fix(telegram): clip mid-stream overflow instead of splitting (#48648)

plugins/platforms/telegram/adapter.py

@@ -2975,11 +2975,18 @@ class TelegramAdapter(BasePlatformAdapter):
                 return rich_result
 
         # Pre-flight: if content already exceeds the limit, split-and-deliver
-        # without round-tripping a doomed edit.
+        # without round-tripping a doomed edit.  During streaming
+        # (finalize=False) we truncate instead of splitting — splitting creates
+        # continuation messages whose IDs become the new edit target, and on
+        # the next token chunk the full accumulated text is re-edited into the
+        # continuation, triggering another split → infinite duplication loop
+        # (#48648).  The full content is delivered when finalize=True.
         if utf16_len(content) > self.MAX_MESSAGE_LENGTH:
-            return await self._edit_overflow_split(
-                chat_id, message_id, content, finalize=finalize, metadata=metadata,
-            )
+            if finalize:
+                return await self._edit_overflow_split(
+                    chat_id, message_id, content, finalize=finalize, metadata=metadata,
+                )
+            content = self._truncate_stream_overflow_preview(content)
 
         try:
             if not finalize:
@@ -3028,9 +3035,18 @@ class TelegramAdapter(BasePlatformAdapter):
                     "[%s] edit_message overflow (%d UTF-16 > %d), splitting",
                     self.name, utf16_len(content), self.MAX_MESSAGE_LENGTH,
                 )
-                return await self._edit_overflow_split(
-                    chat_id, message_id, content, finalize=finalize, metadata=metadata,
+                if finalize:
+                    return await self._edit_overflow_split(
+                        chat_id, message_id, content, finalize=finalize, metadata=metadata,
+                    )
+                # Mid-stream: truncate and retry instead of splitting (#48648).
+                truncated = self._truncate_stream_overflow_preview(content)
+                await self._bot.edit_message_text(
+                    chat_id=int(chat_id),
+                    message_id=int(message_id),
+                    text=truncated,
                 )
+                return SendResult(success=True, message_id=message_id)
             # Flood control / RetryAfter — short waits are retried inline,
             # long waits return a failure immediately so streaming can fall back
             # to a normal final send instead of leaving a truncated partial.
@@ -3093,6 +3109,21 @@ class TelegramAdapter(BasePlatformAdapter):
             )
             return SendResult(success=False, error=str(e))
 
+    def _truncate_stream_overflow_preview(self, content: str) -> str:
+        """Return a one-message preview for oversized streaming edits.
+
+        Streaming edits must keep targeting the original message. Splitting a
+        mid-stream preview creates continuation messages and moves the active
+        message id, so the next accumulated-token edit repeats the overflow
+        cycle (#48648). Final edits still use ``_edit_overflow_split`` to
+        deliver the complete response.
+        """
+        return self.truncate_message(
+            content,
+            self.MAX_MESSAGE_LENGTH,
+            len_fn=utf16_len,
+        )[0]
+
     async def _edit_overflow_split(
         self,
         chat_id: str,

tests/gateway/test_telegram_format.py

@@ -908,12 +908,12 @@ class TestEditMessageStreamingSafety:
 
     @pytest.mark.asyncio
     async def test_message_too_long_splits_into_continuations_not_silent_truncation(self):
-        """When edit_message_text exceeds Telegram's 4096 UTF-16 limit, the
-        adapter must split the content across the existing message + new
-        continuation messages so the user gets the full reply.  Previously
-        the adapter best-effort truncated the content with '…' and returned
-        success=True, dropping everything past the truncation boundary
-        (#19537)."""
+        """When edit_message_text exceeds Telegram's 4096 UTF-16 limit on
+        finalize, the adapter must split the content across the existing
+        message + new continuation messages so the user gets the full reply.
+        Previously the adapter best-effort truncated the content with '…' and
+        returned success=True, dropping everything past the truncation
+        boundary (#19537)."""
         adapter = TelegramAdapter(PlatformConfig(enabled=True, token="fake-token"))
         adapter._bot = MagicMock()
         adapter._bot.edit_message_text = AsyncMock()
@@ -926,7 +926,7 @@ class TestEditMessageStreamingSafety:
 
         # 6000-char content well over the 4096 UTF-16 limit.
         oversized = "x" * 6000
-        result = await adapter.edit_message("123", "456", oversized, finalize=False)
+        result = await adapter.edit_message("123", "456", oversized, finalize=True)
 
         # Adapter reports success with continuations populated.
         assert result.success is True
@@ -961,7 +961,7 @@ class TestEditMessageStreamingSafety:
             "-100123",
             "456",
             "x" * 6000,
-            finalize=False,
+            finalize=True,
             metadata={"thread_id": "17585"},
         )
 
@@ -970,6 +970,60 @@ class TestEditMessageStreamingSafety:
         assert all(kwargs.get("message_thread_id") == 17585 for kwargs in sent_kwargs)
         assert sent_kwargs[0]["reply_to_message_id"] == 456
 
+    @pytest.mark.asyncio
+    async def test_mid_stream_overflow_truncates_instead_of_splitting(self):
+        """During streaming (finalize=False), oversized content must be
+        truncated to fit one message rather than split into continuations.
+        Splitting mid-stream creates new message IDs that the stream consumer
+        then edits with the full accumulated text, causing an infinite
+        duplication loop (#48648)."""
+        adapter = TelegramAdapter(PlatformConfig(enabled=True, token="fake-token"))
+        adapter._bot = MagicMock()
+        adapter._bot.edit_message_text = AsyncMock()
+        _next_id = [1000]
+
+        async def _fake_send(**kwargs):
+            _next_id[0] += 1
+            return SimpleNamespace(message_id=_next_id[0])
+
+        adapter._bot.send_message = AsyncMock(side_effect=_fake_send)
+
+        oversized = "x" * 6000
+        result = await adapter.edit_message("123", "456", oversized, finalize=False)
+
+        # Must NOT create continuation messages during streaming.
+        assert result.success is True
+        assert adapter._bot.send_message.await_count == 0, (
+            "mid-stream overflow must not send continuation messages"
+        )
+        # message_id stays the original — no shift to a new ID.
+        assert result.message_id == "456"
+        # The edit must contain truncated content (≤ MAX_MESSAGE_LENGTH).
+        edited_text = adapter._bot.edit_message_text.call_args.kwargs["text"]
+        assert len(edited_text) <= adapter.MAX_MESSAGE_LENGTH
+
+    @pytest.mark.asyncio
+    async def test_mid_stream_reactive_overflow_retries_truncated_edit(self):
+        """If Telegram rejects a streaming edit as too long, retry with a
+        one-message preview instead of splitting into continuations."""
+        adapter = TelegramAdapter(PlatformConfig(enabled=True, token="fake-token"))
+        adapter._bot = MagicMock()
+        adapter._bot.edit_message_text = AsyncMock(
+            side_effect=[Exception("Bad Request: message is too long"), None]
+        )
+        adapter._bot.send_message = AsyncMock()
+
+        content = "x" * adapter.MAX_MESSAGE_LENGTH
+        result = await adapter.edit_message("123", "456", content, finalize=False)
+
+        assert result.success is True
+        assert result.message_id == "456"
+        adapter._bot.send_message.assert_not_called()
+        assert adapter._bot.edit_message_text.await_count == 2
+        retry_text = adapter._bot.edit_message_text.await_args_list[1].kwargs["text"]
+        assert len(retry_text) <= adapter.MAX_MESSAGE_LENGTH
+
+
 # =========================================================================
 # Telegram guest mention gating
 # =========================================================================