Python pycurl 模块,EFFECTIVE_URL 实例源码

我们从Python开源项目中,提取了以下23个代码示例,用于说明如何使用pycurl.EFFECTIVE_URL

项目:defcon-workshop    作者:devsecops    | 项目源码 | 文件源码
def response_from_conn_object(self, conn, header, body):
        # followlocation
        if conn.getinfo(pycurl.EFFECTIVE_URL) != self.completeUrl:
        self.setFinalUrl(conn.getinfo(pycurl.EFFECTIVE_URL))
        #pycurl reponse headers includes original => remove
        header = header[header.find("\r\n\r\n")+1:]

        self.totaltime = conn.getinfo(pycurl.TOTAL_TIME)

        rp = Response()
        rp.parseResponse(header)
        rp.addContent(body)

        if self.schema=="https" and self.__proxy:
        self.response=Response()
        self.response.parseResponse(rp.getContent())
        else:
        self.response=rp

        return rp
项目:wfuzz    作者:gwen001    | 项目源码 | 文件源码
def response_from_conn_object(self, conn, header, body):
        # followlocation
        if conn.getinfo(pycurl.EFFECTIVE_URL) != self.completeUrl:
        self.setFinalUrl(conn.getinfo(pycurl.EFFECTIVE_URL))
        #pycurl reponse headers includes original => remove
        header = header[header.find("\r\n\r\n")+1:]

        self.totaltime = conn.getinfo(pycurl.TOTAL_TIME)

        rp = Response()
        rp.parseResponse(header)
        rp.addContent(body)

        if self.schema=="https" and self.__proxy:
        self.response=Response()
        self.response.parseResponse(rp.getContent())
        else:
        self.response=rp

        return rp
项目:annotated-py-tornado    作者:hhstore    | 项目源码 | 文件源码
def fetch(self, request, **kwargs):
        """Executes an HTTPRequest, returning an HTTPResponse.

