Avoid copying exact bytes before Huffman decoding

CHANGELOG.rst

@@ -12,6 +12,8 @@ dev
 
 **Bugfixes**
 
+- Avoided copying exact ``bytes`` inputs before Huffman decoding.
+
 
 4.2.0 (2026-06-22)
 ------------------

src/hpack/huffman_table.py

@@ -87,11 +87,9 @@ def decode_huffman(huffman_string: bytes | bytearray | memoryview | None) -> byt
     flags = 0
     decoded_bytes = bytearray()
 
-    # Perversely, bytearrays are a lot more convenient across Python 2 and
-    # Python 3 because they behave *the same way* on both platforms. Given that
-    # we really do want numerical bytes when we iterate here, let's use a
-    # bytearray.
-    huffman_string = bytearray(huffman_string)
+    # Exact bytes already yield the octets this table expects, so skip the copy.
+    if type(huffman_string) is not bytes:
+        huffman_string = bytearray(huffman_string)
 
     # This loop is unrolled somewhat. Because we use a nibble, not a byte, we
     # need to handle each nibble twice. We unroll that: it makes the loop body

tests/test_huffman.py

@@ -7,6 +7,21 @@
 from hypothesis.strategies import binary
 
 
+class BytesWithCustomConversion(bytes):
+    def __bytes__(self):
+        return b'\xff'
+
+
+class BytearrayWithCustomConversion(bytearray):
+    def __bytes__(self):
+        return b'\xff'
+
+
+class BytesWithCustomIteration(bytes):
+    def __iter__(self):
+        return iter(b'\xff')
+
+
 class TestHuffman:
 
     def test_request_huffman_decoder(self):
@@ -34,6 +49,23 @@ def test_request_huffman_encode(self):
 
 
 class TestHuffmanDecoder:
+    def test_huffman_decoder_accepts_byte_buffers(self):
+        encoded = b'\xa8\xeb\x10d\x9c\xbf'
+        interleaved = bytes(value for byte in encoded for value in (byte, 0))
+        buffers = (
+            encoded,
+            BytesWithCustomConversion(encoded),
+            BytesWithCustomIteration(encoded),
+            bytearray(encoded),
+            BytearrayWithCustomConversion(encoded),
+            memoryview(encoded),
+            memoryview(encoded).cast('H'),
+            memoryview(interleaved)[::2],
+        )
+
+        for buffer in buffers:
+            assert decode_huffman(buffer) == b'no-cache'
+
     @given(data=binary())
     @example(b'\xff')
     @example(b'\x5f\xff\xff\xff\xff')