Replace quadratic algo in LineDecoder

Leading to enormous speedups when doing things such as
Response(...).iter_lines() as described on issue #2422

Author: giannitedesco

httpx/_decoders.py

@@ -266,57 +266,24 @@ def __init__(self) -> None:
         self.buffer = ""
 
     def decode(self, text: str) -> typing.List[str]:
-        lines = []
-
-        if text and self.buffer and self.buffer[-1] == "\r":
-            if text.startswith("\n"):
-                # Handle the case where we have an "\r\n" split across
-                # our previous input, and our new chunk.
-                lines.append(self.buffer[:-1] + "\n")
-                self.buffer = ""
-                text = text[1:]
-            else:
-                # Handle the case where we have "\r" at the end of our
-                # previous input.
-                lines.append(self.buffer[:-1] + "\n")
-                self.buffer = ""
-
-        while text:
-            num_chars = len(text)
-            for idx in range(num_chars):
-                char = text[idx]
-                next_char = None if idx + 1 == num_chars else text[idx + 1]
-                if char == "\n":
-                    lines.append(self.buffer + text[: idx + 1])
-                    self.buffer = ""
-                    text = text[idx + 1 :]
-                    break
-                elif char == "\r" and next_char == "\n":
-                    lines.append(self.buffer + text[:idx] + "\n")
-                    self.buffer = ""
-                    text = text[idx + 2 :]
-                    break
-                elif char == "\r" and next_char is not None:
-                    lines.append(self.buffer + text[:idx] + "\n")
-                    self.buffer = ""
-                    text = text[idx + 1 :]
-                    break
-                elif next_char is None:
-                    self.buffer += text
-                    text = ""
-                    break
+        if self.buffer:
+            text = self.buffer + text
+
+        if not text:
+            return []
+
+        lines = text.splitlines(True)
+        if text.endswith("\n"):
+            self.buffer = ""
+        else:
+            remainder = lines.pop()
+            self.buffer = remainder
 
         return lines
 
     def flush(self) -> typing.List[str]:
-        if self.buffer.endswith("\r"):
-            # Handle the case where we had a trailing '\r', which could have
-            # been a '\r\n' pair.
-            lines = [self.buffer[:-1] + "\n"]
-        elif self.buffer:
-            lines = [self.buffer]
-        else:
-            lines = []
+        # this handles stripping any trailing "\r"
+        lines = self.buffer.splitlines(True)
         self.buffer = ""
         return lines
 

tests/test_decoders.py

@@ -257,48 +257,48 @@ def test_line_decoder_nl():
 def test_line_decoder_cr():
     decoder = LineDecoder()
     assert decoder.decode("") == []
-    assert decoder.decode("a\r\rb\rc") == ["a\n", "\n", "b\n"]
+    assert decoder.decode("a\r\rb\rc") == ["a\r", "\r", "b\r"]
     assert decoder.flush() == ["c"]
 
     decoder = LineDecoder()
     assert decoder.decode("") == []
-    assert decoder.decode("a\r\rb\rc\r") == ["a\n", "\n", "b\n"]
-    assert decoder.flush() == ["c\n"]
+    assert decoder.decode("a\r\rb\rc\r") == ["a\r", "\r", "b\r"]
+    assert decoder.flush() == ["c\r"]
 
     # Issue #1033
     decoder = LineDecoder()
     assert decoder.decode("") == []
     assert decoder.decode("12345\r") == []
-    assert decoder.decode("foo ") == ["12345\n"]
+    assert decoder.decode("foo ") == ["12345\r"]
     assert decoder.decode("bar ") == []
     assert decoder.decode("baz\r") == []
-    assert decoder.flush() == ["foo bar baz\n"]
+    assert decoder.flush() == ["foo bar baz\r"]
 
 
 def test_line_decoder_crnl():
     decoder = LineDecoder()
     assert decoder.decode("") == []
-    assert decoder.decode("a\r\n\r\nb\r\nc") == ["a\n", "\n", "b\n"]
+    assert decoder.decode("a\r\n\r\nb\r\nc") == ["a\r\n", "\r\n", "b\r\n"]
     assert decoder.flush() == ["c"]
 
     decoder = LineDecoder()
     assert decoder.decode("") == []
-    assert decoder.decode("a\r\n\r\nb\r\nc\r\n") == ["a\n", "\n", "b\n", "c\n"]
+    assert decoder.decode("a\r\n\r\nb\r\nc\r\n") == ["a\r\n", "\r\n", "b\r\n", "c\r\n"]
     assert decoder.flush() == []
 
     decoder = LineDecoder()
     assert decoder.decode("") == []
     assert decoder.decode("a\r") == []
-    assert decoder.decode("\n\r\nb\r\nc") == ["a\n", "\n", "b\n"]
+    assert decoder.decode("\n\r\nb\r\nc") == ["a\r\n", "\r\n", "b\r\n"]
     assert decoder.flush() == ["c"]
 
     # Issue #1033
     decoder = LineDecoder()
     assert decoder.decode("") == []
-    assert decoder.decode("12345\r\n") == ["12345\n"]
+    assert decoder.decode("12345\r\n") == ["12345\r\n"]
     assert decoder.decode("foo ") == []
     assert decoder.decode("bar ") == []
-    assert decoder.decode("baz\r\n") == ["foo bar baz\n"]
+    assert decoder.decode("baz\r\n") == ["foo bar baz\r\n"]
     assert decoder.flush() == []