Compact consumed data in ReadBuffer.feed() before size check

Previously compaction only happened on read operations. Now feed()
also triggers compaction, ensuring consumed bytes are reclaimed
before new data is appended. This keeps actual memory usage closer
to what max_message_size implies.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: antoineleclair

src/dqlitewire/buffer.py

@@ -48,6 +48,7 @@ def __init__(self, max_message_size: int = DEFAULT_MAX_MESSAGE_SIZE) -> None:
 
     def feed(self, data: bytes) -> None:
         """Add received data to the buffer."""
+        self._maybe_compact()
         unconsumed = len(self._data) - self._pos + len(data)
         if unconsumed > self._max_message_size:
             raise DecodeError(f"Buffer size {unconsumed} exceeds maximum {self._max_message_size}")

tests/test_buffer.py

@@ -272,6 +272,22 @@ def test_skip_message_waits_for_complete_normal_sized_message(self) -> None:
         # Buffer position should not have changed
         assert buf.available() == 24  # 8 header + 16 partial body
 
+    def test_feed_compacts_consumed_data(self) -> None:
+        """feed() should compact consumed data before extending the buffer.
+
+        When _pos is past the compaction threshold, feed() should compact
+        the buffer to reclaim consumed bytes before adding new data.
+        """
+        buf = ReadBuffer(max_message_size=65536)
+        # Simulate a state where data was consumed but not yet compacted
+        # (e.g., by external manipulation or edge case in read patterns)
+        buf._data = bytearray(b"\x00" * 5000)
+        buf._pos = 4500  # 4500 consumed, 500 remaining
+        # feed() should compact (since _pos > 4096) before extending
+        buf.feed(b"\x01" * 100)
+        assert buf._pos == 0
+        assert len(buf._data) == 600  # 500 remaining + 100 new
+
     def test_buffer_compaction(self) -> None:
         buf = ReadBuffer()
         # Feed a lot of small messages to trigger compaction