Quelle  streams.py   Sprache: Python

import asyncio
import collections
import warnings
from typing import (
    Awaitable,
    Callable,
    Deque,
    Final,
    Generic,
    List,
    Optional,
    Tuple,
    TypeVar,
)

from .base_protocol import BaseProtocol
from .helpers import (
    _EXC_SENTINEL,
    BaseTimerContext,
    TimerNoop,
    set_exception,
    set_result,
)
from .log import internal_logger

__all__ = (
    "EMPTY_PAYLOAD",
    "EofStream",
    "StreamReader",
    "DataQueue",
    "FlowControlDataQueue",
)

_T = TypeVar("_T")


class EofStream(Exception):
    """eof stream indication."""


class AsyncStreamIterator(Generic[_T]):
    def __init__(self, read_func: Callable[[], Awaitable[_T]]) -> None:
        self.read_func = read_func

    def __aiter__(self) -> "AsyncStreamIterator[_T]":
        return self

    async def __anext__(self) -> _T:
        try:
            rv = await self.read_func()
        except EofStream:
            raise StopAsyncIteration
        if rv == b"":
            raise StopAsyncIteration
        return rv


class ChunkTupleAsyncStreamIterator:
    def __init__(self, stream: "StreamReader") -> None:
        self._stream = stream

    def __aiter__(self) -> "ChunkTupleAsyncStreamIterator":
        return self

    async def __anext__(self) -> Tuple[bytes, bool]:
        rv = await self._stream.readchunk()
        if rv == (b"", False):
            raise StopAsyncIteration
        return rv


class AsyncStreamReaderMixin:
    def __aiter__(self) -> AsyncStreamIterator[bytes]:
        return AsyncStreamIterator(self.readline)  # type: ignore[attr-defined]

    def iter_chunked(self, n: int) -> AsyncStreamIterator[bytes]:
        """Returns an asynchronous iterator that yields chunks of size n."""
        return AsyncStreamIterator(lambda: self.read(n))  # type: ignore[attr-defined]

    def iter_any(self) -> AsyncStreamIterator[bytes]:
        """Yield all available data as soon as it is received."""
        return AsyncStreamIterator(self.readany)  # type: ignore[attr-defined]

    def iter_chunks(self) -> ChunkTupleAsyncStreamIterator:
        """Yield chunks of data as they are received by the server.

        The yielded objects are tuples
        of (bytes, bool) as returned by the StreamReader.readchunk method.
        """
        return ChunkTupleAsyncStreamIterator(self)  # type: ignore[arg-type]


class StreamReader(AsyncStreamReaderMixin):
    """An enhancement of asyncio.StreamReader.

    Supports asynchronous iteration by line, chunk or as available::

        async for line in reader:
            ...
        async for chunk in reader.iter_chunked(1024):
            ...
        async for slice in reader.iter_any():
            ...

    """

    total_bytes = 0

    def __init__(
        self,
        protocol: BaseProtocol,
        limit: int,
        *,
        timer: Optional[BaseTimerContext] = None,
        loop: Optional[asyncio.AbstractEventLoop] = None,
    ) -> None:
        self._protocol = protocol
        self._low_water = limit
        self._high_water = limit * 2
        if loop is None:
            loop = asyncio.get_event_loop()
        self._loop = loop
        self._size = 0
        self._cursor = 0
        self._http_chunk_splits: Optional[List[int]] = None
        self._buffer: Deque[bytes] = collections.deque()
        self._buffer_offset = 0
        self._eof = False
        self._waiter: Optional[asyncio.Future[None]] = None
        self._eof_waiter: Optional[asyncio.Future[None]] = None
        self._exception: Optional[BaseException] = None
        self._timer = TimerNoop() if timer is None else timer
        self._eof_callbacks: List[Callable[[], None]] = []

    def __repr__(self) -> str:
        info = [self.__class__.__name__]
        if self._size:
            info.append("%d bytes" % self._size)
        if self._eof:
            info.append("eof")
        if self._low_water != 2**16:  # default limit
            info.append("low=%d high=%d" % (self._low_water, self._high_water))
        if self._waiter:
            info.append("w=%r" % self._waiter)
        if self._exception:
            info.append("e=%r" % self._exception)
        return "<%s>" % " ".join(info)

    def get_read_buffer_limits(self) -> Tuple[int, int]:
        return (self._low_water, self._high_water)

    def exception(self) -> Optional[BaseException]:
        return self._exception

    def set_exception(
        self,
        exc: BaseException,
        exc_cause: BaseException = _EXC_SENTINEL,
    ) -> None:
        self._exception = exc
        self._eof_callbacks.clear()

