def test_io(self):
        fp = BytesIO(b'foo')
        resp = HTTPResponse(fp, preload_content=False)

        assert not resp.closed
        assert resp.readable()
        assert not resp.writable()
        with pytest.raises(IOError):
            resp.fileno()

        resp.close()
        assert resp.closed

        # Try closing with a base Response
        try:
            hlr = get_response()
            resp2 = HTTPResponse(hlr.body, preload_content=False)
            assert not resp2.closed
            resp2.close()
            assert resp2.closed
        finally:
            hlr.close()

        # also try when only data is present.
        resp3 = HTTPResponse('foodata')
        with pytest.raises(IOError):
            resp3.fileno()
 def test_retries(self):
     fp = BytesIO(b'')
     resp = HTTPResponse(fp)
     assert resp.retries is None
     retry = Retry()
     resp = HTTPResponse(fp, retries=retry)
     assert resp.retries == retry
Example #3
0
    def test_error_message(self):
        retry = Retry(total=0)
        with pytest.raises(MaxRetryError) as e:
            retry = retry.increment(method='GET',
                                    error=ReadTimeoutError(
                                        None, "/", "read timed out"))
        assert 'Caused by redirect' not in str(e.value)
        assert str(e.value.reason) == 'None: read timed out'

        retry = Retry(total=1)
        with pytest.raises(MaxRetryError) as e:
            retry = retry.increment('POST', '/')
            retry = retry.increment('POST', '/')
        assert 'Caused by redirect' not in str(e.value)
        assert isinstance(e.value.reason, ResponseError)
        assert str(e.value.reason) == ResponseError.GENERIC_ERROR

        retry = Retry(total=1)
        response = HTTPResponse(status=500)
        with pytest.raises(MaxRetryError) as e:
            retry = retry.increment('POST', '/', response=response)
            retry = retry.increment('POST', '/', response=response)
        assert 'Caused by redirect' not in str(e.value)
        msg = ResponseError.SPECIFIC_ERROR.format(status_code=500)
        assert str(e.value.reason) == msg

        retry = Retry(connect=1)
        with pytest.raises(MaxRetryError) as e:
            retry = retry.increment(error=ConnectTimeoutError('conntimeout'))
            retry = retry.increment(error=ConnectTimeoutError('conntimeout'))
        assert 'Caused by redirect' not in str(e.value)
        assert str(e.value.reason) == 'conntimeout'
    def test_preload(self):
        fp = BytesIO(b'foo')

        r = HTTPResponse(fp, preload_content=True)

        assert fp.tell() == len(b'foo')
        assert r.data == b'foo'
    def test_close_midstream(self):
        # A mock fp object that wraps a list and allows closing.
        class MockFP(object):
            self.list = None

            def close(self):
                self.list = None

            def __iter__(self):
                return self

            def __next__(self):
                if not self.list:
                    raise StopIteration()
                return self.list.pop(0)

            next = __next__

        data = [b'fo', b'o']
        fp = MockFP()
        fp.list = data
        resp = HTTPResponse(fp, preload_content=False)
        stream = resp.stream()

        assert next(stream) == b'fo'
        resp.close()
        with pytest.raises(StopIteration):
            next(stream)
    def test_empty_stream(self):
        fp = BytesIO(b'')
        resp = HTTPResponse(fp, preload_content=False)
        stream = resp.stream(decode_content=False)

        with pytest.raises(StopIteration):
            next(stream)
    def test_decode_deflate(self):
        import zlib
        data = zlib.compress(b'foo')

        fp = BytesIO(data)
        r = HTTPResponse(fp, headers={'content-encoding': 'deflate'})

        assert r.data == b'foo'
    def test_reference_read(self):
        fp = BytesIO(b'foo')
        r = HTTPResponse(fp, preload_content=False)

        assert r.read(1) == b'f'
        assert r.read(2) == b'oo'
        assert r.read() == b''
        assert r.read() == b''
Example #9
0
 def test_status_counter(self):
     resp = HTTPResponse(status=400)
     retry = Retry(status=2)
     retry = retry.increment(response=resp)
     retry = retry.increment(response=resp)
     with pytest.raises(MaxRetryError) as e:
         retry.increment(response=resp)
     assert str(e.value.reason) == ResponseError.SPECIFIC_ERROR.format(
         status_code=400)
    def test_double_streaming(self):
        fp = [b'fo', b'o']
        resp = HTTPResponse(fp, preload_content=False)

        stream = list(resp.stream(decode_content=False))
        assert stream == fp

        stream = list(resp.stream(decode_content=False))
        assert stream == []
    def test_streaming(self):
        fp = [b'fo', b'o']
        resp = HTTPResponse(fp, preload_content=False)
        stream = resp.stream(decode_content=False)

        assert next(stream) == b'fo'
        assert next(stream) == b'o'
        with pytest.raises(StopIteration):
            next(stream)
    def test_io_bufferedreader(self):
        fp = BytesIO(b'foo')
        resp = HTTPResponse(fp, preload_content=False)
        br = BufferedReader(resp)

        assert br.read() == b'foo'

        br.close()
        assert resp.closed

        b = b'!tenbytes!'
        fp = BytesIO(b)
        resp = HTTPResponse(fp, preload_content=False)
        br = BufferedReader(resp, 5)

