multipart 1.1.0

Installation

pip install multipart

Features

• Pure python single file module with no dependencies.

• 100% test coverage. Tested with inputs as seen from actual browsers and HTTP clients.

• Parses multiple GB/s on modern hardware (quick tests, no proper benchmark).

• Quickly rejects malicious or broken inputs and emits useful error messages.

• Enforces configurable memory and disk resource limits to prevent DoS attacks.

Limitations: This parser implements multipart/form-data as it is used by actual modern browsers and HTTP clients, which means:

• Just multipart/form-data, not suitable for email parsing.

• No multipart/mixed support (deprecated in RFC 7578).

• No base64 or quoted-printable transfer encoding (deprecated in RFC 7578).

• No encoded-word or name=_charset_ encoding markers (discouraged in RFC 7578).

• No support for clearly broken input (e.g. invalid line breaks or header names).

Usage and examples

For WSGI application developers we strongly suggest using the parse_form_data helper function. It accepts a WSGI environ dictionary and parses both types of form submission (multipart/form-data and application/x-www-form-urlencoded) based on the actual content type of the request. You’ll get two MultiDict instances in return, one for text fields and the other for file uploads:

from multipart import parse_form_data

def wsgi(environ, start_response):
  if environ["REQUEST_METHOD"] == "POST":
    forms, files = parse_form_data(environ)

    title = forms["title"]    # string
    upload = files["upload"]  # MultipartPart
    upload.save_as(...)

The parse_form_data helper function internally uses MultipartParser, a streaming parser that reads from a multipart/form-data encoded binary data stream and emits MultipartPart instances as soon as a part is fully parsed. This is most useful if you want to consume the individual parts as soon as they arrive, instead of waiting for the entire request to be parsed:

from multipart import parse_options_header, MultipartParser

def wsgi(environ, start_response):
  assert environ["REQUEST_METHOD"] == "POST"
  ctype, copts = parse_options_header(environ.get("CONTENT_TYPE", ""))
  boundary = copts.get("boundary")
  charset = copts.get("charset", "utf8")
  assert ctype == "multipart/form-data"

  parser = MultipartParser(environ["wsgi.input"], boundary, charset)
  for part in parser:
    if part.filename:
      print(f"{part.name}: File upload ({part.size} bytes)")
      part.save_as(...)
    elif part.size < 1024:
      print(f"{part.name}: Text field ({part.value!r})")
    else:
      print(f"{part.name}: Test field, but too big to print :/")

The MultipartParser handles IO and file buffering for you, but does so using blocking APIs. If you need absolute control over the parsing process and want to avoid blocking IO at all cost, then have a look at PushMultipartParser, the low-level non-blocking incremental multipart/form-data parser that powers all the other parsers in this library:

from multipart import PushMultipartParser, MultipartSegment

async def process_multipart(reader: asyncio.StreamReader, boundary: str):
  with PushMultipartParser(boundary) as parser:
    while not parser.closed:
      chunk = await reader.read(1024*64)
      for result in parser.parse(chunk):
        if isinstance(result, MultipartSegment):
          print(f"== Start of segment: {result.name}")
          for header, value in result.headerlist:
            print(f"{header}: {value}")
        elif result:  # Result is a non-empty bytearray
          print(f"[received {len(result)} bytes of data]")
        else:         # Result is None
          print(f"== End of segment")

Changelog

• 1.1

  • Some of these fixes changed behavior to match documentation or specification, none of them should be a surprise. Existing apps should be able to upgrade without change.

  • fix: Fail faster on input with invalid line breaks (#55)

  • fix: Allow empty segment names (#56)

  • fix: Avoid ResourceWarning when using parse_form_data (#57)

  • fix: MultipartPart now always has a sensible content type.

  • fix: Actually check parser state on context manager exit.

  • fix: Honor Content-Length header, if present.

  • perf: Reduce overhead for small segments (-21%)

  • perf: Reduce write overhead for large uploads (-2%)

• 1.0

  • A completely new, fast, non-blocking PushMultipartParser parser, which now serves as the basis for all other parsers.

  • The new parser is stricter and rejects clearly broken input quicker, even in non-strict mode (e.g. invalid line breaks or header names). This should not affect data sent by actual browsers or HTTP clients.

  • Default charset for MultipartParser headers and text fields changed to utf8, as recommended by W3C HTTP.

  • Default disk and memory limits for MultipartParser increased, but multiple other limits added for finer control. Check if the the new defaults still fit your needs.

  • Undocumented APIs deprecated or removed, some of which were not strictly private. This includes parameters for MultipartParser and some MultipartPart methods, but those should not be used by anyone but the parser itself.

• 0.2.5

  • Don’t test semicolon separators in urlencoded data (#33)

  • Add python-requires directive, indicating Python 3.5 or later is required and preventing older Pythons from attempting to download this version (#32)

  • Add official support for Python 3.10-3.12 (#38, #48)

  • Default value of copy_file should be 2 ** 16, not 2 * 16 (#41)

  • Update URL for Bottle (#42)

• 0.2.4

  • Consistently decode non-utf8 URL-encoded form-data

• 0.2.3

  • Import MutableMapping from collections.abc (#23)

  • Fix a few more ResourceWarnings in the test suite (#24)

  • Allow stream to contain data before first boundary (#25)

• 0.2.2

  • Fix #21 ResourceWarnings on Python 3

• 0.2.1

  • Fix #20 empty payload

• 0.2

  • Dropped support for Python versions below 3.6. Stay on 0.1 if you need Python 2.5+ support.

• 0.1

  • First release