = mappedTags['login'];
+    } else {
+      // Normal prefixes are `nick!user@host`. Fall back to the login tag (or
+      // null) if the expected `!` is absent so an unexpected prefix shape can
+      // never throw a RangeError and take down the whole message.
+      final bangIndex = prefix.indexOf('!');
+      user = bangIndex == -1
+          ? mappedTags['login']
+          : prefix.substring(0, bangIndex);
+    }
 
     // If there is an associated message, set it.
     String? message;
@@ -1153,6 +1166,50 @@ class IRCMessage {
     );
   }
 
+  /// Unescapes an IRCv3 message-tag value.
+  ///
+  /// The spec defines these sequences: `\:` → `;`, `\s` → space, `\\` → `\`,
+  /// `\r` → CR, `\n` → LF. Any other `\x` collapses to `x`, and a lone trailing
+  /// backslash is dropped. Done in a single pass so sequences can't be
+  /// double-unescaped (e.g. `\\s` correctly yields `\s`, not a space).
+  static String _unescapeTagValue(String value) {
+    // Fast path: the vast majority of tag values contain no escape sequences.
+    if (!value.contains('\\')) return value;
+
+    final buffer = StringBuffer();
+    for (var i = 0; i < value.length; i++) {
+      if (value[i] != '\\') {
+        buffer.write(value[i]);
+        continue;
+      }
+
+      // A trailing backslash with nothing after it is dropped.
+      if (i + 1 == value.length) break;
+
+      switch (value[i + 1]) {
+        case ':':
+          buffer.write(';');
+          break;
+        case 's':
+          buffer.write(' ');
+          break;
+        case '\\':
+          buffer.write('\\');
+          break;
+        case 'r':
+          buffer.write('\r');
+          break;
+        case 'n':
+          buffer.write('\n');
+          break;
+        default:
+          buffer.write(value[i + 1]);
+      }
+      i++; // Skip the escaped character we just consumed.
+    }
+    return buffer.toString();
+  }
+
   factory IRCMessage.createNotice({
     required String message,
     VoidCallback? actionCallback,

lib/screens/channel/chat/stores/chat_store.dart

@@ -519,10 +519,19 @@ abstract class ChatStoreBase with Store {
     for (final message in data.trimRight().split('\r\n')) {
       // debugPrint('$message\n');
       if (message.startsWith('@')) {
-        final parsedIRCMessage = IRCMessage.fromString(
-          message,
-          userLogin: auth.user.details?.login,
-        );
+        // Parse defensively: a single malformed message must never abort the
+        // loop, or one bad frame would drop every other message batched with
+        // it (e.g. an entire recent-messages backlog). Skip and keep going.
+        final IRCMessage parsedIRCMessage;
+        try {
+          parsedIRCMessage = IRCMessage.fromString(
+            message,
+            userLogin: auth.user.details?.login,
+          );
+        } catch (e) {
+          debugPrint('Failed to parse IRC message: $e');
+          continue;
+        }
 
         if (parsedIRCMessage.user != null) {
           chatDetailsStore.chatUsers.add(parsedIRCMessage.user!);