        # This is necessary to make sure the "no bytes left" part of `readinto`
        # gets tested.
        assert len(br.read(5)) == 5
        assert len(br.read(5)) == 5
        assert len(br.read(5)) == 0
Example #13
0
 def test_backoff_reset_after_redirect(self):
     retry = Retry(total=100, redirect=5, backoff_factor=0.2)
     retry = retry.increment(method='GET')
     retry = retry.increment(method='GET')
     assert retry.get_backoff_time() == 0.4
     redirect_response = HTTPResponse(status=302,
                                      headers={'location': 'test'})
     retry = retry.increment(method='GET', response=redirect_response)
     assert retry.get_backoff_time() == 0
     retry = retry.increment(method='GET')
     retry = retry.increment(method='GET')
     assert retry.get_backoff_time() == 0.4
    def test_deflate_streaming(self):
        import zlib
        data = zlib.compress(b'foo')

        fp = BytesIO(data)
        resp = HTTPResponse(fp, headers={'content-encoding': 'deflate'},
                            preload_content=False)
        stream = resp.stream()

        assert next(stream) == b'foo'
        with pytest.raises(StopIteration):
            next(stream)
    def test_extract(self):
        request = urllib.request.Request('http://google.com')
        cookiejar = http_cookiejar.CookieJar()
        response = HTTPResponse()

        cookies = [
            "sessionhash=abcabcabcabcab; path=/; HttpOnly",
            "lastvisit=1348253375; expires=Sat, 21-Sep-2050 18:49:35 GMT; path=/"
        ]
        for c in cookies:
            response.headers.add('set-cookie', c)
        cookiejar.extract_cookies(response, request)
        assert len(cookiejar) == len(cookies)
    def test_chunked_decoding_gzip(self):
        import zlib
        compress = zlib.compressobj(6, zlib.DEFLATED, 16 + zlib.MAX_WBITS)
        data = compress.compress(b'foo')
        data += compress.flush()

        fp = BytesIO(data)
        r = HTTPResponse(fp, headers={'content-encoding': 'gzip'},
                         preload_content=False)

        assert r.read(1) == b'f'
        assert r.read(2) == b'oo'
        assert r.read() == b''
        assert r.read() == b''
    def test_deflate2_streaming(self):
        import zlib
        compress = zlib.compressobj(6, zlib.DEFLATED, -zlib.MAX_WBITS)
        data = compress.compress(b'foo')
        data += compress.flush()

        fp = BytesIO(data)
        resp = HTTPResponse(fp, headers={'content-encoding': 'deflate'},
                            preload_content=False)
        stream = resp.stream()

        assert next(stream) == b'foo'
        with pytest.raises(StopIteration):
            next(stream)
    def test_chunked_decoding_deflate(self):
        import zlib
        data = zlib.compress(b'foo')

        fp = BytesIO(data)
        r = HTTPResponse(fp, headers={'content-encoding': 'deflate'},
                         preload_content=False)

        assert r.read(1) == b'f'
        # Buffer in case we need to switch to the raw stream
        assert r._decoder._data is None
        assert r.read(2) == b'oo'
        assert r.read() == b''
        assert r.read() == b''
    def test_chunked_decoding_deflate2(self):
        import zlib
        compress = zlib.compressobj(6, zlib.DEFLATED, -zlib.MAX_WBITS)
        data = compress.compress(b'foo')
        data += compress.flush()

        fp = BytesIO(data)
        r = HTTPResponse(fp, headers={'content-encoding': 'deflate'},
                         preload_content=False)

        assert r.read(1) == b'f'
        # Once we've decoded data, we just stream to the decoder; no buffering
        assert r._decoder._data is None
        assert r.read(2) == b'oo'
        assert r.read() == b''
        assert r.read() == b''
    def test_streaming_tell(self):
        fp = [b'fo', b'o']
        resp = HTTPResponse(fp, preload_content=False)
        stream = resp.stream(decode_content=False)

        position = 0

        position += len(next(stream))
        assert 2 == position
        assert 2 == resp.tell()

        position += len(next(stream))
        assert 3 == position
        assert 3 == resp.tell()

        with pytest.raises(StopIteration):
            next(stream)
    def test_gzipped_streaming_tell(self):
        import zlib
        compress = zlib.compressobj(6, zlib.DEFLATED, 16 + zlib.MAX_WBITS)
        uncompressed_data = b'foo'
        data = compress.compress(uncompressed_data)
        data += compress.flush()

        fp = BytesIO(data)
        resp = HTTPResponse(fp, headers={'content-encoding': 'gzip'},
                            preload_content=False)
        stream = resp.stream()