        waiter = self._waiter
        if waiter is not None:
            self._waiter = None
            set_exception(waiter, exc, exc_cause)

        waiter = self._eof_waiter
        if waiter is not None:
            self._eof_waiter = None
            set_exception(waiter, exc, exc_cause)

    def on_eof(self, callback: Callable[[], None]) -> None:
        if self._eof:
            try:
                callback()
            except Exception:
                internal_logger.exception("Exception in eof callback")
        else:
            self._eof_callbacks.append(callback)

    def feed_eof(self) -> None:
        self._eof = True

        waiter = self._waiter
        if waiter is not None:
            self._waiter = None
            set_result(waiter, None)

        waiter = self._eof_waiter
        if waiter is not None:
            self._eof_waiter = None
            set_result(waiter, None)

        for cb in self._eof_callbacks:
            try:
                cb()
            except Exception:
                internal_logger.exception("Exception in eof callback")

        self._eof_callbacks.clear()

    def is_eof(self) -> bool:
        """Return True if 'feed_eof' was called."""
        return self._eof

    def at_eof(self) -> bool:
        """Return True if the buffer is empty and 'feed_eof' was called."""
        return self._eof and not self._buffer

    async def wait_eof(self) -> None:
        if self._eof:
            return

        assert self._eof_waiter is None
        self._eof_waiter = self._loop.create_future()
        try:
            await self._eof_waiter
        finally:
            self._eof_waiter = None

    def unread_data(self, data: bytes) -> None:
        """rollback reading some data from stream, inserting it to buffer head."""
        warnings.warn(
            "unread_data() is deprecated "
            "and will be removed in future releases (#3260)",
            DeprecationWarning,
            stacklevel=2,
        )
        if not data:
            return

        if self._buffer_offset:
            self._buffer[0] = self._buffer[0][self._buffer_offset :]
            self._buffer_offset = 0
        self._size += len(data)
        self._cursor -= len(data)
        self._buffer.appendleft(data)
        self._eof_counter = 0

    # TODO: size is ignored, remove the param later
    def feed_data(self, data: bytes, size: int = 0) -> None:
        assert not self._eof, "feed_data after feed_eof"

        if not data:
            return

        data_len = len(data)
        self._size += data_len
        self._buffer.append(data)
        self.total_bytes += data_len

        waiter = self._waiter
        if waiter is not None:
            self._waiter = None
            set_result(waiter, None)

        if self._size > self._high_water and not self._protocol._reading_paused:
            self._protocol.pause_reading()

    def begin_http_chunk_receiving(self) -> None:
        if self._http_chunk_splits is None:
            if self.total_bytes:
                raise RuntimeError(
                    "Called begin_http_chunk_receiving when" "some data was already fed"
                )
            self._http_chunk_splits = []

    def end_http_chunk_receiving(self) -> None:
        if self._http_chunk_splits is None:
            raise RuntimeError(
                "Called end_chunk_receiving without calling "
                "begin_chunk_receiving first"
            )

        # self._http_chunk_splits contains logical byte offsets from start of
        # the body transfer. Each offset is the offset of the end of a chunk.
        # "Logical" means bytes, accessible for a user.
        # If no chunks containing logical data were received, current position
        # is difinitely zero.
        pos = self._http_chunk_splits[-1] if self._http_chunk_splits else 0

        if self.total_bytes == pos:
            # We should not add empty chunks here. So we check for that.
            # Note, when chunked + gzip is used, we can receive a chunk
            # of compressed data, but that data may not be enough for gzip FSM
            # to yield any uncompressed data. That's why current position may
            # not change after receiving a chunk.
            return

        self._http_chunk_splits.append(self.total_bytes)

        # wake up readchunk when end of http chunk received
        waiter = self._waiter
        if waiter is not None:
            self._waiter = None
            set_result(waiter, None)

    async def _wait(self, func_name: str) -> None:
        if not self._protocol.connected:
            raise RuntimeError("Connection closed.")

        # StreamReader uses a future to link the protocol feed_data() method
        # to a read coroutine. Running two read coroutines at the same time
        # would have an unexpected behaviour. It would not possible to know
        # which coroutine would get the next data.
        if self._waiter is not None:
            raise RuntimeError(
                "%s() called while another coroutine is "
                "already waiting for incoming data" % func_name
            )

        waiter = self._waiter = self._loop.create_future()
        try:
            with self._timer:
                await waiter
        finally:
            self._waiter = None

    async def readline(self) -> bytes:
        return await self.readuntil()

    async def readuntil(self, separator: bytes = b"\n") -> bytes:
        seplen = len(separator)
        if seplen == 0:
            raise ValueError("Separator should be at least one-byte string")

