import base64
import binascii
import json
import re
import sys
import uuid
import warnings
from collections import deque
from collections.abc import Mapping, Sequence
from types import TracebackType
from typing import (
    TYPE_CHECKING,
    Any,
    Deque,
    Dict,
    Iterator,
    List,
    Optional,
    Tuple,
    Type,
    Union,
    cast,
)
from urllib.parse import parse_qsl, unquote, urlencode

from multidict import CIMultiDict, CIMultiDictProxy

from .abc import AbstractStreamWriter
from .compression_utils import (
    DEFAULT_MAX_DECOMPRESS_SIZE,
    ZLibCompressor,
    ZLibDecompressor,
)
from .hdrs import (
    CONTENT_DISPOSITION,
    CONTENT_ENCODING,
    CONTENT_LENGTH,
    CONTENT_TRANSFER_ENCODING,
    CONTENT_TYPE,
)
from .helpers import CHAR, TOKEN, parse_mimetype, reify
from .http import HeadersParser
from .log import internal_logger
from .payload import (
    JsonPayload,
    LookupError,
    Order,
    Payload,
    StringPayload,
    get_payload,
    payload_type,
)
from .streams import StreamReader

if sys.version_info >= (3, 11):
    from typing import Self
else:
    from typing import TypeVar

    Self = TypeVar("Self", bound="BodyPartReader")

__all__ = (
    "MultipartReader",
    "MultipartWriter",
    "BodyPartReader",
    "BadContentDispositionHeader",
    "BadContentDispositionParam",
    "parse_content_disposition",
    "content_disposition_filename",
)


if TYPE_CHECKING:
    from .client_reqrep import ClientResponse


class BadContentDispositionHeader(RuntimeWarning):
    pass


class BadContentDispositionParam(RuntimeWarning):
    pass


def parse_content_disposition(
    header: Optional[str],
) -> Tuple[Optional[str], Dict[str, str]]:
    def is_token(string: str) -> bool:
        return bool(string) and TOKEN >= set(string)

    def is_quoted(string: str) -> bool:
        return string[0] == string[-1] == '"'

    def is_rfc5987(string: str) -> bool:
        return is_token(string) and string.count("'") == 2

    def is_extended_param(string: str) -> bool:
        return string.endswith("*")

    def is_continuous_param(string: str) -> bool:
        pos = string.find("*") + 1
        if not pos:
            return False
        substring = string[pos:-1] if string.endswith("*") else string[pos:]
        return substring.isdigit()

    def unescape(text: str, *, chars: str = "".join(map(re.escape, CHAR))) -> str:
        return re.sub(f"\\\\([{chars}])", "\\1", text)

    if not header:
        return None, {}

    disptype, *parts = header.split(";")
    if not is_token(disptype):
        warnings.warn(BadContentDispositionHeader(header))
        return None, {}

    params: Dict[str, str] = {}
    while parts:
        item = parts.pop(0)

        if not item:  # To handle trailing semicolons
            warnings.warn(BadContentDispositionHeader(header))
            continue

        if "=" not in item:
            warnings.warn(BadContentDispositionHeader(header))
            return None, {}

        key, value = item.split("=", 1)
        key = key.lower().strip()
        value = value.lstrip()

        if key in params:
            warnings.warn(BadContentDispositionHeader(header))
            return None, {}

        if not is_token(key):
            warnings.warn(BadContentDispositionParam(item))
            continue

        elif is_continuous_param(key):
            if is_quoted(value):
                value = unescape(value[1:-1])
            elif not is_token(value):
                warnings.warn(BadContentDispositionParam(item))
                continue

        elif is_extended_param(key):
            if is_rfc5987(value):
                encoding, _, value = value.split("'", 2)
                encoding = encoding or "utf-8"
            else:
                warnings.warn(BadContentDispositionParam(item))
                continue

            try:
                value = unquote(value, encoding, "strict")
            except UnicodeDecodeError:  # pragma: nocover
                warnings.warn(BadContentDispositionParam(item))
                continue

        else:
            failed = True
            if is_quoted(value):
                failed = False
                value = unescape(value[1:-1].lstrip("\\/"))
            elif is_token(value):
                failed = False
            elif parts:
                # maybe just ; in filename, in any case this is just
                # one case fix, for proper fix we need to redesign parser
                _value = f"{value};{parts[0]}"
                if is_quoted(_value):
                    parts.pop(0)
                    value = unescape(_value[1:-1].lstrip("\\/"))
                    failed = False

