初探selenium3原理

从一个启动浏览器并打开百度网页的代码开始

from selenium import webdriver
driver = webdriver.chrome()
driver.get('https://www.baidu.com')

from selenium import webdriver
​ 这代码表示从selenium导入webdriver。进入selenium, 发现webdriver是一个包，那么导入的其实是webdriver包下的`___init__.py`文件

from .firefox.webdriver import WebDriver as Firefox  # noqa
from .firefox.firefox_profile import FirefoxProfile  # noqa
from .firefox.options import Options as FirefoxOptions  # noqa
#实例化的是.chrome.webdriver里的webDriver
from .chrome.webdriver import WebDriver as Chrome  # noqa
from .chrome.options import Options as ChromeOptions  # noqa
from .ie.webdriver import WebDriver as Ie  # noqa
from .ie.options import Options as IeOptions  # noqa
from .edge.webdriver import WebDriver as Edge  # noqa
from .opera.webdriver import WebDriver as Opera  # noqa
from .safari.webdriver import WebDriver as Safari  # noqa
from .blackberry.webdriver import WebDriver as BlackBerry #noqa
from .phantomjs.webdriver import WebDriver as PhantomJS # noqa
from .android.webdriver import WebDriver as Android  # noqa
from .webkitgtk.webdriver import WebDriver as WebKitGTK # noqa
from .webkitgtk.options import Options as WebKitGTKOptions
from .remote.webdriver import WebDriver as Remote  # noqa
from .common.desired_capabilities import DesiredCapabilities
from .common.action_chains import ActionChains  # noqa
from .common.touch_actions import TouchActions  # noqa
from .common.proxy import Proxy  # noqa

打开chrome.webdriver文件，下面只展示出相关代码

#selenium/webdriver/chrome/webdriver.py
import warnings
from selenium.webdriver.remote.webdriver import WebDriver as RemoteWebDriver
from .remote_connection import ChromeRemoteConnection
from .service import Service
from .options import Options
class WebDriver(RemoteWebDriver):

    def __init__(self, executable_path="chromedriver", port=0,
                 options=None, service_args=None,
                 desired_capabilities=None, service_log_path=None,
                 chrome_options=None):
        """
        参数：
         - executable_path - chromedriver的执行路径 默认在环境变里中查找
         - port -http连接的端口号
         - desired_capabilities: 一般浏览器的字典对象
         - options: ChromeOptions的实例
        """

        #………………………………省略…………………………………………
        #第1步 实例化一个Service对象
        self.service = Service(
            executable_path,
            port=port,
            service_args=service_args,
            log_path=service_log_path)
        #第2步 调用了service的start方法
        self.service.start()
        #………………………………省略…………………………………………

WebDriver构造方法中最先实例化Service类，我们实例化chrome() 并没有参数，所以Service 的参数 executable_path="chromedriver" port=0，其余都是None

打开Chrome目录Service文件, 只有以下代码

#selenium/webdriver/chrome/service.py
from selenium.webdriver.common import service

class Service(service.Service):
    """
    实例化Service对象 管理ChromeDriver的启动和停止
    """
    def __init__(self, executable_path, port=0, service_args=None,
                 log_path=None, env=None):
        """
         参数:
         - service_args : chromedriver 的参数 列表形式
         - log_path :  chromedriver的日志路径
        """

        self.service_args = service_args or []
        if log_path:
            self.service_args.append('--log-path=%s' % log_path)
        #第1步 调用复类的构造方法
        service.Service.__init__(self, executable_path, port=port, env=env,
                                 start_error_message="Please see https://sites.google.com/a/chromium.org/chromedriver/home")

    #重写父类方法 获取命令行的参数
    def command_line_args(self):
        return ["--port=%d" % self.port] + self.service_args

该类继承了selenium.webdriver.common目录下 service 类，并重写了父类的command_line_args方法。构造方法中调用了父类的构造方法。

