mitmproxy/test/test_console_contentview.py

332 lines
9.8 KiB
Python
Raw Normal View History

import os
from nose.plugins.skip import SkipTest
if os.name == "nt":
raise SkipTest("Skipped on Windows.")
import sys
import libmproxy.console.contentview as cv
from libmproxy import utils, flow, encoding
import tutils
try:
import pyamf
except ImportError:
pyamf = None
class TestContentView:
def test_trailer(self):
txt = []
cv.trailer(5, txt, 1000)
assert not txt
cv.trailer(cv.VIEW_CUTOFF + 10, txt, cv.VIEW_CUTOFF)
assert txt
def test_view_auto(self):
v = cv.ViewAuto()
f = v(
flow.ODictCaseless(),
"foo",
1000
)
assert f[0] == "Raw"
f = v(
flow.ODictCaseless(
[["content-type", "text/html"]],
),
"<html></html>",
1000
)
assert f[0] == "HTML"
f = v(
flow.ODictCaseless(
[["content-type", "text/flibble"]],
),
"foo",
1000
)
assert f[0] == "Raw"
f = v(
flow.ODictCaseless(
[["content-type", "text/flibble"]],
),
"<xml></xml>",
1000
)
assert f[0].startswith("XML")
def test_view_urlencoded(self):
d = utils.urlencode([("one", "two"), ("three", "four")])
2012-08-18 05:08:17 +00:00
v = cv.ViewURLEncoded()
assert v([], d, 100)
d = utils.urlencode([("adsfa", "")])
v = cv.ViewURLEncoded()
assert v([], d, 100)
def test_view_html(self):
2012-08-18 05:08:17 +00:00
v = cv.ViewHTML()
s = "<html><br><br></br><p>one</p></html>"
2012-08-18 05:08:17 +00:00
assert v([], s, 1000)
s = "gobbledygook"
2012-08-18 05:08:17 +00:00
assert not v([], s, 1000)
def test_view_html_outline(self):
v = cv.ViewHTMLOutline()
s = "<html><br><br></br><p>one</p></html>"
assert v([], s, 1000)
def test_view_json(self):
cv.VIEW_CUTOFF = 100
2012-08-18 05:08:17 +00:00
v = cv.ViewJSON()
assert v([], "{}", 1000)
assert not v([], "{", 1000)
assert v([], "[" + ",".join(["0"]*cv.VIEW_CUTOFF) + "]", 1000)
assert v([], "[1, 2, 3, 4, 5]", 5)
def test_view_xml(self):
2012-08-18 05:08:17 +00:00
v = cv.ViewXML()
assert v([], "<foo></foo>", 1000)
assert not v([], "<foo>", 1000)
s = """<?xml version="1.0" encoding="UTF-8"?>
<?xml-stylesheet title="XSL_formatting"?>
2012-04-07 10:15:31 +00:00
<rss
xmlns:media="http://search.yahoo.com/mrss/"
xmlns:atom="http://www.w3.org/2005/Atom"
version="2.0">
</rss>
"""
2012-08-18 05:08:17 +00:00
assert v([], s, 1000)
def test_view_raw(self):
2012-08-18 05:08:17 +00:00
v = cv.ViewRaw()
assert v([], "foo", 1000)
def test_view_javascript(self):
2012-08-18 05:08:17 +00:00
v = cv.ViewJavaScript()
assert v([], "[1, 2, 3]", 100)
assert v([], "[1, 2, 3", 100)
assert v([], "function(a){[1, 2, 3]}", 100)
def test_view_hex(self):
2012-08-18 05:08:17 +00:00
v = cv.ViewHex()
assert v([], "foo", 1000)
def test_view_image(self):
2012-08-18 05:08:17 +00:00
v = cv.ViewImage()
p = tutils.test_data.path("data/image.png")
2013-06-15 22:23:44 +00:00
assert v([], file(p,"rb").read(), sys.maxint)
p = tutils.test_data.path("data/image.gif")
2013-06-15 22:23:44 +00:00
assert v([], file(p,"rb").read(), sys.maxint)
p = tutils.test_data.path("data/image-err1.jpg")
2013-06-15 22:23:44 +00:00
assert v([], file(p,"rb").read(), sys.maxint)
p = tutils.test_data.path("data/image.ico")
2013-06-15 22:23:44 +00:00
assert v([], file(p,"rb").read(), sys.maxint)
2012-08-18 05:08:17 +00:00
assert not v([], "flibble", sys.maxint)
def test_view_multipart(self):
2012-08-18 05:08:17 +00:00
view = cv.ViewMultipart()
v = """
--AaB03x
Content-Disposition: form-data; name="submit-name"
Larry
--AaB03x
""".strip()
h = flow.ODictCaseless(
[("Content-Type", "multipart/form-data; boundary=AaB03x")]
)
2012-08-18 05:08:17 +00:00
assert view(h, v, 1000)
h = flow.ODictCaseless()
2012-08-18 05:08:17 +00:00
assert not view(h, v, 1000)
h = flow.ODictCaseless(
[("Content-Type", "multipart/form-data")]
)
2012-08-18 05:08:17 +00:00
assert not view(h, v, 1000)
h = flow.ODictCaseless(
[("Content-Type", "unparseable")]
)
2012-08-18 05:08:17 +00:00
assert not view(h, v, 1000)
def test_get_content_view(self):
r = cv.get_content_view(
2012-08-18 05:08:17 +00:00
cv.get("Raw"),
[["content-type", "application/json"]],
"[1, 2, 3]",
1000,
lambda x: None
)
assert "Raw" in r[0]
r = cv.get_content_view(
2012-08-18 05:08:17 +00:00
cv.get("Auto"),
[["content-type", "application/json"]],
"[1, 2, 3]",
1000,
lambda x: None
)
assert r[0] == "JSON"
r = cv.get_content_view(
2012-08-18 05:08:17 +00:00
cv.get("Auto"),
[["content-type", "application/json"]],
"[1, 2",
1000,
lambda x: None
)
assert "Raw" in r[0]
r = cv.get_content_view(
cv.get("AMF"),
[],