            if failed:
                warnings.warn(BadContentDispositionHeader(header))
                return None, {}

        params[key] = value

    return disptype.lower(), params


def content_disposition_filename(
    params: Mapping[str, str], name: str = "filename"
) -> Optional[str]:
    name_suf = "%s*" % name
    if not params:
        return None
    elif name_suf in params:
        return params[name_suf]
    elif name in params:
        return params[name]
    else:
        parts = []
        fnparams = sorted(
            (key, value) for key, value in params.items() if key.startswith(name_suf)
        )
        for num, (key, value) in enumerate(fnparams):
            _, tail = key.split("*", 1)
            if tail.endswith("*"):
                tail = tail[:-1]
            if tail == str(num):
                parts.append(value)
            else:
                break
        if not parts:
            return None
        value = "".join(parts)
        if "'" in value:
            encoding, _, value = value.split("'", 2)
            encoding = encoding or "utf-8"
            return unquote(value, encoding, "strict")
        return value


class MultipartResponseWrapper:
    """Wrapper around the MultipartReader.

    It takes care about
    underlying connection and close it when it needs in.
    """

    def __init__(
        self,
        resp: "ClientResponse",
        stream: "MultipartReader",
    ) -> None:
        self.resp = resp
        self.stream = stream

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

    async def __anext__(
        self,
    ) -> Union["MultipartReader", "BodyPartReader"]:
        part = await self.next()
        if part is None:
            raise StopAsyncIteration
        return part

    def at_eof(self) -> bool:
        """Returns True when all response data had been read."""
        return self.resp.content.at_eof()

    async def next(
        self,
    ) -> Optional[Union["MultipartReader", "BodyPartReader"]]:
        """Emits next multipart reader object."""
        item = await self.stream.next()
        if self.stream.at_eof():
            await self.release()
        return item

    async def release(self) -> None:
        """Release the connection gracefully.

        All remaining content is read to the void.
        """
        await self.resp.release()


class BodyPartReader:
    """Multipart reader for single body part."""

    chunk_size = 8192

    def __init__(
        self,
        boundary: bytes,
        headers: "CIMultiDictProxy[str]",
        content: StreamReader,
        *,
        subtype: str = "mixed",
        default_charset: Optional[str] = None,
        max_decompress_size: int = DEFAULT_MAX_DECOMPRESS_SIZE,
    ) -> None:
        self.headers = headers
        self._boundary = boundary
        self._boundary_len = len(boundary) + 2  # Boundary + \r\n
        self._content = content
        self._default_charset = default_charset
        self._at_eof = False
        self._is_form_data = subtype == "form-data"
        # https://datatracker.ietf.org/doc/html/rfc7578#section-4.8
        length = None if self._is_form_data else self.headers.get(CONTENT_LENGTH, None)
        self._length = int(length) if length is not None else None
        self._read_bytes = 0
        self._unread: Deque[bytes] = deque()
        self._prev_chunk: Optional[bytes] = None
        self._content_eof = 0
        self._cache: Dict[str, Any] = {}
        self._max_decompress_size = max_decompress_size

    def __aiter__(self: Self) -> Self:
        return self

    async def __anext__(self) -> bytes:
        part = await self.next()
        if part is None:
            raise StopAsyncIteration
        return part

    async def next(self) -> Optional[bytes]:
        item = await self.read()
        if not item:
            return None
        return item

    async def read(self, *, decode: bool = False) -> bytes:
        """Reads body part data.

        decode: Decodes data following by encoding
                method from Content-Encoding header. If it missed
                data remains untouched
        """
        if self._at_eof:
            return b""
        data = bytearray()
        while not self._at_eof:
            data.extend(await self.read_chunk(self.chunk_size))
        if decode:
            return await self.decode(data)
        return data

    async def read_chunk(self, size: int = chunk_size) -> bytes:
        """Reads body part content chunk of the specified size.

        size: chunk size
        """
        if self._at_eof:
            return b""
        if self._length:
            chunk = await self._read_chunk_from_length(size)
        else:
            chunk = await self._read_chunk_from_stream(size)

