Fix parsing failure for large JSON outputs split across buffer boundaries

src/claude_code_sdk/_internal/transport/subprocess_cli.py

@@ -183,11 +183,22 @@ class SubprocessCLITransport(Transport):
             tg.start_soon(read_stderr)
 
             try:
+                # store incomplete buffered json output
+                incomplete_json_line_str = None
                 async for line in self._stdout_stream:
                     line_str = line.strip()
                     if not line_str:
                         continue
 
+                    if incomplete_json_line_str:
+                        assert not line_str.startswith("{"), (
+                            "New line must complete incomplete JSON"
+                        )
+                        line_str = incomplete_json_line_str + line_str
+
+                    # Store parsed JSON output for the line
+                    parsed_json_outputs = []
+
                     # Split on newlines in case multiple JSON objects are buffered together
                     json_lines = line_str.split("\n")
 
@@ -198,16 +209,31 @@ class SubprocessCLITransport(Transport):
 
                         try:
                             data = json.loads(json_line)
-                            try:
-                                yield data
-                            except GeneratorExit:
-                                # Handle generator cleanup gracefully
-                                return
+
+                            # line has been parsed, reseting incomplete_json_line_str
+                            incomplete_json_line_str = None
+
+                            # Yield later to avoid duplicates on incomplete line_str
+                            parsed_json_outputs.append(data)
                         except json.JSONDecodeError as e:
                             if json_line.startswith("{") or json_line.startswith("["):
-                                raise SDKJSONDecodeError(json_line, e) from e
+                                incomplete_json_line_str = line_str
+
+                                # raise error only if output is complete JSON but unable to parse
+                                if json_line.endswith("}") or json_line.startswith("]"):
+                                    raise SDKJSONDecodeError(json_line, e) from e
                             continue
 
+                    if incomplete_json_line_str:
+                        continue
+
+                    for json_output_data in parsed_json_outputs:
+                        try:
+                            yield json_output_data
+                        except GeneratorExit:
+                            # Handle generator cleanup gracefully
+                            return
+
             except anyio.ClosedResourceError:
                 pass
 

tests/test_subprocess_buffering.py

@@ -6,7 +6,9 @@ from typing import Any
 from unittest.mock import AsyncMock, MagicMock
 
 import anyio
+import pytest
 
+from claude_code_sdk._errors import CLIJSONDecodeError
 from claude_code_sdk._internal.transport.subprocess_cli import SubprocessCLITransport
 from claude_code_sdk.types import ClaudeCodeOptions
 
@@ -139,3 +141,148 @@ class TestSubprocessBuffering:
             assert messages[1]["id"] == "res1"
 