        If an error occurs during the fetch, we raise an HTTPError.
        """
        if not isinstance(request, HTTPRequest):
           request = HTTPRequest(url=request, **kwargs)
        buffer = cStringIO.StringIO()
        headers = httputil.HTTPHeaders()
        try:
            _curl_setup_request(self._curl, request, buffer, headers)
            self._curl.perform()
            code = self._curl.getinfo(pycurl.HTTP_CODE)
            effective_url = self._curl.getinfo(pycurl.EFFECTIVE_URL)
            buffer.seek(0)
            response = HTTPResponse(
                request=request, code=code, headers=headers,
                buffer=buffer, effective_url=effective_url)
            if code < 200 or code >= 300:
                raise HTTPError(code, response=response)
            return response
        except pycurl.error, e:
            buffer.close()
            raise CurlError(*e)
项目:noc-orchestrator    作者:DirceuSilvaLabs    | 项目源码 | 文件源码
def _finish(self, curl, curl_error=None, curl_message=None):
        info = curl.info
        curl.info = None
        self._multi.remove_handle(curl)
        self._free_list.append(curl)
        buffer = info["buffer"]
        if curl_error:
            error = CurlError(curl_error, curl_message)
            code = error.code
            effective_url = None
            buffer.close()
            buffer = None
        else:
            error = None
            code = curl.getinfo(pycurl.HTTP_CODE)
            effective_url = curl.getinfo(pycurl.EFFECTIVE_URL)
            buffer.seek(0)
        # the various curl timings are documented at
        # http://curl.haxx.se/libcurl/c/curl_easy_getinfo.html
        time_info = dict(
            queue=info["curl_start_time"] - info["request"].start_time,
            namelookup=curl.getinfo(pycurl.NAMELOOKUP_TIME),
            connect=curl.getinfo(pycurl.CONNECT_TIME),
            pretransfer=curl.getinfo(pycurl.PRETRANSFER_TIME),
            starttransfer=curl.getinfo(pycurl.STARTTRANSFER_TIME),
            total=curl.getinfo(pycurl.TOTAL_TIME),
            redirect=curl.getinfo(pycurl.REDIRECT_TIME),
        )
        try:
            info["callback"](HTTPResponse(
                request=info["request"], code=code, headers=info["headers"],
                buffer=buffer, effective_url=effective_url, error=error,
                reason=info['headers'].get("X-Http-Reason", None),
                request_time=time.time() - info["curl_start_time"],
                time_info=time_info))
        except Exception:
            self.handle_callback_exception(info["callback"])
项目:noc-orchestrator    作者:DirceuSilvaLabs    | 项目源码 | 文件源码
def _finish(self, curl, curl_error=None, curl_message=None):
        info = curl.info
        curl.info = None
        self._multi.remove_handle(curl)
        self._free_list.append(curl)
        buffer = info["buffer"]
        if curl_error:
            error = CurlError(curl_error, curl_message)
            code = error.code
            effective_url = None
            buffer.close()
            buffer = None
        else:
            error = None
            code = curl.getinfo(pycurl.HTTP_CODE)
            effective_url = curl.getinfo(pycurl.EFFECTIVE_URL)
            buffer.seek(0)
        # the various curl timings are documented at
        # http://curl.haxx.se/libcurl/c/curl_easy_getinfo.html
        time_info = dict(
            queue=info["curl_start_time"] - info["request"].start_time,
            namelookup=curl.getinfo(pycurl.NAMELOOKUP_TIME),
            connect=curl.getinfo(pycurl.CONNECT_TIME),
            pretransfer=curl.getinfo(pycurl.PRETRANSFER_TIME),
            starttransfer=curl.getinfo(pycurl.STARTTRANSFER_TIME),
            total=curl.getinfo(pycurl.TOTAL_TIME),
            redirect=curl.getinfo(pycurl.REDIRECT_TIME),
        )
        try:
            info["callback"](HTTPResponse(
                request=info["request"], code=code, headers=info["headers"],
                buffer=buffer, effective_url=effective_url, error=error,
                reason=info['headers'].get("X-Http-Reason", None),
                request_time=time.time() - info["curl_start_time"],
                time_info=time_info))
        except Exception:
            self.handle_callback_exception(info["callback"])
项目:noc-orchestrator    作者:DirceuSilvaLabs    | 项目源码 | 文件源码
def _finish(self, curl, curl_error=None, curl_message=None):
        info = curl.info
        curl.info = None
        self._multi.remove_handle(curl)
        self._free_list.append(curl)
        buffer = info["buffer"]
        if curl_error:
            error = CurlError(curl_error, curl_message)
            code = error.code
            effective_url = None
            buffer.close()
            buffer = None
        else:
            error = None
            code = curl.getinfo(pycurl.HTTP_CODE)
            effective_url = curl.getinfo(pycurl.EFFECTIVE_URL)
            buffer.seek(0)
        # the various curl timings are documented at
        # http://curl.haxx.se/libcurl/c/curl_easy_getinfo.html
        time_info = dict(
            queue=info["curl_start_time"] - info["request"].start_time,
            namelookup=curl.getinfo(pycurl.NAMELOOKUP_TIME),
            connect=curl.getinfo(pycurl.CONNECT_TIME),
            pretransfer=curl.getinfo(pycurl.PRETRANSFER_TIME),
            starttransfer=curl.getinfo(pycurl.STARTTRANSFER_TIME),
            total=curl.getinfo(pycurl.TOTAL_TIME),
            redirect=curl.getinfo(pycurl.REDIRECT_TIME),
        )
        try:
            info["callback"](HTTPResponse(
                request=info["request"], code=code, headers=info["headers"],
                buffer=buffer, effective_url=effective_url, error=error,
                reason=info['headers'].get("X-Http-Reason", None),
                request_time=time.time() - info["curl_start_time"],
                time_info=time_info))
        except Exception:
            self.handle_callback_exception(info["callback"])
项目:download-manager    作者:thispc    | 项目源码 | 文件源码
def load(self, url, get={}, post={}, referer=True, cookies=True, just_header=False, multipart=False, decode=False):
        """ load and returns a given page """

        self.setRequestContext(url, get, post, referer, cookies, multipart)

        self.header = ""

        self.c.setopt(pycurl.HTTPHEADER, self.headers)

        if just_header:
            self.c.setopt(pycurl.FOLLOWLOCATION, 0)
            self.c.setopt(pycurl.NOBODY, 1)
            self.c.perform()
            rep = self.header

            self.c.setopt(pycurl.FOLLOWLOCATION, 1)
            self.c.setopt(pycurl.NOBODY, 0)

        else:
            self.c.perform()
            rep = self.getResponse()

        self.c.setopt(pycurl.POSTFIELDS, "")
        self.lastEffectiveURL = self.c.getinfo(pycurl.EFFECTIVE_URL)
        self.code = self.verifyHeader()

        self.addCookies()

        if decode:
            rep = self.decodeResponse(rep)