        if self._exception is not None:
            raise self._exception

        chunk = b""
        chunk_size = 0
        not_enough = True

        while not_enough:
            while self._buffer and not_enough:
                offset = self._buffer_offset
                ichar = self._buffer[0].find(separator, offset) + 1
                # Read from current offset to found separator or to the end.
                data = self._read_nowait_chunk(
                    ichar - offset + seplen - 1 if ichar else -1
                )
                chunk += data
                chunk_size += len(data)
                if ichar:
                    not_enough = False

                if chunk_size > self._high_water:
                    raise ValueError("Chunk too big")

            if self._eof:
                break

            if not_enough:
                await self._wait("readuntil")

        return chunk

    async def read(self, n: int = -1) -> bytes:
        if self._exception is not None:
            raise self._exception

        # migration problem; with DataQueue you have to catch
        # EofStream exception, so common way is to run payload.read() inside
        # infinite loop. what can cause real infinite loop with StreamReader
        # lets keep this code one major release.
        if __debug__:
            if self._eof and not self._buffer:
                self._eof_counter = getattr(self, "_eof_counter", 0) + 1
                if self._eof_counter > 5:
                    internal_logger.warning(
                        "Multiple access to StreamReader in eof state, "
                        "might be infinite loop.",
                        stack_info=True,
                    )

        if not n:
            return b""

        if n < 0:
            # This used to just loop creating a new waiter hoping to
            # collect everything in self._buffer, but that would
            # deadlock if the subprocess sends more than self.limit
            # bytes.  So just call self.readany() until EOF.
            blocks = []
            while True:
                block = await self.readany()
                if not block:
                    break
                blocks.append(block)
            return b"".join(blocks)

        # TODO: should be `if` instead of `while`
        # because waiter maybe triggered on chunk end,
        # without feeding any data
        while not self._buffer and not self._eof:
            await self._wait("read")

        return self._read_nowait(n)

    async def readany(self) -> bytes:
        if self._exception is not None:
            raise self._exception

        # TODO: should be `if` instead of `while`
        # because waiter maybe triggered on chunk end,
        # without feeding any data
        while not self._buffer and not self._eof:
            await self._wait("readany")

        return self._read_nowait(-1)

    async def readchunk(self) -> Tuple[bytes, bool]:
        """Returns a tuple of (data, end_of_http_chunk).

        When chunked transfer
        encoding is used, end_of_http_chunk is a boolean indicating if the end
        of the data corresponds to the end of a HTTP chunk , otherwise it is
        always False.
        """
        while True:
            if self._exception is not None:
                raise self._exception

            while self._http_chunk_splits:
                pos = self._http_chunk_splits.pop(0)
                if pos == self._cursor:
                    return (b"", True)
                if pos > self._cursor:
                    return (self._read_nowait(pos - self._cursor), True)
                internal_logger.warning(
                    "Skipping HTTP chunk end due to data "
                    "consumption beyond chunk boundary"
                )

            if self._buffer:
                return (self._read_nowait_chunk(-1), False)
                # return (self._read_nowait(-1), False)

            if self._eof:
                # Special case for signifying EOF.
                # (b'', True) is not a final return value actually.
                return (b"", False)

            await self._wait("readchunk")

    async def readexactly(self, n: int) -> bytes:
        if self._exception is not None:
            raise self._exception

        blocks: List[bytes] = []
        while n > 0:
            block = await self.read(n)
            if not block:
                partial = b"".join(blocks)
                raise asyncio.IncompleteReadError(partial, len(partial) + n)
            blocks.append(block)
            n -= len(block)

        return b"".join(blocks)

    def read_nowait(self, n: int = -1) -> bytes:
        # default was changed to be consistent with .read(-1)
        #
        # I believe the most users don't know about the method and
        # they are not affected.
        if self._exception is not None:
            raise self._exception

        if self._waiter and not self._waiter.done():
            raise RuntimeError(
                "Called while some coroutine is waiting for incoming data."
            )

        return self._read_nowait(n)

    def _read_nowait_chunk(self, n: int) -> bytes:
        first_buffer = self._buffer[0]
        offset = self._buffer_offset
        if n != -1 and len(first_buffer) - offset > n:
            data = first_buffer[offset : offset + n]
            self._buffer_offset += n

        elif offset:
            self._buffer.popleft()
            data = first_buffer[offset:]
            self._buffer_offset = 0

        else:
            data = self._buffer.popleft()

        self._size -= len(data)
        self._cursor += len(data)

        chunk_splits = self._http_chunk_splits
        # Prevent memory leak: drop useless chunk splits
        while chunk_splits and chunk_splits[0] < self._cursor:
            chunk_splits.pop(0)

        if self._size < self._low_water and self._protocol._reading_paused:
            self._protocol.resume_reading()
        return data