         anyio.run(_test)
+
+    def test_incomplete_json_across_multiple_lines(self) -> None:
+        """Test parsing when JSON is split across multiple lines due to buffering."""
+
+        async def _test() -> None:
+            # Large JSON that gets split across multiple lines
+            json_obj = {
+                "type": "assistant",
+                "message": {
+                    "content": [
+                        {
+                            "type": "text",
+                            "text": "This is a very long response that might get split",
+                        },
+                        {
+                            "type": "tool_use",
+                            "id": "tool_123",
+                            "name": "Read",
+                            "input": {"file_path": "/very/long/path/to/file.py"},
+                        },
+                    ]
+                },
+            }
+
+            complete_json = json.dumps(json_obj)
+
+            # Split the JSON at an arbitrary point to simulate buffering
+            split_point = len(complete_json) // 2
+            first_part = complete_json[:split_point]
+            second_part = complete_json[split_point:]
+
+            transport = SubprocessCLITransport(
+                prompt="test", options=ClaudeCodeOptions(), cli_path="/usr/bin/claude"
+            )
+
+            mock_process = MagicMock()
+            mock_process.returncode = None
+            mock_process.wait = AsyncMock(return_value=None)
+            transport._process = mock_process
+
+            # Simulate receiving the JSON in two parts
+            transport._stdout_stream = MockTextReceiveStream([first_part, second_part])
+            transport._stderr_stream = MockTextReceiveStream([])
+
+            messages: list[Any] = []
+            async for msg in transport.receive_messages():
+                messages.append(msg)
+
+            # Should parse as one complete message
+            assert len(messages) == 1
+            assert messages[0]["type"] == "assistant"
+            assert len(messages[0]["message"]["content"]) == 2
+
+        anyio.run(_test)
+
+    def test_malformed_complete_json_raises_error(self) -> None:
+        """Test that malformed but seemingly complete JSON raises an error."""
+
+        async def _test() -> None:
+            # JSON that looks complete but is malformed
+            malformed_json = '{"type": "message", "invalid": unquoted_value}'
+
+            transport = SubprocessCLITransport(
+                prompt="test", options=ClaudeCodeOptions(), cli_path="/usr/bin/claude"
+            )
+
+            mock_process = MagicMock()
+            mock_process.returncode = None
+            mock_process.wait = AsyncMock(return_value=None)
+            transport._process = mock_process
+            transport._stdout_stream = MockTextReceiveStream([malformed_json])
+            transport._stderr_stream = MockTextReceiveStream([])
+
+            # Should raise CLIJSONDecodeError for malformed complete JSON
+            # The exception will be wrapped in an ExceptionGroup due to anyio task group
+            with pytest.raises(Exception) as exc_info:
+                messages: list[Any] = []
+                async for msg in transport.receive_messages():
+                    messages.append(msg)
+
+            # Verify the actual exception is CLIJSONDecodeError
+            assert len(exc_info.value.exceptions) == 1
+            assert isinstance(exc_info.value.exceptions[0], CLIJSONDecodeError)
+
+        anyio.run(_test)
+
+    def test_no_duplicate_output_when_incomplete_json_followed_by_valid(self) -> None:
+        """Test that we don't duplicate output when incomplete JSON is followed by valid JSON."""
+
+        async def _test() -> None:
+            # First valid JSON
+            valid_json1 = {"type": "user", "message": "first message"}
+
+            # Second valid JSON
+            valid_json2 = {"type": "assistant", "message": "second message"}
+
+            # Large JSON that will be incomplete
+            large_json = {
+                "type": "result",
+                "data": "Very large data " * 200,  # Make it large enough to split
+                "status": "completed",
+            }
+
+            valid_str1 = json.dumps(valid_json1)
+            valid_str2 = json.dumps(valid_json2)
+            large_str = json.dumps(large_json)
+
+            # Split the large JSON
+            split_point = len(large_str) // 2
+            large_part1 = large_str[:split_point]
+            large_part2 = large_str[split_point:]
+
+            # Create line that has: valid JSON + newline + valid JSON + newline + incomplete large JSON
+            combined_line = valid_str1 + "\n" + valid_str2 + "\n" + large_part1
+
+            lines = [
+                combined_line,  # First line: 2 valid JSONs + start of large JSON
+                large_part2,  # Second line: completion of large JSON
+            ]
+
+            transport = SubprocessCLITransport(
+                prompt="test", options=ClaudeCodeOptions(), cli_path="/usr/bin/claude"
+            )
+
+            mock_process = MagicMock()
+            mock_process.returncode = None
+            mock_process.wait = AsyncMock(return_value=None)
+            transport._process = mock_process
+            transport._stdout_stream = MockTextReceiveStream(lines)
+            transport._stderr_stream = MockTextReceiveStream([])
+
+            messages: list[Any] = []
+            async for msg in transport.receive_messages():
+                messages.append(msg)
+
+            # Should have exactly 3 messages: 2 from first line + 1 completed large JSON
+            assert len(messages) == 3
+            assert messages[0]["type"] == "user"
+            assert messages[0]["message"] == "first message"
+            assert messages[1]["type"] == "assistant"
+            assert messages[1]["message"] == "second message"
+            assert messages[2]["type"] == "result"
+            assert messages[2]["status"] == "completed"
+
+        anyio.run(_test)