        return rep
项目:My-Web-Server-Framework-With-Python2.7    作者:syjsu    | 项目源码 | 文件源码
def _finish(self, curl, curl_error=None, curl_message=None):
        info = curl.info
        curl.info = None
        self._multi.remove_handle(curl)
        self._free_list.append(curl)
        buffer = info["buffer"]
        if curl_error:
            error = CurlError(curl_error, curl_message)
            code = error.code
            effective_url = None
            buffer.close()
            buffer = None
        else:
            error = None
            code = curl.getinfo(pycurl.HTTP_CODE)
            effective_url = curl.getinfo(pycurl.EFFECTIVE_URL)
            buffer.seek(0)
        # the various curl timings are documented at
        # http://curl.haxx.se/libcurl/c/curl_easy_getinfo.html
        time_info = dict(
            queue=info["curl_start_time"] - info["request"].start_time,
            namelookup=curl.getinfo(pycurl.NAMELOOKUP_TIME),
            connect=curl.getinfo(pycurl.CONNECT_TIME),
            pretransfer=curl.getinfo(pycurl.PRETRANSFER_TIME),
            starttransfer=curl.getinfo(pycurl.STARTTRANSFER_TIME),
            total=curl.getinfo(pycurl.TOTAL_TIME),
            redirect=curl.getinfo(pycurl.REDIRECT_TIME),
        )
        try:
            info["callback"](HTTPResponse(
                request=info["request"], code=code, headers=info["headers"],
                buffer=buffer, effective_url=effective_url, error=error,
                reason=info['headers'].get("X-Http-Reason", None),
                request_time=time.time() - info["curl_start_time"],
                time_info=time_info))
        except Exception:
            self.handle_callback_exception(info["callback"])
项目:time2go    作者:twitchyliquid64    | 项目源码 | 文件源码
def _finish(self, curl, curl_error=None, curl_message=None):
        info = curl.info
        curl.info = None
        self._multi.remove_handle(curl)
        self._free_list.append(curl)
        buffer = info["buffer"]
        if curl_error:
            error = CurlError(curl_error, curl_message)
            code = error.code
            effective_url = None
            buffer.close()
            buffer = None
        else:
            error = None
            code = curl.getinfo(pycurl.HTTP_CODE)
            effective_url = curl.getinfo(pycurl.EFFECTIVE_URL)
            buffer.seek(0)
        # the various curl timings are documented at
        # http://curl.haxx.se/libcurl/c/curl_easy_getinfo.html
        time_info = dict(
            queue=info["curl_start_time"] - info["request"].start_time,
            namelookup=curl.getinfo(pycurl.NAMELOOKUP_TIME),
            connect=curl.getinfo(pycurl.CONNECT_TIME),
            pretransfer=curl.getinfo(pycurl.PRETRANSFER_TIME),
            starttransfer=curl.getinfo(pycurl.STARTTRANSFER_TIME),
            total=curl.getinfo(pycurl.TOTAL_TIME),
            redirect=curl.getinfo(pycurl.REDIRECT_TIME),
            )
        try:
            info["callback"](HTTPResponse(
                request=info["request"], code=code, headers=info["headers"],
                buffer=buffer, effective_url=effective_url, error=error,
                request_time=time.time() - info["curl_start_time"],
                time_info=time_info))
        except Exception:
            self.handle_callback_exception(info["callback"])
项目:annotated-py-tornado    作者:hhstore    | 项目源码 | 文件源码
def _finish(self, curl, curl_error=None, curl_message=None):
        info = curl.info
        curl.info = None
        self._multi.remove_handle(curl)
        self._free_list.append(curl)
        buffer = info["buffer"]
        if curl_error:
            error = CurlError(curl_error, curl_message)
            code = error.code
            body = None
            effective_url = None
            buffer.close()
            buffer = None
        else:
            error = None
            code = curl.getinfo(pycurl.HTTP_CODE)
            effective_url = curl.getinfo(pycurl.EFFECTIVE_URL)
            buffer.seek(0)
        try:
            info["callback"](HTTPResponse(
                request=info["request"], code=code, headers=info["headers"],
                buffer=buffer, effective_url=effective_url, error=error,
                request_time=time.time() - info["start_time"]))
        except (KeyboardInterrupt, SystemExit):
            raise
        except:
            logging.error("Exception in callback %r", info["callback"],
                          exc_info=True)
