daphne/tests/http_strategies.py
2018-02-07 14:15:28 -08:00

138 lines
3.6 KiB
Python

import string
from urllib import parse
from hypothesis import strategies
HTTP_METHODS = ["OPTIONS", "GET", "HEAD", "POST", "PUT", "DELETE", "TRACE", "CONNECT"]
# Unicode characters of the "Letter" category
letters = strategies.characters(whitelist_categories=("Lu", "Ll", "Lt", "Lm", "Lo", "Nl"))
def http_method():
return strategies.sampled_from(HTTP_METHODS)
def _http_path_portion():
alphabet = string.ascii_letters + string.digits + "-._~"
return strategies.text(min_size=1, average_size=10, max_size=128, alphabet=alphabet)
def http_path():
"""
Returns a URL path (not encoded).
"""
return strategies.lists(
_http_path_portion(),
min_size=0,
max_size=10,
).map(lambda s: "/" + "/".join(s))
def http_body():
"""
Returns random binary body data.
"""
return strategies.binary(min_size=0, average_size=600, max_size=1500)
def valid_bidi(value):
"""
Rejects strings which nonsensical Unicode text direction flags.
Relying on random Unicode characters means that some combinations don't make sense, from a
direction of text point of view. This little helper just rejects those.
"""
try:
value.encode("idna")
except UnicodeError:
return False
else:
return True
def _domain_label():
return strategies.text(
alphabet=letters,
min_size=1,
average_size=6,
max_size=63,
).filter(valid_bidi)
def international_domain_name():
"""
Returns a byte string of a domain name, IDNA-encoded.
"""
return strategies.lists(
_domain_label(),
min_size=2,
average_size=2,
).map(lambda s: (".".join(s)).encode("idna"))
def _query_param():
return strategies.text(
alphabet=letters,
min_size=1,
average_size=10,
max_size=255,
).map(lambda s: s.encode("utf8"))
def query_params():
"""
Returns a list of two-tuples byte strings, ready for encoding with urlencode.
We're aiming for a total length of a URL below 2083 characters, so this strategy
ensures that the total urlencoded query string is not longer than 1500 characters.
"""
return strategies.lists(
strategies.tuples(_query_param(), _query_param()),
min_size=0,
average_size=5,
).filter(lambda x: len(parse.urlencode(x)) < 1500)
def header_name():
"""
Strategy returning something that looks like a HTTP header field
https://en.wikipedia.org/wiki/List_of_HTTP_header_fields suggests they are between 4
and 20 characters long
"""
return strategies.text(
alphabet=string.ascii_letters + string.digits + "-",
min_size=1,
max_size=30,
)
def header_value():
"""
Strategy returning something that looks like a HTTP header value
"For example, the Apache 2.3 server by default limits the size of each field to 8190 bytes"
https://en.wikipedia.org/wiki/List_of_HTTP_header_fields
"""
return strategies.text(
alphabet=string.ascii_letters + string.digits + string.punctuation.replace(",", "") + " /t",
min_size=1,
average_size=40,
max_size=8190,
).filter(lambda s: len(s.encode("utf8")) < 8190)
def headers():
"""
Strategy returning a list of tuples, containing HTTP header fields and their values.
"[Apache 2.3] there can be at most 100 header fields in a single request."
https://en.wikipedia.org/wiki/List_of_HTTP_header_fields
"""
return strategies.lists(
strategies.tuples(header_name(), header_value()),
min_size=0,
average_size=10,
max_size=100,
)