#selenium/webdriver/common/service.py
import errno
import os
import platform
import subprocess
from subprocess import PIPE
import time
from selenium.common.exceptions import WebDriverException
from selenium.webdriver.common import utils

try:
    from subprocess import DEVNULL
    _HAS_NATIVE_DEVNULL = True
except ImportError:
    DEVNULL = -3
    _HAS_NATIVE_DEVNULL = False

class Service(object):

    def __init__(self, executable, port=0, log_file=DEVNULL, env=None, start_error_message=""):
        self.path = executable

        self.port = port
        #默认自动获取一个端口
        if self.port == 0:
            self.port = utils.free_port()

        if not _HAS_NATIVE_DEVNULL and log_file == DEVNULL:
            log_file = open(os.devnull, 'wb')

        self.start_error_message = start_error_message
        self.log_file = log_file
        #默认获取系统的环境变量
        self.env = env or os.environ

    @property
    def service_url(self):
        """
        Gets the url of the Service
        """
        return "http://%s" % utils.join_host_port('localhost', self.port)

    def command_line_args(self):
        raise NotImplemented("This method needs to be implemented in a sub class")

    def start(self):
        """
        Starts the Service.

        :Exceptions:
         - WebDriverException : Raised either when it can't start the service
           or when it can't connect to the service
        """
        try:
            #启动chromedriver程序 参数为 --port=端口号 输入输出到devnull空设备
            cmd = [self.path]
            cmd.extend(self.command_line_args())
            self.process = subprocess.Popen(cmd, env=self.env,
                                            close_fds=platform.system() != 'Windows',
                                            stdout=self.log_file,
                                            stderr=self.log_file,
                                            stdin=PIPE)
        except TypeError:
            raise
        except OSError as err:
            if err.errno == errno.ENOENT:
                raise WebDriverException(
                    "'%s' executable needs to be in PATH. %s" % (
                        os.path.basename(self.path), self.start_error_message)
                )
            elif err.errno == errno.EACCES:
                raise WebDriverException(
                    "'%s' executable may have wrong permissions. %s" % (
                        os.path.basename(self.path), self.start_error_message)
                )
            else:
                raise
        except Exception as e:
            raise WebDriverException(
                "The executable %s needs to be available in the path. %s\n%s" %
                (os.path.basename(self.path), self.start_error_message, str(e)))
        count = 0
        #检测是否subprocess进程是否还在，不在则抛出异常
        #检测是否http协议是否链接 若无链接等待30秒抛出异常
        while True:
            self.assert_process_still_running()
            if self.is_connectable():
                break
            count += 1
            time.sleep(1)
            if count == 30:
                raise WebDriverException("Can not connect to the Service %s" % self.path)

    def assert_process_still_running(self):
         return_code = self.process.poll()
         if return_code is not None:
             raise WebDriverException(
                  'Service %s unexpectedly exited. Status code was: %s'
                   % (self.path, return_code)
             )
　　 #判断是否正在连接，等待30秒后抛出webdriver异常
    def is_connectable(self):
          return utils.is_connectable(self.port)

由上代码可知Serivce的实例化 获取一个端口。
然后调用了service对象的start方法。该方法用subprocess启动chromedriver程序 并检测是否正在连接。
现在再来看最开始chrome 的webDriver类,  此类继承了selenium.webdriver.remote下的webdriver并调用了父类的构造方法。

#selenium/webdriver/remote/webdriver.py
import warnings
from selenium.webdriver.remote.webdriver import WebDriver as RemoteWebDriver
from .remote_connection import ChromeRemoteConnection
from .service import Service
from .options import Options
class WebDriver(RemoteWebDriver):
    """
    Controls the ChromeDriver and allows you to drive the browser.