        # For the case of base64 data, we must read a fragment of size with a
        # remainder of 0 by dividing by 4 for string without symbols \n or \r
        encoding = self.headers.get(CONTENT_TRANSFER_ENCODING)
        if encoding and encoding.lower() == "base64":
            stripped_chunk = b"".join(chunk.split())
            remainder = len(stripped_chunk) % 4

            while remainder != 0 and not self.at_eof():
                over_chunk_size = 4 - remainder
                over_chunk = b""

                if self._prev_chunk:
                    over_chunk = self._prev_chunk[:over_chunk_size]
                    self._prev_chunk = self._prev_chunk[len(over_chunk) :]

                if len(over_chunk) != over_chunk_size:
                    over_chunk += await self._content.read(4 - len(over_chunk))

                if not over_chunk:
                    self._at_eof = True

                stripped_chunk += b"".join(over_chunk.split())
                chunk += over_chunk
                remainder = len(stripped_chunk) % 4

        self._read_bytes += len(chunk)
        if self._read_bytes == self._length:
            self._at_eof = True
        if self._at_eof and await self._content.readline() != b"\r\n":
            raise ValueError("Reader did not read all the data or it is malformed")
        return chunk

    async def _read_chunk_from_length(self, size: int) -> bytes:
        # Reads body part content chunk of the specified size.
        # The body part must has Content-Length header with proper value.
        assert self._length is not None, "Content-Length required for chunked read"
        chunk_size = min(size, self._length - self._read_bytes)
        chunk = await self._content.read(chunk_size)
        if self._content.at_eof():
            self._at_eof = True
        return chunk

    async def _read_chunk_from_stream(self, size: int) -> bytes:
        # Reads content chunk of body part with unknown length.
        # The Content-Length header for body part is not necessary.
        assert (
            size >= self._boundary_len
        ), "Chunk size must be greater or equal than boundary length + 2"
        first_chunk = self._prev_chunk is None
        if first_chunk:
            # We need to re-add the CRLF that got removed from headers parsing.
            self._prev_chunk = b"\r\n" + await self._content.read(size)

        chunk = b""
        # content.read() may return less than size, so we need to loop to ensure
        # we have enough data to detect the boundary.
        while len(chunk) < self._boundary_len:
            chunk += await self._content.read(size)
            self._content_eof += int(self._content.at_eof())
            if self._content_eof > 2:
                raise ValueError("Reading after EOF")
            if self._content_eof:
                break
        if len(chunk) > size:
            self._content.unread_data(chunk[size:])
            chunk = chunk[:size]

        assert self._prev_chunk is not None
        window = self._prev_chunk + chunk
        sub = b"\r\n" + self._boundary
        if first_chunk:
            idx = window.find(sub)
        else:
            idx = window.find(sub, max(0, len(self._prev_chunk) - len(sub)))
        if idx >= 0:
            # pushing boundary back to content
            with warnings.catch_warnings():
                warnings.filterwarnings("ignore", category=DeprecationWarning)
                self._content.unread_data(window[idx:])
            self._prev_chunk = self._prev_chunk[:idx]
            chunk = window[len(self._prev_chunk) : idx]
            if not chunk:
                self._at_eof = True
        result = self._prev_chunk[2 if first_chunk else 0 :]  # Strip initial CRLF
        self._prev_chunk = chunk
        return result

    async def readline(self) -> bytes:
        """Reads body part by line by line."""
        if self._at_eof:
            return b""

        if self._unread:
            line = self._unread.popleft()
        else:
            line = await self._content.readline()

        if line.startswith(self._boundary):
            # the very last boundary may not come with \r\n,
            # so set single rules for everyone
            sline = line.rstrip(b"\r\n")
            boundary = self._boundary
            last_boundary = self._boundary + b"--"
            # ensure that we read exactly the boundary, not something alike
            if sline == boundary or sline == last_boundary:
                self._at_eof = True
                self._unread.append(line)
                return b""
        else:
            next_line = await self._content.readline()
            if next_line.startswith(self._boundary):
                line = line[:-2]  # strip CRLF but only once
            self._unread.append(next_line)

        return line

    async def release(self) -> None:
        """Like read(), but reads all the data to the void."""
        if self._at_eof:
            return
        while not self._at_eof:
            await self.read_chunk(self.chunk_size)

    async def text(self, *, encoding: Optional[str] = None) -> str:
        """Like read(), but assumes that body part contains text data."""
        data = await self.read(decode=True)
        # see https://www.w3.org/TR/html5/forms.html#multipart/form-data-encoding-algorithm
        # and https://dvcs.w3.org/hg/xhr/raw-file/tip/Overview.html#dom-xmlhttprequest-send
        encoding = encoding or self.get_charset(default="utf-8")
        return data.decode(encoding)