    def _read_nowait(self, n: int) -> bytes:
        """Read not more than n bytes, or whole buffer if n == -1"""
        self._timer.assert_timeout()

        chunks = []
        while self._buffer:
            chunk = self._read_nowait_chunk(n)
            chunks.append(chunk)
            if n != -1:
                n -= len(chunk)
                if n == 0:
                    break

        return b"".join(chunks) if chunks else b""


class EmptyStreamReader(StreamReader):  # lgtm [py/missing-call-to-init]
    def __init__(self) -> None:
        self._read_eof_chunk = False

    def __repr__(self) -> str:
        return "<%s>" % self.__class__.__name__

    def exception(self) -> Optional[BaseException]:
        return None

    def set_exception(
        self,
        exc: BaseException,
        exc_cause: BaseException = _EXC_SENTINEL,
    ) -> None:
        pass

    def on_eof(self, callback: Callable[[], None]) -> None:
        try:
            callback()
        except Exception:
            internal_logger.exception("Exception in eof callback")

    def feed_eof(self) -> None:
        pass

    def is_eof(self) -> bool:
        return True

    def at_eof(self) -> bool:
        return True

    async def wait_eof(self) -> None:
        return

    def feed_data(self, data: bytes, n: int = 0) -> None:
        pass

    async def readline(self) -> bytes:
        return b""

    async def read(self, n: int = -1) -> bytes:
        return b""

    # TODO add async def readuntil

    async def readany(self) -> bytes:
        return b""

    async def readchunk(self) -> Tuple[bytes, bool]:
        if not self._read_eof_chunk:
            self._read_eof_chunk = True
            return (b"", False)

        return (b"", True)

    async def readexactly(self, n: int) -> bytes:
        raise asyncio.IncompleteReadError(b"", n)

    def read_nowait(self, n: int = -1) -> bytes:
        return b""


EMPTY_PAYLOAD: Final[StreamReader] = EmptyStreamReader()


class DataQueue(Generic[_T]):
    """DataQueue is a general-purpose blocking queue with one reader."""

    def __init__(self, loop: asyncio.AbstractEventLoop) -> None:
        self._loop = loop
        self._eof = False
        self._waiter: Optional[asyncio.Future[None]] = None
        self._exception: Optional[BaseException] = None
        self._size = 0
        self._buffer: Deque[Tuple[_T, int]] = collections.deque()

    def __len__(self) -> int:
        return len(self._buffer)

    def is_eof(self) -> bool:
        return self._eof

    def at_eof(self) -> bool:
        return self._eof and not self._buffer

    def exception(self) -> Optional[BaseException]:
        return self._exception

    def set_exception(
        self,
        exc: BaseException,
        exc_cause: BaseException = _EXC_SENTINEL,
    ) -> None:
        self._eof = True
        self._exception = exc

        waiter = self._waiter
        if waiter is not None:
            self._waiter = None
            set_exception(waiter, exc, exc_cause)

    def feed_data(self, data: _T, size: int = 0) -> None:
        self._size += size
        self._buffer.append((data, size))

        waiter = self._waiter
        if waiter is not None:
            self._waiter = None
            set_result(waiter, None)

    def feed_eof(self) -> None:
        self._eof = True

        waiter = self._waiter
        if waiter is not None:
            self._waiter = None
            set_result(waiter, None)

    async def read(self) -> _T:
        if not self._buffer and not self._eof:
            assert not self._waiter
            self._waiter = self._loop.create_future()
            try:
                await self._waiter
            except (asyncio.CancelledError, asyncio.TimeoutError):
                self._waiter = None
                raise

        if self._buffer:
            data, size = self._buffer.popleft()
            self._size -= size
            return data
        else:
            if self._exception is not None:
                raise self._exception
            else:
                raise EofStream

    def __aiter__(self) -> AsyncStreamIterator[_T]:
        return AsyncStreamIterator(self.read)


class FlowControlDataQueue(DataQueue[_T]):
    """FlowControlDataQueue resumes and pauses an underlying stream.

    It is a destination for parsed data.
    """

    def __init__(
        self, protocol: BaseProtocol, limit: int, *, loop: asyncio.AbstractEventLoop
    ) -> None:
        super().__init__(loop=loop)

        self._protocol = protocol
        self._limit = limit * 2

    def feed_data(self, data: _T, size: int = 0) -> None:
        super().feed_data(data, size)

        if self._size > self._limit and not self._protocol._reading_paused:
            self._protocol.pause_reading()

    async def read(self) -> _T:
        try:
            return await super().read()
        finally:
            if self._size < self._limit and self._protocol._reading_paused:
                self._protocol.resume_reading()