    You will need to download the ChromeDriver executable from
    http://chromedriver.storage.googleapis.com/index.html
    """
    def __init__(self, executable_path="chromedriver", port=0,
                 options=None, service_args=None,
                 desired_capabilities=None, service_log_path=None,
                 chrome_options=None):
           #………………………………省略…………………………………………

        if options is None:
            # desired_capabilities stays as passed in
            if desired_capabilities is None:
                  #第1步 创建一个浏览器的字典对象
                desired_capabilities = self.create_options().to_capabilities()
        else:
            if desired_capabilities is None:
                desired_capabilities = options.to_capabilities()
            else:
                desired_capabilities.update(options.to_capabilities())
        #………………………………省略…………………………………………

          #第二步调用 复类的构造方法
          try:
            RemoteWebDriver.__init__(
                self,
                command_executor=ChromeRemoteConnection(
                    remote_server_addr=self.service.service_url),
                desired_capabilities=desired_capabilities)
        except Exception:
            self.quit()
            raise
        self._is_remote = False

    def create_options(self):
        return Options()
      

首先创建一个浏览器的字典对象，然后调用了to_capabilities()方法。

Options的to_capabilities()方法是返回一个caps字典对象

chrome浏览器返回的caps字典对象为：
    {
     'browserName': 'chrome',
     'version': '',
     'platform': 'ANY',
     'goog:chromeOptions': {'extensions': [], 'args': []}
     }

接下来看看 RemoteWebDriver的构造方法

 RemoteWebDriver.__init__(
            self,
            command_executor=ChromeRemoteConnection(
                remote_server_addr=self.service.service_url),
            desired_capabilities=desired_capabilities)

传入了2个参数 一个是 ChromeRemoteConnection类的实例对象，  一个是前面获取到的浏览器字典对象。
来看看ChromeRemoteConnection类。继承了RemoteConnection，调用了父类的构造方法并往self._commands添加里几个command键值对

#selenium/webdriver/chrome/remote_connection.py
from selenium.webdriver.remote.remote_connection import RemoteConnection

class ChromeRemoteConnection(RemoteConnection):

    def __init__(self, remote_server_addr, keep_alive=True):
        RemoteConnection.__init__(self, remote_server_addr, keep_alive)
        self._commands["launchApp"] = ('POST', '/session/$sessionId/chromium/launch_app')
        self._commands["setNetworkConditions"] = ('POST', '/session/$sessionId/chromium/network_conditions')
        self._commands["getNetworkConditions"] = ('GET', '/session/$sessionId/chromium/network_conditions')

#selenium/webdriver/remote/remote_connection.py
class RemoteConnection(object):
    """A connection with the Remote WebDriver server.

    Communicates with the server using the WebDriver wire protocol:
    https://github.com/SeleniumHQ/selenium/wiki/JsonWireProtocol"""

    def __init__(self, remote_server_addr, keep_alive=False, resolve_ip=True):
        # Attempt to resolve the hostname and get an IP address.
        self.keep_alive = keep_alive
        parsed_url = parse.urlparse(remote_server_addr)
        addr = parsed_url.hostname
        if parsed_url.hostname and resolve_ip:
            port = parsed_url.port or None
            if parsed_url.scheme == "https":
                ip = parsed_url.hostname
            elif port and not common_utils.is_connectable(port, parsed_url.hostname):
                ip = None
                LOGGER.info('Could not connect to port {} on host '
                            '{}'.format(port, parsed_url.hostname))
            else:
                ip = common_utils.find_connectable_ip(parsed_url.hostname,
                                                      port=port)
            if ip:
                netloc = ip
                addr = netloc
                if parsed_url.port:
                    netloc = common_utils.join_host_port(netloc,
                                                         parsed_url.port)
                if parsed_url.username:
                    auth = parsed_url.username
                    if parsed_url.password:
                        auth += ':%s' % parsed_url.password
                    netloc = '%s@%s' % (auth, netloc)
                remote_server_addr = parse.urlunparse(
                    (parsed_url.scheme, netloc, parsed_url.path,
                     parsed_url.params, parsed_url.query, parsed_url.fragment))
            else:
                LOGGER.info('Could not get IP address for host: %s' %
                            parsed_url.hostname)