项目:annotated-py-tornado    作者:hhstore    | 项目源码 | 文件源码
def _finish(self, curl, curl_error=None, curl_message=None):
        info = curl.info
        curl.info = None
        self._multi.remove_handle(curl)
        self._free_list.append(curl)
        buffer = info["buffer"]
        if curl_error:
            error = CurlError(curl_error, curl_message)
            code = error.code
            effective_url = None
            buffer.close()
            buffer = None
        else:
            error = None
            code = curl.getinfo(pycurl.HTTP_CODE)
            effective_url = curl.getinfo(pycurl.EFFECTIVE_URL)
            buffer.seek(0)
        try:
            info["callback"](HTTPResponse(
                request=info["request"], code=code, headers=info["headers"],
                buffer=buffer, effective_url=effective_url, error=error,
                request_time=time.time() - info["start_time"]))
        except (KeyboardInterrupt, SystemExit):
            raise
        except:
            logging.error("Exception in callback %r", info["callback"],
                          exc_info=True)
项目:annotated-py-tornado    作者:hhstore    | 项目源码 | 文件源码
def _finish(self, curl, curl_error=None, curl_message=None):
        info = curl.info
        curl.info = None
        self._multi.remove_handle(curl)
        self._free_list.append(curl)
        buffer = info["buffer"]
        if curl_error:
            error = CurlError(curl_error, curl_message)
            code = error.code
            effective_url = None
            buffer.close()
            buffer = None
        else:
            error = None
            code = curl.getinfo(pycurl.HTTP_CODE)
            effective_url = curl.getinfo(pycurl.EFFECTIVE_URL)
            buffer.seek(0)
        # the various curl timings are documented at
        # http://curl.haxx.se/libcurl/c/curl_easy_getinfo.html
        time_info = dict(
            queue=info["curl_start_time"] - info["request"].start_time,
            namelookup=curl.getinfo(pycurl.NAMELOOKUP_TIME),
            connect=curl.getinfo(pycurl.CONNECT_TIME),
            pretransfer=curl.getinfo(pycurl.PRETRANSFER_TIME),
            starttransfer=curl.getinfo(pycurl.STARTTRANSFER_TIME),
            total=curl.getinfo(pycurl.TOTAL_TIME),
            redirect=curl.getinfo(pycurl.REDIRECT_TIME),
        )
        try:
            info["callback"](HTTPResponse(
                request=info["request"], code=code, headers=info["headers"],
                buffer=buffer, effective_url=effective_url, error=error,
                request_time=time.time() - info["curl_start_time"],
                time_info=time_info))
        except Exception:
            self.handle_callback_exception(info["callback"])
项目:annotated-py-tornado    作者:hhstore    | 项目源码 | 文件源码
def _finish(self, curl, curl_error=None, curl_message=None):
        info = curl.info
        curl.info = None
        self._multi.remove_handle(curl)
        self._free_list.append(curl)
        buffer = info["buffer"]
        if curl_error:
            error = CurlError(curl_error, curl_message)
            code = error.code
            effective_url = None
            buffer.close()
            buffer = None
        else:
            error = None
            code = curl.getinfo(pycurl.HTTP_CODE)
            effective_url = curl.getinfo(pycurl.EFFECTIVE_URL)
            buffer.seek(0)
        # the various curl timings are documented at
        # http://curl.haxx.se/libcurl/c/curl_easy_getinfo.html
        time_info = dict(
            queue=info["curl_start_time"] - info["request"].start_time,
            namelookup=curl.getinfo(pycurl.NAMELOOKUP_TIME),
            connect=curl.getinfo(pycurl.CONNECT_TIME),
            pretransfer=curl.getinfo(pycurl.PRETRANSFER_TIME),
            starttransfer=curl.getinfo(pycurl.STARTTRANSFER_TIME),
            total=curl.getinfo(pycurl.TOTAL_TIME),
            redirect=curl.getinfo(pycurl.REDIRECT_TIME),
        )
        try:
            info["callback"](HTTPResponse(
                request=info["request"], code=code, headers=info["headers"],
                buffer=buffer, effective_url=effective_url, error=error,
                reason=info['headers'].get("X-Http-Reason", None),
                request_time=time.time() - info["curl_start_time"],
                time_info=time_info))
        except Exception:
            self.handle_callback_exception(info["callback"])
项目:deprecated_thedap    作者:unitedvote    | 项目源码 | 文件源码
