ok
Direktori : /lib/python3.6/site-packages/requests_toolbelt/utils/ |
Current File : //lib/python3.6/site-packages/requests_toolbelt/utils/dump.py |
"""This module provides functions for dumping information about responses.""" import collections from requests import compat __all__ = ('dump_response', 'dump_all') HTTP_VERSIONS = { 9: b'0.9', 10: b'1.0', 11: b'1.1', } _PrefixSettings = collections.namedtuple('PrefixSettings', ['request', 'response']) class PrefixSettings(_PrefixSettings): def __new__(cls, request, response): request = _coerce_to_bytes(request) response = _coerce_to_bytes(response) return super(PrefixSettings, cls).__new__(cls, request, response) def _get_proxy_information(response): if getattr(response.connection, 'proxy_manager', False): proxy_info = {} request_url = response.request.url if request_url.startswith('https://'): proxy_info['method'] = 'CONNECT' proxy_info['request_path'] = request_url return proxy_info return None def _format_header(name, value): return (_coerce_to_bytes(name) + b': ' + _coerce_to_bytes(value) + b'\r\n') def _build_request_path(url, proxy_info): uri = compat.urlparse(url) proxy_url = proxy_info.get('request_path') if proxy_url is not None: request_path = _coerce_to_bytes(proxy_url) return request_path, uri request_path = _coerce_to_bytes(uri.path) if uri.query: request_path += b'?' + _coerce_to_bytes(uri.query) return request_path, uri def _dump_request_data(request, prefixes, bytearr, proxy_info=None): if proxy_info is None: proxy_info = {} prefix = prefixes.request method = _coerce_to_bytes(proxy_info.pop('method', request.method)) request_path, uri = _build_request_path(request.url, proxy_info) # <prefix><METHOD> <request-path> HTTP/1.1 bytearr.extend(prefix + method + b' ' + request_path + b' HTTP/1.1\r\n') # <prefix>Host: <request-host> OR host header specified by user headers = request.headers.copy() host_header = _coerce_to_bytes(headers.pop('Host', uri.netloc)) bytearr.extend(prefix + b'Host: ' + host_header + b'\r\n') for name, value in headers.items(): bytearr.extend(prefix + _format_header(name, value)) bytearr.extend(prefix + b'\r\n') if request.body: if isinstance(request.body, compat.basestring): bytearr.extend(prefix + _coerce_to_bytes(request.body)) else: # In the event that the body is a file-like object, let's not try # to read everything into memory. bytearr.extend(b'<< Request body is not a string-like type >>') bytearr.extend(b'\r\n') def _dump_response_data(response, prefixes, bytearr): prefix = prefixes.response # Let's interact almost entirely with urllib3's response raw = response.raw # Let's convert the version int from httplib to bytes version_str = HTTP_VERSIONS.get(raw.version, b'?') # <prefix>HTTP/<version_str> <status_code> <reason> bytearr.extend(prefix + b'HTTP/' + version_str + b' ' + str(raw.status).encode('ascii') + b' ' + _coerce_to_bytes(response.reason) + b'\r\n') headers = raw.headers for name in headers.keys(): for value in headers.getlist(name): bytearr.extend(prefix + _format_header(name, value)) bytearr.extend(prefix + b'\r\n') bytearr.extend(response.content) def _coerce_to_bytes(data): if not isinstance(data, bytes) and hasattr(data, 'encode'): data = data.encode('utf-8') # Don't bail out with an exception if data is None return data if data is not None else b'' def dump_response(response, request_prefix=b'< ', response_prefix=b'> ', data_array=None): """Dump a single request-response cycle's information. This will take a response object and dump only the data that requests can see for that single request-response cycle. Example:: import requests from requests_toolbelt.utils import dump resp = requests.get('https://api.github.com/users/sigmavirus24') data = dump.dump_response(resp) print(data.decode('utf-8')) :param response: The response to format :type response: :class:`requests.Response` :param request_prefix: (*optional*) Bytes to prefix each line of the request data :type request_prefix: :class:`bytes` :param response_prefix: (*optional*) Bytes to prefix each line of the response data :type response_prefix: :class:`bytes` :param data_array: (*optional*) Bytearray to which we append the request-response cycle data :type data_array: :class:`bytearray` :returns: Formatted bytes of request and response information. :rtype: :class:`bytearray` """ data = data_array if data_array is not None else bytearray() prefixes = PrefixSettings(request_prefix, response_prefix) if not hasattr(response, 'request'): raise ValueError('Response has no associated request') proxy_info = _get_proxy_information(response) _dump_request_data(response.request, prefixes, data, proxy_info=proxy_info) _dump_response_data(response, prefixes, data) return data def dump_all(response, request_prefix=b'< ', response_prefix=b'> '): """Dump all requests and responses including redirects. This takes the response returned by requests and will dump all request-response pairs in the redirect history in order followed by the final request-response. Example:: import requests from requests_toolbelt.utils import dump resp = requests.get('https://httpbin.org/redirect/5') data = dump.dump_all(resp) print(data.decode('utf-8')) :param response: The response to format :type response: :class:`requests.Response` :param request_prefix: (*optional*) Bytes to prefix each line of the request data :type request_prefix: :class:`bytes` :param response_prefix: (*optional*) Bytes to prefix each line of the response data :type response_prefix: :class:`bytes` :returns: Formatted bytes of request and response information. :rtype: :class:`bytearray` """ data = bytearray() history = list(response.history[:]) history.append(response) for response in history: dump_response(response, request_prefix, response_prefix, data) return data