        self._url = remote_server_addr
        if keep_alive:
            self._conn = httplib.HTTPConnection(
                str(addr), str(parsed_url.port), timeout=self._timeout)

        self._commands = {
            Command.STATUS: ('GET', '/status'),
            Command.NEW_SESSION: ('POST', '/session'),
            Command.GET_ALL_SESSIONS: ('GET', '/sessions'),
            Command.QUIT: ('DELETE', '/session/$sessionId'),
            Command.GET_CURRENT_WINDOW_HANDLE:
                ('GET', '/session/$sessionId/window_handle'),
            Command.W3C_GET_CURRENT_WINDOW_HANDLE:
                ('GET', '/session/$sessionId/window'),
            Command.GET_WINDOW_HANDLES:
                ('GET', '/session/$sessionId/window_handles'),

          #................省略.................
        }

     #最终发送命令到远程服务器的方法
     def execute(self, command, params):
        command_info = self._commands[command]
        assert command_info is not None, 'Unrecognised command %s' % command
        path = string.Template(command_info[1]).substitute(params)
        if hasattr(self, 'w3c') and self.w3c and isinstance(params, dict) and 'sessionId' in params:
            del params['sessionId']
        data = utils.dump_json(params)
        url = '%s%s' % (self._url, path)
        return self._request(command_info[0], url, body=data)

     #返回带有JSON解析的字典
     def _request(self, method, url, body=None):
        """
        Send an HTTP request to the remote server.

        :Args:
         - method - A string for the HTTP method to send the request with.
         - url - A string for the URL to send the request to.
         - body - A string for request body. Ignored unless method is POST or PUT.

        :Returns:
          A dictionary with the server's parsed JSON response.
        """
        LOGGER.debug('%s %s %s' % (method, url, body))

        parsed_url = parse.urlparse(url)
        headers = self.get_remote_connection_headers(parsed_url, self.keep_alive)
        resp = None
        if body and method != 'POST' and method != 'PUT':
            body = None

        if self.keep_alive:
            resp = self._conn.request(method, url, body=body, headers=headers)

            statuscode = resp.status
        else:
            http = urllib3.PoolManager(timeout=self._timeout)
            resp = http.request(method, url, body=body, headers=headers)

            statuscode = resp.status
            if not hasattr(resp, 'getheader'):
                if hasattr(resp.headers, 'getheader'):
                    resp.getheader = lambda x: resp.headers.getheader(x)
                elif hasattr(resp.headers, 'get'):
                    resp.getheader = lambda x: resp.headers.get(x)

        data = resp.data.decode('UTF-8')
        try:
            if 300 <= statuscode < 304:
                return self._request('GET', resp.getheader('location'))
            if 399 < statuscode <= 500:
                return {'status': statuscode, 'value': data}
            content_type = []
            if resp.getheader('Content-Type') is not None:
                content_type = resp.getheader('Content-Type').split(';')
            if not any([x.startswith('image/png') for x in content_type]):

                try:
                    data = utils.load_json(data.strip())
                except ValueError:
                    if 199 < statuscode < 300:
                        status = ErrorCode.SUCCESS
                    else:
                        status = ErrorCode.UNKNOWN_ERROR
                    return {'status': status, 'value': data.strip()}

                # Some of the drivers incorrectly return a response
                # with no 'value' field when they should return null.
                if 'value' not in data:
                    data['value'] = None
                return data
            else:
                data = {'status': 0, 'value': data}
                return data
        finally:
            LOGGER.debug("Finished Request")
            resp.close()

构造方法中主要是把localhost域名换成127.0.0.1，通过urllib.parse.urlparse把要处理的url解析6大部分。

urlparse返回的是一个名字元组对象scheme, netloc, path, params, query, fragment。netloc包括hostname和port。

调用 common_utils.find_connectable_ip()方法获取hostname对应的ip地址，最后urllib.parse.urlunparse()重新组成url并赋值给self._url