    async def json(self, *, encoding: Optional[str] = None) -> Optional[Dict[str, Any]]:
        """Like read(), but assumes that body parts contains JSON data."""
        data = await self.read(decode=True)
        if not data:
            return None
        encoding = encoding or self.get_charset(default="utf-8")
        return cast(Dict[str, Any], json.loads(data.decode(encoding)))

    async def form(self, *, encoding: Optional[str] = None) -> List[Tuple[str, str]]:
        """Like read(), but assumes that body parts contain form urlencoded data."""
        data = await self.read(decode=True)
        if not data:
            return []
        if encoding is not None:
            real_encoding = encoding
        else:
            real_encoding = self.get_charset(default="utf-8")
        try:
            decoded_data = data.rstrip().decode(real_encoding)
        except UnicodeDecodeError:
            raise ValueError("data cannot be decoded with %s encoding" % real_encoding)

        return parse_qsl(
            decoded_data,
            keep_blank_values=True,
            encoding=real_encoding,
        )

    def at_eof(self) -> bool:
        """Returns True if the boundary was reached or False otherwise."""
        return self._at_eof

    async def decode(self, data: bytes) -> bytes:
        """Decodes data.

        Decoding is done according the specified Content-Encoding
        or Content-Transfer-Encoding headers value.
        """
        if CONTENT_TRANSFER_ENCODING in self.headers:
            data = self._decode_content_transfer(data)
        # https://datatracker.ietf.org/doc/html/rfc7578#section-4.8
        if not self._is_form_data and CONTENT_ENCODING in self.headers:
            return await self._decode_content(data)
        return data

    async def _decode_content(self, data: bytes) -> bytes:
        encoding = self.headers.get(CONTENT_ENCODING, "").lower()
        if encoding == "identity":
            return data
        if encoding in {"deflate", "gzip"}:
            return await ZLibDecompressor(
                encoding=encoding,
                suppress_deflate_header=True,
            ).decompress(data, max_length=self._max_decompress_size)

        raise RuntimeError(f"unknown content encoding: {encoding}")

    def _decode_content_transfer(self, data: bytes) -> bytes:
        encoding = self.headers.get(CONTENT_TRANSFER_ENCODING, "").lower()

        if encoding == "base64":
            return base64.b64decode(data)
        elif encoding == "quoted-printable":
            return binascii.a2b_qp(data)
        elif encoding in ("binary", "8bit", "7bit"):
            return data
        else:
            raise RuntimeError(f"unknown content transfer encoding: {encoding}")

    def get_charset(self, default: str) -> str:
        """Returns charset parameter from Content-Type header or default."""
        ctype = self.headers.get(CONTENT_TYPE, "")
        mimetype = parse_mimetype(ctype)
        return mimetype.parameters.get("charset", self._default_charset or default)

    @reify
    def name(self) -> Optional[str]:
        """Returns name specified in Content-Disposition header.

        If the header is missing or malformed, returns None.
        """
        _, params = parse_content_disposition(self.headers.get(CONTENT_DISPOSITION))
        return content_disposition_filename(params, "name")

    @reify
    def filename(self) -> Optional[str]:
        """Returns filename specified in Content-Disposition header.

        Returns None if the header is missing or malformed.
        """
        _, params = parse_content_disposition(self.headers.get(CONTENT_DISPOSITION))
        return content_disposition_filename(params, "filename")


@payload_type(BodyPartReader, order=Order.try_first)
class BodyPartReaderPayload(Payload):
    _value: BodyPartReader
    # _autoclose = False (inherited) - Streaming reader that may have resources

    def __init__(self, value: BodyPartReader, *args: Any, **kwargs: Any) -> None:
        super().__init__(value, *args, **kwargs)

        params: Dict[str, str] = {}
        if value.name is not None:
            params["name"] = value.name
        if value.filename is not None:
            params["filename"] = value.filename

        if params:
            self.set_content_disposition("attachment", True, **params)

    def decode(self, encoding: str = "utf-8", errors: str = "strict") -> str:
        raise TypeError("Unable to decode.")

    async def as_bytes(self, encoding: str = "utf-8", errors: str = "strict") -> bytes:
        """Raises TypeError as body parts should be consumed via write().

        This is intentional: BodyPartReader payloads are designed for streaming
        large data (potentially gigabytes) and must be consumed only once via
        the write() method to avoid memory exhaustion. They cannot be buffered
        in memory for reuse.
        """
        raise TypeError("Unable to read body part as bytes. Use write() to consume.")

    async def write(self, writer: AbstractStreamWriter) -> None:
        field = self._value
        chunk = await field.read_chunk(size=2**16)
        while chunk:
            await writer.write(await field.decode(chunk))
            chunk = await field.read_chunk(size=2**16)


class MultipartReader:
    """Multipart body reader."""