def _finish(self, curl, curl_error=None, curl_message=None):
        info = curl.info
        curl.info = None
        self._multi.remove_handle(curl)
        self._free_list.append(curl)
        buffer = info["buffer"]
        if curl_error:
            error = CurlError(curl_error, curl_message)
            code = error.code
            effective_url = None
            buffer.close()
            buffer = None
        else:
            error = None
            code = curl.getinfo(pycurl.HTTP_CODE)
            effective_url = curl.getinfo(pycurl.EFFECTIVE_URL)
            buffer.seek(0)
        # the various curl timings are documented at
        # http://curl.haxx.se/libcurl/c/curl_easy_getinfo.html
        time_info = dict(
            queue=info["curl_start_time"] - info["request"].start_time,
            namelookup=curl.getinfo(pycurl.NAMELOOKUP_TIME),
            connect=curl.getinfo(pycurl.CONNECT_TIME),
            pretransfer=curl.getinfo(pycurl.PRETRANSFER_TIME),
            starttransfer=curl.getinfo(pycurl.STARTTRANSFER_TIME),
            total=curl.getinfo(pycurl.TOTAL_TIME),
            redirect=curl.getinfo(pycurl.REDIRECT_TIME),
            )
        try:
            info["callback"](HTTPResponse(
                request=info["request"], code=code, headers=info["headers"],
                buffer=buffer, effective_url=effective_url, error=error,
                request_time=time.time() - info["curl_start_time"],
                time_info=time_info))
        except Exception:
            self.handle_callback_exception(info["callback"])
项目:get_started_with_respeaker    作者:respeaker    | 项目源码 | 文件源码
def _finish(self, curl, curl_error=None, curl_message=None):
        info = curl.info
        curl.info = None
        self._multi.remove_handle(curl)
        self._free_list.append(curl)
        buffer = info["buffer"]
        if curl_error:
            error = CurlError(curl_error, curl_message)
            code = error.code
            effective_url = None
            buffer.close()
            buffer = None
        else:
            error = None
            code = curl.getinfo(pycurl.HTTP_CODE)
            effective_url = curl.getinfo(pycurl.EFFECTIVE_URL)
            buffer.seek(0)
        # the various curl timings are documented at
        # http://curl.haxx.se/libcurl/c/curl_easy_getinfo.html
        time_info = dict(
            queue=info["curl_start_time"] - info["request"].start_time,
            namelookup=curl.getinfo(pycurl.NAMELOOKUP_TIME),
            connect=curl.getinfo(pycurl.CONNECT_TIME),
            pretransfer=curl.getinfo(pycurl.PRETRANSFER_TIME),
            starttransfer=curl.getinfo(pycurl.STARTTRANSFER_TIME),
            total=curl.getinfo(pycurl.TOTAL_TIME),
            redirect=curl.getinfo(pycurl.REDIRECT_TIME),
        )
        try:
            info["callback"](HTTPResponse(
                request=info["request"], code=code, headers=info["headers"],
                buffer=buffer, effective_url=effective_url, error=error,
                request_time=time.time() - info["curl_start_time"],
                time_info=time_info))
        except Exception:
            self.handle_callback_exception(info["callback"])
项目:teleport    作者:eomsoft    | 项目源码 | 文件源码
def _finish(self, curl, curl_error=None, curl_message=None):
        info = curl.info
        curl.info = None
        self._multi.remove_handle(curl)
        self._free_list.append(curl)
        buffer = info["buffer"]
        if curl_error:
            error = CurlError(curl_error, curl_message)
            code = error.code
            effective_url = None
            buffer.close()
            buffer = None
        else:
            error = None
            code = curl.getinfo(pycurl.HTTP_CODE)
            effective_url = curl.getinfo(pycurl.EFFECTIVE_URL)
            buffer.seek(0)
        # the various curl timings are documented at
        # http://curl.haxx.se/libcurl/c/curl_easy_getinfo.html
        time_info = dict(
            queue=info["curl_start_time"] - info["request"].start_time,
            namelookup=curl.getinfo(pycurl.NAMELOOKUP_TIME),
            connect=curl.getinfo(pycurl.CONNECT_TIME),
            pretransfer=curl.getinfo(pycurl.PRETRANSFER_TIME),
            starttransfer=curl.getinfo(pycurl.STARTTRANSFER_TIME),
            total=curl.getinfo(pycurl.TOTAL_TIME),
            redirect=curl.getinfo(pycurl.REDIRECT_TIME),
        )
        try:
            info["callback"](HTTPResponse(
