sanic/sanic/request.py

144 lines
4.6 KiB
Python
Raw Normal View History

from cgi import parse_header
from collections import namedtuple
from httptools import parse_url
2016-10-03 08:27:54 +01:00
from urllib.parse import parse_qs
2016-10-06 09:23:52 +01:00
from ujson import loads as json_loads
2016-10-03 08:27:54 +01:00
from .log import log
2016-10-09 23:28:31 +01:00
class RequestParameters(dict):
"""
Hosts a dict with lists as values where get returns the first
value of the list and getlist returns the whole shebang
"""
2016-10-09 23:28:31 +01:00
def __init__(self, *args, **kwargs):
self.super = super()
self.super.__init__(*args, **kwargs)
2016-10-09 23:28:31 +01:00
def get(self, name, default=None):
values = self.super.get(name)
return values[0] if values else default
2016-10-09 23:28:31 +01:00
def getlist(self, name, default=None):
return self.super.get(name, default)
2016-10-03 08:27:54 +01:00
class Request:
2016-10-09 23:28:31 +01:00
__slots__ = (
'url', 'headers', 'version', 'method',
'query_string', 'body',
'parsed_json', 'parsed_args', 'parsed_form', 'parsed_files',
2016-10-09 23:28:31 +01:00
)
2016-10-03 08:27:54 +01:00
def __init__(self, url_bytes, headers, version, method):
# TODO: Content-Encoding detection
url_parsed = parse_url(url_bytes)
self.url = url_parsed.path.decode('utf-8')
2016-10-03 08:27:54 +01:00
self.headers = headers
self.version = version
self.method = method
self.query_string = url_parsed.query.decode('utf-8') if url_parsed.query else None
2016-10-03 08:27:54 +01:00
# Init but do not inhale
self.body = None
self.parsed_json = None
2016-10-09 23:28:31 +01:00
self.parsed_form = None
self.parsed_files = None
2016-10-03 08:27:54 +01:00
self.parsed_args = None
@property
def json(self):
if not self.parsed_json:
2016-10-09 23:28:31 +01:00
try:
self.parsed_json = json_loads(self.body)
except:
pass
2016-10-03 08:27:54 +01:00
2016-10-09 23:28:31 +01:00
return self.parsed_json
@property
def form(self):
if self.parsed_form is None:
self.parsed_form = {}
self.parsed_files = {}
content_type, parameters = parse_header(self.headers.get('Content-Type'))
2016-10-09 23:28:31 +01:00
try:
if content_type is None or content_type == 'application/x-www-form-urlencoded':
self.parsed_form = RequestParameters(parse_qs(self.body.decode('utf-8')))
elif content_type == 'multipart/form-data':
# TODO: Stream this instead of reading to/from memory
boundary = parameters['boundary'].encode('utf-8')
self.parsed_form, self.parsed_files = parse_multipart_form(self.body, boundary)
except Exception as e:
log.exception(e)
2016-10-09 23:28:31 +01:00
pass
return self.parsed_form
2016-10-03 08:27:54 +01:00
@property
def files(self):
if self.parsed_files is None:
_ = self.form # compute form to get files
return self.parsed_files
2016-10-03 08:27:54 +01:00
@property
def args(self):
if self.parsed_args is None:
if self.query_string:
2016-10-09 23:28:31 +01:00
self.parsed_args = RequestParameters(parse_qs(self.query_string))
2016-10-03 08:27:54 +01:00
else:
self.parsed_args = {}
return self.parsed_args
File = namedtuple('File', ['type', 'body', 'name'])
def parse_multipart_form(body, boundary):
"""
Parses a request body and returns fields and files
:param body: Bytes request body
:param boundary: Bytes multipart boundary
:return: fields (dict), files (dict)
"""
files = {}
fields = {}
form_parts = body.split(boundary)
for form_part in form_parts[1:-1]:
file_name = None
file_type = None
field_name = None
line_index = 2
line_end_index = 0
while not line_end_index == -1:
line_end_index = form_part.find(b'\r\n', line_index)
form_line = form_part[line_index:line_end_index].decode('utf-8')
line_index = line_end_index + 2
if not form_line:
break
colon_index = form_line.index(':')
form_header_field = form_line[0:colon_index]
form_header_value, form_parameters = parse_header(form_line[colon_index + 2:])
if form_header_field == 'Content-Disposition':
if 'filename' in form_parameters:
file_name = form_parameters['filename']
field_name = form_parameters.get('name')
elif form_header_field == 'Content-Type':
file_type = form_header_value
post_data = form_part[line_index:-4]
if file_name or file_type:
files[field_name] = File(type=file_type, name=file_name, body=post_data)
else:
fields[field_name] = post_data.decode('utf-8')
return fields, files