    #: Response wrapper, used when multipart readers constructs from response.
    response_wrapper_cls = MultipartResponseWrapper
    #: Multipart reader class, used to handle multipart/* body parts.
    #: None points to type(self)
    multipart_reader_cls: Optional[Type["MultipartReader"]] = None
    #: Body part reader class for non multipart/* content types.
    part_reader_cls = BodyPartReader

    def __init__(self, headers: Mapping[str, str], content: StreamReader) -> None:
        self._mimetype = parse_mimetype(headers[CONTENT_TYPE])
        assert self._mimetype.type == "multipart", "multipart/* content type expected"
        if "boundary" not in self._mimetype.parameters:
            raise ValueError(
                "boundary missed for Content-Type: %s" % headers[CONTENT_TYPE]
            )

        self.headers = headers
        self._boundary = ("--" + self._get_boundary()).encode()
        self._content = content
        self._default_charset: Optional[str] = None
        self._last_part: Optional[Union["MultipartReader", BodyPartReader]] = None
        self._at_eof = False
        self._at_bof = True
        self._unread: List[bytes] = []

    def __aiter__(self: Self) -> Self:
        return self

    async def __anext__(
        self,
    ) -> Optional[Union["MultipartReader", BodyPartReader]]:
        part = await self.next()
        if part is None:
            raise StopAsyncIteration
        return part

    @classmethod
    def from_response(
        cls,
        response: "ClientResponse",
    ) -> MultipartResponseWrapper:
        """Constructs reader instance from HTTP response.

        :param response: :class:`~aiohttp.client.ClientResponse` instance
        """
        obj = cls.response_wrapper_cls(
            response, cls(response.headers, response.content)
        )
        return obj

    def at_eof(self) -> bool:
        """Returns True if the final boundary was reached, false otherwise."""
        return self._at_eof

    async def next(
        self,
    ) -> Optional[Union["MultipartReader", BodyPartReader]]:
        """Emits the next multipart body part."""
        # So, if we're at BOF, we need to skip till the boundary.
        if self._at_eof:
            return None
        await self._maybe_release_last_part()
        if self._at_bof:
            await self._read_until_first_boundary()
            self._at_bof = False
        else:
            await self._read_boundary()
        if self._at_eof:  # we just read the last boundary, nothing to do there
            return None

        part = await self.fetch_next_part()
        # https://datatracker.ietf.org/doc/html/rfc7578#section-4.6
        if (
            self._last_part is None
            and self._mimetype.subtype == "form-data"
            and isinstance(part, BodyPartReader)
        ):
            _, params = parse_content_disposition(part.headers.get(CONTENT_DISPOSITION))
            if params.get("name") == "_charset_":
                # Longest encoding in https://encoding.spec.whatwg.org/encodings.json
                # is 19 characters, so 32 should be more than enough for any valid encoding.
                charset = await part.read_chunk(32)
                if len(charset) > 31:
                    raise RuntimeError("Invalid default charset")
                self._default_charset = charset.strip().decode()
                part = await self.fetch_next_part()
        self._last_part = part
        return self._last_part

    async def release(self) -> None:
        """Reads all the body parts to the void till the final boundary."""
        while not self._at_eof:
            item = await self.next()
            if item is None:
                break
            await item.release()

    async def fetch_next_part(
        self,
    ) -> Union["MultipartReader", BodyPartReader]:
        """Returns the next body part reader."""
        headers = await self._read_headers()
        return self._get_part_reader(headers)

    def _get_part_reader(
        self,
        headers: "CIMultiDictProxy[str]",
    ) -> Union["MultipartReader", BodyPartReader]:
        """Dispatches the response by the `Content-Type` header.

        Returns a suitable reader instance.