        # Read everything
        payload = next(stream)
        assert payload == uncompressed_data

        assert len(data) == resp.tell()

        with pytest.raises(StopIteration):
            next(stream)
Example #22
0
    def test_history(self):
        retry = Retry(total=10, method_whitelist=frozenset(['GET', 'POST']))
        assert retry.history == tuple()
        connection_error = ConnectTimeoutError('conntimeout')
        retry = retry.increment('GET', '/test1', None, connection_error)
        history = (RequestHistory('GET', '/test1', connection_error, None,
                                  None), )
        assert retry.history == history

        read_error = ReadTimeoutError(None, "/test2", "read timed out")
        retry = retry.increment('POST', '/test2', None, read_error)
        history = (RequestHistory('GET', '/test1', connection_error, None,
                                  None),
                   RequestHistory('POST', '/test2', read_error, None, None))
        assert retry.history == history

        response = HTTPResponse(status=500)
        retry = retry.increment('GET', '/test3', response, None)
        history = (RequestHistory('GET', '/test1', connection_error, None,
                                  None),
                   RequestHistory('POST', '/test2', read_error, None, None),
                   RequestHistory('GET', '/test3', None, 500, None))
        assert retry.history == history
    def test_mock_httpresponse_stream(self):
        # Mock out a HTTP Request that does enough to make it through urllib3's
        # read() and close() calls, and also exhausts and underlying file
        # object.
        class MockHTTPRequest(object):
            self.fp = None

            def read(self, amt):
                data = self.fp.read(amt)
                if not data:
                    self.fp = None

                return data

            def close(self):
                self.fp = None

            def __iter__(self):
                return self

            def __next__(self):
                if self.fp is None:
                    raise StopIteration()
                return self.read(1)

            next = __next__
        bio = BytesIO(b'foo')
        fp = MockHTTPRequest()
        fp.fp = bio
        resp = HTTPResponse(fp, preload_content=False)
        stream = resp.stream()

        assert next(stream) == b'f'
        assert next(stream) == b'o'
        assert next(stream) == b'o'
        with pytest.raises(StopIteration):
            next(stream)
 def test_getheader(self):
     headers = {'host': 'example.com'}
     r = HTTPResponse(headers=headers)
     assert r.getheader('host') == 'example.com'
    def test_deflate_streaming_tell_intermediate_point(self):
        # Ensure that ``tell()`` returns the correct number of bytes when
        # part-way through streaming compressed content.
        import zlib

        NUMBER_OF_READS = 10

        class MockCompressedDataReading(BytesIO):
            """
            A ByteIO-like reader returning ``payload`` in ``NUMBER_OF_READS``
            calls to ``read``.
            """

            def __init__(self, payload, payload_part_size):
                self.payloads = [
                    payload[i*payload_part_size:(i+1)*payload_part_size]
                    for i in range(NUMBER_OF_READS+1)]
                self.consumed = 0

                assert b"".join(self.payloads) == payload

            def read(self, _):
                # Amount is unused.
                if len(self.payloads) > 0:
                    payload = self.payloads.pop(0)
                    self.consumed += len(payload)
                    return payload
                return b""

            def __iter__(self):
                return self

            def __next__(self):
                if not self.payloads:
                    raise StopIteration()
                return self.read(None)

            next = __next__

        uncompressed_data = zlib.decompress(ZLIB_PAYLOAD)

        payload_part_size = len(ZLIB_PAYLOAD) // NUMBER_OF_READS
        fp = MockCompressedDataReading(ZLIB_PAYLOAD, payload_part_size)
        resp = HTTPResponse(fp, headers={'content-encoding': 'deflate'},
                            preload_content=False)
        parts = []
        stream = resp.stream(1)

        for part in stream:
            parts.append(part)
            assert resp.tell() == fp.consumed

        end_of_stream = resp.tell()

        with pytest.raises(StopIteration):
            next(stream)

        # Check that the payload is equal to the uncompressed data
        payload = b"".join(parts)
        assert uncompressed_data == payload

        # Check that the end of the stream is in the correct place
        assert len(ZLIB_PAYLOAD) == end_of_stream
 def test_get_case_insensitive_headers(self):
     headers = {'host': 'example.com'}
     r = HTTPResponse(headers=headers)
     assert r.headers.get('host') == 'example.com'
     assert r.headers.get('Host') == 'example.com'
 def test_cache_content(self):
     r = HTTPResponse('foo')
     assert r.data == 'foo'
     assert r._body == 'foo'
 def test_default(self):
     r = HTTPResponse()
     assert r.data == b''
 def test_decode_bad_data(self):
     fp = BytesIO(b'\x00' * 10)
     with pytest.raises(DecodeError):
         HTTPResponse(fp, headers={'content-encoding': 'deflate'})
 def test_none(self):
     r = HTTPResponse(None)
     assert r.data == b''