                request=info["request"], code=code, headers=info["headers"],
                buffer=buffer, effective_url=effective_url, error=error,
                reason=info['headers'].get("X-Http-Reason", None),
                request_time=time.time() - info["curl_start_time"],
                time_info=time_info))
        except Exception:
            self.handle_callback_exception(info["callback"])
项目:projects-2017-2    作者:ncss    | 项目源码 | 文件源码
def _finish(self, curl, curl_error=None, curl_message=None):
        info = curl.info
        curl.info = None
        self._multi.remove_handle(curl)
        self._free_list.append(curl)
        buffer = info["buffer"]
        if curl_error:
            error = CurlError(curl_error, curl_message)
            code = error.code
            effective_url = None
            buffer.close()
            buffer = None
        else:
            error = None
            code = curl.getinfo(pycurl.HTTP_CODE)
            effective_url = curl.getinfo(pycurl.EFFECTIVE_URL)
            buffer.seek(0)
        # the various curl timings are documented at
        # http://curl.haxx.se/libcurl/c/curl_easy_getinfo.html
        time_info = dict(
            queue=info["curl_start_time"] - info["request"].start_time,
            namelookup=curl.getinfo(pycurl.NAMELOOKUP_TIME),
            connect=curl.getinfo(pycurl.CONNECT_TIME),
            pretransfer=curl.getinfo(pycurl.PRETRANSFER_TIME),
            starttransfer=curl.getinfo(pycurl.STARTTRANSFER_TIME),
            total=curl.getinfo(pycurl.TOTAL_TIME),
            redirect=curl.getinfo(pycurl.REDIRECT_TIME),
        )
        try:
            info["callback"](HTTPResponse(
                request=info["request"], code=code, headers=info["headers"],
                buffer=buffer, effective_url=effective_url, error=error,
                reason=info['headers'].get("X-Http-Reason", None),
                request_time=time.time() - info["curl_start_time"],
                time_info=time_info))
        except Exception:
            self.handle_callback_exception(info["callback"])
项目:stash-scanner    作者:senuido    | 项目源码 | 文件源码
def info(c):
    "Return a dictionary with all info on the last response."
    m = {}
    m['effective-url'] = c.getinfo(pycurl.EFFECTIVE_URL)
    m['http-code'] = c.getinfo(pycurl.HTTP_CODE)
    m['total-time'] = c.getinfo(pycurl.TOTAL_TIME)
    m['namelookup-time'] = c.getinfo(pycurl.NAMELOOKUP_TIME)
    m['connect-time'] = c.getinfo(pycurl.CONNECT_TIME)
    m['pretransfer-time'] = c.getinfo(pycurl.PRETRANSFER_TIME)
    m['redirect-time'] = c.getinfo(pycurl.REDIRECT_TIME)
    m['redirect-count'] = c.getinfo(pycurl.REDIRECT_COUNT)
    # m['size-upload'] = c.getinfo(pycurl.SIZE_UPLOAD)
    m['size-download'] = c.getinfo(pycurl.SIZE_DOWNLOAD)
    # m['speed-upload'] = c.getinfo(pycurl.SPEED_UPLOAD)
    m['header-size'] = c.getinfo(pycurl.HEADER_SIZE)
    m['request-size'] = c.getinfo(pycurl.REQUEST_SIZE)
    m['content-length-download'] = c.getinfo(pycurl.CONTENT_LENGTH_DOWNLOAD)
    m['content-length-upload'] = c.getinfo(pycurl.CONTENT_LENGTH_UPLOAD)
    m['content-type'] = c.getinfo(pycurl.CONTENT_TYPE)
    m['response-code'] = c.getinfo(pycurl.RESPONSE_CODE)
    m['speed-download'] = c.getinfo(pycurl.SPEED_DOWNLOAD)
    # m['ssl-verifyresult'] = c.getinfo(pycurl.SSL_VERIFYRESULT)
    m['filetime'] = c.getinfo(pycurl.INFO_FILETIME)
    m['starttransfer-time'] = c.getinfo(pycurl.STARTTRANSFER_TIME)
    m['redirect-time'] = c.getinfo(pycurl.REDIRECT_TIME)
    m['redirect-count'] = c.getinfo(pycurl.REDIRECT_COUNT)
    m['http-connectcode'] = c.getinfo(pycurl.HTTP_CONNECTCODE)
    # m['httpauth-avail'] = c.getinfo(pycurl.HTTPAUTH_AVAIL)
    # m['proxyauth-avail'] = c.getinfo(pycurl.PROXYAUTH_AVAIL)
    # m['os-errno'] = c.getinfo(pycurl.OS_ERRNO)
    m['num-connects'] = c.getinfo(pycurl.NUM_CONNECTS)
    # m['ssl-engines'] = c.getinfo(pycurl.SSL_ENGINES)
    # m['cookielist'] = c.getinfo(pycurl.INFO_COOKIELIST)
    # m['lastsocket'] = c.getinfo(pycurl.LASTSOCKET)
    # m['ftp-entry-path'] = c.getinfo(pycurl.FTP_ENTRY_PATH)
    return m
项目:PyQYT    作者:collinsctk    | 项目源码 | 文件源码
def _finish(self, curl, curl_error=None, curl_message=None):
        info = curl.info
        curl.info = None
        self._multi.remove_handle(curl)
        self._free_list.append(curl)
        buffer = info["buffer"]
        if curl_error:
            error = CurlError(curl_error, curl_message)
            code = error.code