        :param dict headers: Response headers
        """
        ctype = headers.get(CONTENT_TYPE, "")
        mimetype = parse_mimetype(ctype)

        if mimetype.type == "multipart":
            if self.multipart_reader_cls is None:
                return type(self)(headers, self._content)
            return self.multipart_reader_cls(headers, self._content)
        else:
            return self.part_reader_cls(
                self._boundary,
                headers,
                self._content,
                subtype=self._mimetype.subtype,
                default_charset=self._default_charset,
            )

    def _get_boundary(self) -> str:
        boundary = self._mimetype.parameters["boundary"]
        if len(boundary) > 70:
            raise ValueError("boundary %r is too long (70 chars max)" % boundary)

        return boundary

    async def _readline(self) -> bytes:
        if self._unread:
            return self._unread.pop()
        return await self._content.readline()

    async def _read_until_first_boundary(self) -> None:
        while True:
            chunk = await self._readline()
            if chunk == b"":
                raise ValueError(
                    "Could not find starting boundary %r" % (self._boundary)
                )
            chunk = chunk.rstrip()
            if chunk == self._boundary:
                return
            elif chunk == self._boundary + b"--":
                self._at_eof = True
                return

    async def _read_boundary(self) -> None:
        chunk = (await self._readline()).rstrip()
        if chunk == self._boundary:
            pass
        elif chunk == self._boundary + b"--":
            self._at_eof = True
            epilogue = await self._readline()
            next_line = await self._readline()

            # the epilogue is expected and then either the end of input or the
            # parent multipart boundary, if the parent boundary is found then
            # it should be marked as unread and handed to the parent for
            # processing
            if next_line[:2] == b"--":
                self._unread.append(next_line)
            # otherwise the request is likely missing an epilogue and both
            # lines should be passed to the parent for processing
            # (this handles the old behavior gracefully)
            else:
                self._unread.extend([next_line, epilogue])
        else:
            raise ValueError(f"Invalid boundary {chunk!r}, expected {self._boundary!r}")

    async def _read_headers(self) -> "CIMultiDictProxy[str]":
        lines = []
        while True:
            chunk = await self._content.readline()
            chunk = chunk.rstrip(b"\r\n")
            lines.append(chunk)
            if not chunk:
                break
        parser = HeadersParser()
        headers, raw_headers = parser.parse_headers(lines)
        return headers

    async def _maybe_release_last_part(self) -> None:
        """Ensures that the last read body part is read completely."""
        if self._last_part is not None:
            if not self._last_part.at_eof():
                await self._last_part.release()
            self._unread.extend(self._last_part._unread)
            self._last_part = None


_Part = Tuple[Payload, str, str]


class MultipartWriter(Payload):
    """Multipart body writer."""

    _value: None
    # _consumed = False (inherited) - Can be encoded multiple times
    _autoclose = True  # No file handles, just collects parts in memory

    def __init__(self, subtype: str = "mixed", boundary: Optional[str] = None) -> None:
        boundary = boundary if boundary is not None else uuid.uuid4().hex
        # The underlying Payload API demands a str (utf-8), not bytes,
        # so we need to ensure we don't lose anything during conversion.
        # As a result, require the boundary to be ASCII only.
        # In both situations.

        try:
            self._boundary = boundary.encode("ascii")
        except UnicodeEncodeError:
            raise ValueError("boundary should contain ASCII only chars") from None
        ctype = f"multipart/{subtype}; boundary={self._boundary_value}"

        super().__init__(None, content_type=ctype)

        self._parts: List[_Part] = []
        self._is_form_data = subtype == "form-data"

    def __enter__(self) -> "MultipartWriter":
        return self

    def __exit__(
        self,
        exc_type: Optional[Type[BaseException]],
        exc_val: Optional[BaseException],
        exc_tb: Optional[TracebackType],
    ) -> None:
        pass

    def __iter__(self) -> Iterator[_Part]:
        return iter(self._parts)

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

    def __bool__(self) -> bool:
        return True

    _valid_tchar_regex = re.compile(rb"\A[!#$%&'*+\-.^_`|~\w]+\Z")
    _invalid_qdtext_char_regex = re.compile(rb"[\x00-\x08\x0A-\x1F\x7F]")

    @property
    def _boundary_value(self) -> str:
        """Wrap boundary parameter value in quotes, if necessary.