"[1, 2",
1000,
lambda x: None
)
assert "Raw" in r[0]
r = cv.get_content_view(
2012-08-18 05:08:17 +00:00
cv.get("Auto"),
[
["content-type", "application/json"],
["content-encoding", "gzip"]
],
encoding.encode('gzip', "[1, 2, 3]"),
1000,
lambda x: None
)
assert "decoded gzip" in r[0]
assert "JSON" in r[0]
r = cv.get_content_view(
2012-08-18 05:08:17 +00:00
cv.get("XML"),
[
["content-type", "application/json"],
["content-encoding", "gzip"]
],
encoding.encode('gzip', "[1, 2, 3]"),
1000,
lambda x: None
)
assert "decoded gzip" in r[0]
2012-04-07 10:15:31 +00:00
assert "Raw" in r[0]
if pyamf:
def test_view_amf_request():
v = cv.ViewAMF()
p = tutils.test_data.path("data/amf01")
2013-06-15 22:23:44 +00:00
assert v([], file(p,"rb").read(), sys.maxint)
p = tutils.test_data.path("data/amf02")
2013-06-15 22:23:44 +00:00
assert v([], file(p,"rb").read(), sys.maxint)
def test_view_amf_response():
v = cv.ViewAMF()
p = tutils.test_data.path("data/amf03")
2013-06-15 22:23:44 +00:00
assert v([], file(p,"rb").read(), sys.maxint)
if cv.ViewProtobuf.is_available():
def test_view_protobuf_request():
v = cv.ViewProtobuf()
p = tutils.test_data.path("data/protobuf01")
2013-06-15 22:23:44 +00:00
content_type, output = v([], file(p,"rb").read(), sys.maxint)
assert content_type == "Protobuf"
assert output[0].text == '1: "3bbc333c-e61c-433b-819a-0b9a8cc103b8"'
def test_get_by_shortcut():
assert cv.get_by_shortcut("h")
2013-12-24 01:28:20 +00:00
def test_search_highlights():
# Default text in requests is content. We will search for nt once, and
# expect the first bit to be highlighted. We will do it again and expect the
# second to be.
f = tutils.tflowview()
f.search("nt")
text_object = tutils.get_body_line(f.last_displayed_body, 0)
assert text_object.get_text() == ('content', [(None, 2), (f.highlight_color, 2)])
f.search("nt")
text_object = tutils.get_body_line(f.last_displayed_body, 1)
assert text_object.get_text() == ('content', [(None, 5), (f.highlight_color, 2)])
2013-12-26 04:04:18 +00:00
def test_search_returns_useful_messages():
f = tutils.tflowview()
# original string is content. this string should not be in there.
response = f.search("oranges and other fruit.")
assert response == "no matches for 'oranges and other fruit.'"
def test_search_highlights_clears_prev():
f = tutils.tflowview(request_contents="this is string\nstring is string")
f.search("string")
text_object = tutils.get_body_line(f.last_displayed_body, 0)
assert text_object.get_text() == ('this is string', [(None, 8), (f.highlight_color, 6)])
# search again, it should not be highlighted again.
f.search("string")
text_object = tutils.get_body_line(f.last_displayed_body, 0)
assert text_object.get_text() != ('this is string', [(None, 8), (f.highlight_color, 6)])
def test_search_highlights_multi_line():
f = tutils.tflowview(request_contents="this is string\nstring is string")
# should highlight the first line.
f.search("string")
text_object = tutils.get_body_line(f.last_displayed_body, 0)
assert text_object.get_text() == ('this is string', [(None, 8), (f.highlight_color, 6)])
# should highlight second line, first appearance of string.
f.search("string")
text_object = tutils.get_body_line(f.last_displayed_body, 1)
2013-12-25 08:08:20 +00:00
print text_object.get_text()
assert text_object.get_text() == ('string is string', [(None, 0), (f.highlight_color, 6)])
# should highlight third line, second appearance of string.
f.search("string")
text_object = tutils.get_body_line(f.last_displayed_body, 1)
assert text_object.get_text() == ('string is string', [(None, 10), (f.highlight_color, 6)])
2013-12-26 04:04:18 +00:00
def test_search_loops():
f = tutils.tflowview(request_contents="this is string\nstring is string")
# get to the end.
f.search("string")
f.search("string")
f.search("string")
# should highlight the first line.
message = f.search("string")
text_object = tutils.get_body_line(f.last_displayed_body, 0)
assert text_object.get_text() == ('this is string', [(None, 8), (f.highlight_color, 6)])
assert message == "search hit BOTTOM, continuing at TOP"
def test_search_focuses():
f = tutils.tflowview(request_contents="this is string\nstring is string")
# should highlight the first line.
f.search("string")
# should be focusing on the 2nd text line.
f.search("string")
text_object = tutils.get_body_line(f.last_displayed_body, 1)
assert f.last_displayed_body.focus == text_object
2013-12-24 01:28:20 +00:00