            effective_url = None
            buffer.close()
            buffer = None
        else:
            error = None
            code = curl.getinfo(pycurl.HTTP_CODE)
            effective_url = curl.getinfo(pycurl.EFFECTIVE_URL)
            buffer.seek(0)
        # the various curl timings are documented at
        # http://curl.haxx.se/libcurl/c/curl_easy_getinfo.html
        time_info = dict(
            queue=info["curl_start_time"] - info["request"].start_time,
            namelookup=curl.getinfo(pycurl.NAMELOOKUP_TIME),
            connect=curl.getinfo(pycurl.CONNECT_TIME),
            pretransfer=curl.getinfo(pycurl.PRETRANSFER_TIME),
            starttransfer=curl.getinfo(pycurl.STARTTRANSFER_TIME),
            total=curl.getinfo(pycurl.TOTAL_TIME),
            redirect=curl.getinfo(pycurl.REDIRECT_TIME),
        )
        try:
            info["callback"](HTTPResponse(
                request=info["request"], code=code, headers=info["headers"],
                buffer=buffer, effective_url=effective_url, error=error,
                reason=info['headers'].get("X-Http-Reason", None),
                request_time=time.time() - info["curl_start_time"],
                time_info=time_info))
        except Exception:
            self.handle_callback_exception(info["callback"])
项目:alfred-10000ft-scripts    作者:jceelen    | 项目源码 | 文件源码
def info(self):
        "Return a dictionary with all info on the last response."
        m = {}
        m['effective-url'] = self.handle.getinfo(pycurl.EFFECTIVE_URL)
        m['http-code'] = self.handle.getinfo(pycurl.HTTP_CODE)
        m['total-time'] = self.handle.getinfo(pycurl.TOTAL_TIME)
        m['namelookup-time'] = self.handle.getinfo(pycurl.NAMELOOKUP_TIME)
        m['connect-time'] = self.handle.getinfo(pycurl.CONNECT_TIME)
        m['pretransfer-time'] = self.handle.getinfo(pycurl.PRETRANSFER_TIME)
        m['redirect-time'] = self.handle.getinfo(pycurl.REDIRECT_TIME)
        m['redirect-count'] = self.handle.getinfo(pycurl.REDIRECT_COUNT)
        m['size-upload'] = self.handle.getinfo(pycurl.SIZE_UPLOAD)
        m['size-download'] = self.handle.getinfo(pycurl.SIZE_DOWNLOAD)
        m['speed-upload'] = self.handle.getinfo(pycurl.SPEED_UPLOAD)
        m['header-size'] = self.handle.getinfo(pycurl.HEADER_SIZE)
        m['request-size'] = self.handle.getinfo(pycurl.REQUEST_SIZE)
        m['content-length-download'] = self.handle.getinfo(pycurl.CONTENT_LENGTH_DOWNLOAD)
        m['content-length-upload'] = self.handle.getinfo(pycurl.CONTENT_LENGTH_UPLOAD)
        m['content-type'] = self.handle.getinfo(pycurl.CONTENT_TYPE)
        m['response-code'] = self.handle.getinfo(pycurl.RESPONSE_CODE)
        m['speed-download'] = self.handle.getinfo(pycurl.SPEED_DOWNLOAD)
        m['ssl-verifyresult'] = self.handle.getinfo(pycurl.SSL_VERIFYRESULT)
        m['filetime'] = self.handle.getinfo(pycurl.INFO_FILETIME)
        m['starttransfer-time'] = self.handle.getinfo(pycurl.STARTTRANSFER_TIME)
        m['redirect-time'] = self.handle.getinfo(pycurl.REDIRECT_TIME)
        m['redirect-count'] = self.handle.getinfo(pycurl.REDIRECT_COUNT)
        m['http-connectcode'] = self.handle.getinfo(pycurl.HTTP_CONNECTCODE)
        m['httpauth-avail'] = self.handle.getinfo(pycurl.HTTPAUTH_AVAIL)
        m['proxyauth-avail'] = self.handle.getinfo(pycurl.PROXYAUTH_AVAIL)
        m['os-errno'] = self.handle.getinfo(pycurl.OS_ERRNO)
        m['num-connects'] = self.handle.getinfo(pycurl.NUM_CONNECTS)
        m['ssl-engines'] = self.handle.getinfo(pycurl.SSL_ENGINES)
        m['cookielist'] = self.handle.getinfo(pycurl.INFO_COOKIELIST)
        m['lastsocket'] = self.handle.getinfo(pycurl.LASTSOCKET)
        m['ftp-entry-path'] = self.handle.getinfo(pycurl.FTP_ENTRY_PATH)
        return m
项目:falsy    作者:pingf    | 项目源码 | 文件源码
def perform(cls):
        if cls._futures:
            while True:
                status, num_active = cls._multi.perform()
                if status != pycurl.E_CALL_MULTI_PERFORM:
                    break
            while True:
                num_ready, success, fail = cls._multi.info_read()
                for c in success:
                    cc = cls._futures.pop(c)
                    result = curl_result(c)
                    result['url'] = c._raw_url
                    result['id'] = c._raw_id
                    result['state'] = 'normal'
                    result['spider'] = 'pycurl'
                    result['payload'] = payload = c._raw_payload