        Reads self.boundary and returns a unicode string.
        """
        # Refer to RFCs 7231, 7230, 5234.
        #
        # parameter      = token "=" ( token / quoted-string )
        # token          = 1*tchar
        # quoted-string  = DQUOTE *( qdtext / quoted-pair ) DQUOTE
        # qdtext         = HTAB / SP / %x21 / %x23-5B / %x5D-7E / obs-text
        # obs-text       = %x80-FF
        # quoted-pair    = "\" ( HTAB / SP / VCHAR / obs-text )
        # tchar          = "!" / "#" / "$" / "%" / "&" / "'" / "*"
        #                  / "+" / "-" / "." / "^" / "_" / "`" / "|" / "~"
        #                  / DIGIT / ALPHA
        #                  ; any VCHAR, except delimiters
        # VCHAR           = %x21-7E
        value = self._boundary
        if re.match(self._valid_tchar_regex, value):
            return value.decode("ascii")  # cannot fail

        if re.search(self._invalid_qdtext_char_regex, value):
            raise ValueError("boundary value contains invalid characters")

        # escape %x5C and %x22
        quoted_value_content = value.replace(b"\\", b"\\\\")
        quoted_value_content = quoted_value_content.replace(b'"', b'\\"')

        return '"' + quoted_value_content.decode("ascii") + '"'

    @property
    def boundary(self) -> str:
        return self._boundary.decode("ascii")

    def append(self, obj: Any, headers: Optional[Mapping[str, str]] = None) -> Payload:
        if headers is None:
            headers = CIMultiDict()

        if isinstance(obj, Payload):
            obj.headers.update(headers)
            return self.append_payload(obj)
        else:
            try:
                payload = get_payload(obj, headers=headers)
            except LookupError:
                raise TypeError("Cannot create payload from %r" % obj)
            else:
                return self.append_payload(payload)

    def append_payload(self, payload: Payload) -> Payload:
        """Adds a new body part to multipart writer."""
        encoding: Optional[str] = None
        te_encoding: Optional[str] = None
        if self._is_form_data:
            # https://datatracker.ietf.org/doc/html/rfc7578#section-4.7
            # https://datatracker.ietf.org/doc/html/rfc7578#section-4.8
            assert (
                not {CONTENT_ENCODING, CONTENT_LENGTH, CONTENT_TRANSFER_ENCODING}
                & payload.headers.keys()
            )
            # Set default Content-Disposition in case user doesn't create one
            if CONTENT_DISPOSITION not in payload.headers:
                name = f"section-{len(self._parts)}"
                payload.set_content_disposition("form-data", name=name)
        else:
            # compression
            encoding = payload.headers.get(CONTENT_ENCODING, "").lower()
            if encoding and encoding not in ("deflate", "gzip", "identity"):
                raise RuntimeError(f"unknown content encoding: {encoding}")
            if encoding == "identity":
                encoding = None

            # te encoding
            te_encoding = payload.headers.get(CONTENT_TRANSFER_ENCODING, "").lower()
            if te_encoding not in ("", "base64", "quoted-printable", "binary"):
                raise RuntimeError(f"unknown content transfer encoding: {te_encoding}")
            if te_encoding == "binary":
                te_encoding = None

            # size
            size = payload.size
            if size is not None and not (encoding or te_encoding):
                payload.headers[CONTENT_LENGTH] = str(size)

        self._parts.append((payload, encoding, te_encoding))  # type: ignore[arg-type]
        return payload

    def append_json(
        self, obj: Any, headers: Optional[Mapping[str, str]] = None
    ) -> Payload:
        """Helper to append JSON part."""
        if headers is None:
            headers = CIMultiDict()

        return self.append_payload(JsonPayload(obj, headers=headers))

    def append_form(
        self,
        obj: Union[Sequence[Tuple[str, str]], Mapping[str, str]],
        headers: Optional[Mapping[str, str]] = None,
    ) -> Payload:
        """Helper to append form urlencoded part."""
        assert isinstance(obj, (Sequence, Mapping))

        if headers is None:
            headers = CIMultiDict()

        if isinstance(obj, Mapping):
            obj = list(obj.items())
        data = urlencode(obj, doseq=True)

        return self.append_payload(
            StringPayload(
                data, headers=headers, content_type="application/x-www-form-urlencoded"
            )
        )

    @property
    def size(self) -> Optional[int]:
        """Size of the payload."""
        total = 0
        for part, encoding, te_encoding in self._parts:
            part_size = part.size
            if encoding or te_encoding or part_size is None:
                return None

            total += int(
                2
                + len(self._boundary)
                + 2
                + part_size  # b'--'+self._boundary+b'\r\n'
                + len(part._binary_headers)
                + 2  # b'\r\n'
            )

        total += 2 + len(self._boundary) + 4  # b'--'+self._boundary+b'--\r\n'
        return total

    def decode(self, encoding: str = "utf-8", errors: str = "strict") -> str:
        """Return string representation of the multipart data.