                    # post_func = payload.get('post_func')
                    # if type(post_func) == str:
                    #     post_func = load(post_func)
                    # if post_func:
                    #     result = post_func(payload, result)

                    cc.set_result(result)
                for c, err_num, err_msg in fail:
                    print('error:', err_num, err_msg, c.getinfo(pycurl.EFFECTIVE_URL))
                    result = curl_result(c)

                    result['url'] = c._raw_url
                    result['id'] = c._raw_id
                    result['state'] = 'error'
                    result['spider'] = 'pycurl'
                    result['error_code'] = err_num
                    result['error_desc'] = err_msg

                    result['payload'] = payload = c._raw_payload

                    # post_func = payload.get('post_func')
                    # if type(post_func) == str:
                    #     post_func = load(post_func)
                    # if post_func:
                    #     result2 = post_func(payload, result)
                    #     if type(result2) is dict and len(result2) >= len(result):
                    #         result = result2
                    cls._futures.pop(c).set_exception(CurlLoop.CurlException(code=err_num, desc=err_msg, data=result))
                if num_ready == 0:
                    break
项目:falsy    作者:pingf    | 项目源码 | 文件源码
def curl_result(c):
    effective_url = c.getinfo(pycurl.EFFECTIVE_URL)
    primary_ip = c.getinfo(pycurl.PRIMARY_IP)
    primary_port = c.getinfo(pycurl.PRIMARY_PORT)
    local_ip = c.getinfo(pycurl.LOCAL_IP)
    local_port = c.getinfo(pycurl.LOCAL_PORT)
    speed_download = c.getinfo(pycurl.SPEED_DOWNLOAD)
    size_download = c.getinfo(pycurl.SIZE_DOWNLOAD)
    redirect_time = c.getinfo(pycurl.REDIRECT_TIME)
    redirect_count = c.getinfo(pycurl.REDIRECT_COUNT)
    redirect_url = c.getinfo(pycurl.REDIRECT_URL)
    http_code = c.getinfo(pycurl.HTTP_CODE)
    response_code = c.getinfo(pycurl.RESPONSE_CODE)
    total_time = c.getinfo(pycurl.TOTAL_TIME)
    content_type = c.getinfo(pycurl.CONTENT_TYPE)
    namelookup_time = c.getinfo(pycurl.NAMELOOKUP_TIME)
    info_filetime = c.getinfo(pycurl.INFO_FILETIME)
    http_connectcode = c.getinfo(pycurl.HTTP_CONNECTCODE)
    starttransfer_time = c.getinfo(pycurl.STARTTRANSFER_TIME)
    pretransfer_time = c.getinfo(pycurl.PRETRANSFER_TIME)
    header_size = c.getinfo(pycurl.HEADER_SIZE)
    request_size = c.getinfo(pycurl.REQUEST_SIZE)
    ssl_verifyresult = c.getinfo(pycurl.SSL_VERIFYRESULT)
    num_connects = c.getinfo(pycurl.NUM_CONNECTS)

    return {
        'effective_url': effective_url,
        'primary_ip': primary_ip,
        'primary_port': primary_port,
        'local_ip': local_ip,
        'local_port': local_port,
        'speed_download': speed_download,
        'size_download': size_download,
        'redirect_time': redirect_time,
        'redirect_count': redirect_count,
        'redirect_url': redirect_url,
        'http_code': http_code,
        'response_code': response_code,
        'total_time': total_time,
        'content_type': content_type,
        'namelookup_time': namelookup_time,
        'info_filetime': info_filetime,
        'http_connectcode': http_connectcode,
        'starttransfer_time': starttransfer_time,
        'pretransfer_time': pretransfer_time,
        'header_size': header_size,
        'request_size': request_size,
        'ssl_verifyresult': ssl_verifyresult,
        'num_connects': num_connects,
        # 'proxy_ssl_verifyresult': proxy_ssl_verifyresult,
        # 'app_connecttime': app_connecttime,

    }
项目:pyload-requests    作者:pyload    | 项目源码 | 文件源码
def load(self, url, get={}, post={}, referer=True, cookies=True,
             just_header=False, multipart=False, decode=False):
        """
        Load and returns a given page.
        """
        self.set_request_context(url, get, post, referer, cookies, multipart)

        # TODO: use http/rfc message instead
        self.header = ""

        if "header" in self.options:
            # TODO
            # print("custom header not implemented")
            self.setopt(pycurl.HTTPHEADER, self.options['header'])

        if just_header:
            self.setopt(pycurl.FOLLOWLOCATION, 0)
            self.setopt(pycurl.NOBODY, 1)  # TODO: nobody= no post?

            # overwrite HEAD request, we want a common request type
            if post:
                self.setopt(pycurl.CUSTOMREQUEST, 'POST')
            else:
                self.setopt(pycurl.CUSTOMREQUEST, 'GET')

            try:
                self.c.perform()
                rep = self.header
            finally:
                self.setopt(pycurl.FOLLOWLOCATION, 1)
                self.setopt(pycurl.NOBODY, 0)
                self.unsetopt(pycurl.CUSTOMREQUEST)

        else:
            self.c.perform()
            rep = self.get_response()

        self.setopt(pycurl.POSTFIELDS, '')
        self.last_url = safequote(url)
        self.last_effective_url = self.c.getinfo(pycurl.EFFECTIVE_URL)
        if self.last_effective_url:
            self.last_url = self.last_effective_url
        self.code = self.verify_header()

        if cookies:
            self.parse_cookies()

        if decode:
            rep = self.decode_response(rep)

        return rep