        WARNING: This method may do blocking I/O if parts contain file payloads.
        It should not be called in the event loop. Use as_bytes().decode() instead.
        """
        return "".join(
            "--"
            + self.boundary
            + "\r\n"
            + part._binary_headers.decode(encoding, errors)
            + part.decode()
            for part, _e, _te in self._parts
        )

    async def as_bytes(self, encoding: str = "utf-8", errors: str = "strict") -> bytes:
        """Return bytes representation of the multipart data.

        This method is async-safe and calls as_bytes on underlying payloads.
        """
        parts: List[bytes] = []

        # Process each part
        for part, _e, _te in self._parts:
            # Add boundary
            parts.append(b"--" + self._boundary + b"\r\n")

            # Add headers
            parts.append(part._binary_headers)

            # Add payload content using as_bytes for async safety
            part_bytes = await part.as_bytes(encoding, errors)
            parts.append(part_bytes)

            # Add trailing CRLF
            parts.append(b"\r\n")

        # Add closing boundary
        parts.append(b"--" + self._boundary + b"--\r\n")

        return b"".join(parts)

    async def write(
        self, writer: AbstractStreamWriter, close_boundary: bool = True
    ) -> None:
        """Write body."""
        for part, encoding, te_encoding in self._parts:
            if self._is_form_data:
                # https://datatracker.ietf.org/doc/html/rfc7578#section-4.2
                assert CONTENT_DISPOSITION in part.headers
                assert "name=" in part.headers[CONTENT_DISPOSITION]

            await writer.write(b"--" + self._boundary + b"\r\n")
            await writer.write(part._binary_headers)

            if encoding or te_encoding:
                w = MultipartPayloadWriter(writer)
                if encoding:
                    w.enable_compression(encoding)
                if te_encoding:
                    w.enable_encoding(te_encoding)
                await part.write(w)  # type: ignore[arg-type]
                await w.write_eof()
            else:
                await part.write(writer)

            await writer.write(b"\r\n")

        if close_boundary:
            await writer.write(b"--" + self._boundary + b"--\r\n")

    async def close(self) -> None:
        """
        Close all part payloads that need explicit closing.

        IMPORTANT: This method must not await anything that might not finish
        immediately, as it may be called during cleanup/cancellation. Schedule
        any long-running operations without awaiting them.
        """
        if self._consumed:
            return
        self._consumed = True

        # Close all parts that need explicit closing
        # We catch and log exceptions to ensure all parts get a chance to close
        # we do not use asyncio.gather() here because we are not allowed
        # to suspend given we may be called during cleanup
        for idx, (part, _, _) in enumerate(self._parts):
            if not part.autoclose and not part.consumed:
                try:
                    await part.close()
                except Exception as exc:
                    internal_logger.error(
                        "Failed to close multipart part %d: %s", idx, exc, exc_info=True
                    )


class MultipartPayloadWriter:
    def __init__(self, writer: AbstractStreamWriter) -> None:
        self._writer = writer
        self._encoding: Optional[str] = None
        self._compress: Optional[ZLibCompressor] = None
        self._encoding_buffer: Optional[bytearray] = None

    def enable_encoding(self, encoding: str) -> None:
        if encoding == "base64":
            self._encoding = encoding
            self._encoding_buffer = bytearray()
        elif encoding == "quoted-printable":
            self._encoding = "quoted-printable"

    def enable_compression(
        self, encoding: str = "deflate", strategy: Optional[int] = None
    ) -> None:
        self._compress = ZLibCompressor(
            encoding=encoding,
            suppress_deflate_header=True,
            strategy=strategy,
        )

    async def write_eof(self) -> None:
        if self._compress is not None:
            chunk = self._compress.flush()
            if chunk:
                self._compress = None
                await self.write(chunk)

        if self._encoding == "base64":
            if self._encoding_buffer:
                await self._writer.write(base64.b64encode(self._encoding_buffer))

    async def write(self, chunk: bytes) -> None:
        if self._compress is not None:
            if chunk:
                chunk = await self._compress.compress(chunk)
                if not chunk:
                    return

        if self._encoding == "base64":
            buf = self._encoding_buffer
            assert buf is not None
            buf.extend(chunk)

            if buf:
                div, mod = divmod(len(buf), 3)
                enc_chunk, self._encoding_buffer = (buf[: div * 3], buf[div * 3 :])
                if enc_chunk:
                    b64chunk = base64.b64encode(enc_chunk)
                    await self._writer.write(b64chunk)
        elif self._encoding == "quoted-printable":
            await self._writer.write(binascii.b2a_qp(chunk))
        else:
            await self._writer.write(chunk)