初始化里self._commands 字典，value为具体执行的命令的字典。

RemoteConnection类的实例方法execute调用 _request方法最终实现发送命令到远程服务器。

他们是通过wire protocol有线协议 这种协议是点对点方式进行通信的。首先前端将这个点击转换成json格式的字符串，然后通过wire protocl协议传递给服务器

RemoteWebDriver类的构造方法 更新capabilities字典 主要调用start_session传入capabilities字典

start_session方法 根据capabilities字典创建一个新的会话并获取session_id。

另外还实例化了错误处理handle，文件查找file_detector（默认实例化是LocalFileDetector）。一个页面切换的SwitchTo对象。

#selenium/webdriver/remote/webdriver.py
class WebDriver(object):

    _web_element_cls = WebElement

    def __init__(self, command_executor='http://127.0.0.1:4444/wd/hub',
                 desired_capabilities=None, browser_profile=None, proxy=None,
                 keep_alive=False, file_detector=None, options=None):
        """
        创建一个driver使用 wire协议发送命令
        参数:
         - command_executor -  远程服务器的url 'http://127.0.0.1:端口号'
         - desired_capabilities - A dictionary of capabilities to request when
             starting the browser session. 必选参数
         - proxy - 一个selenium.webdriver.common.proxy.Proxy 对象. 可选的
         - file_detector - 自定义文件检测器对象. 默认使用LocalFileDetector()
         - options - options.Options类的实例
        """
        capabilities = {}
        if options is not None:
            capabilities = options.to_capabilities()
        if desired_capabilities is not None:
            if not isinstance(desired_capabilities, dict):
                raise WebDriverException("Desired Capabilities must be a dictionary")
            else:
                  #更新capabilities字典
                capabilities.update(desired_capabilities)
        if proxy is not None:
            warnings.warn("Please use FirefoxOptions to set proxy",
                          DeprecationWarning)
            proxy.add_to_capabilities(capabilities)
        self.command_executor = command_executor
        if type(self.command_executor) is bytes or isinstance(self.command_executor, str):
            self.command_executor = RemoteConnection(command_executor, keep_alive=keep_alive)
        self._is_remote = True
        #控制浏览器会话的字符串id
        self.session_id = None
        self.capabilities = {}
        #errorhandler.ErrorHandler 处理错误的handler
        self.error_handler = ErrorHandler()
        self.start_client()
        if browser_profile is not None:
            warnings.warn("Please use FirefoxOptions to set browser profile",
                          DeprecationWarning)
        #核心代码 开始一个会话
        self.start_session(capabilities, browser_profile)
        #实例化页面切换对象
        self._switch_to = SwitchTo(self)
        #app
        self._mobile = Mobile(self)
        #默认实例化LocalFileDetector对象
        self.file_detector = file_detector or LocalFileDetector

   def start_session(self, capabilities, browser_profile=None):
        """
        根据capabilities字典创建一个新的会话
        browser_profile FirefoxProfile的一个对象 只有火狐浏览器
        """
        if not isinstance(capabilities, dict):
            raise InvalidArgumentException("Capabilities must be a dictionary")
        if browser_profile:
            if "moz:firefoxOptions" in capabilities:
                capabilities["moz:firefoxOptions"]["profile"] = browser_profile.encoded
            else:
                capabilities.update({'firefox_profile': browser_profile.encoded})
        """
        _make_w3c_caps return dict
            {
               "firstMatch": [{}],
               "alwaysMatch": {
                   'browserName': 'chrome',
                   'version': '',
                   'platformName': 'any',
                   'goog:chromeOptions': {'extensions': [], 'args': []}
                   }
             }
        """
        w3c_caps = _make_w3c_caps(capabilities)
        parameters = {"capabilities": w3c_caps,
                      "desiredCapabilities": capabilities}
        #Command.NEW_SESSION: ('POST', '/session'),
        response = self.execute(Command.NEW_SESSION, parameters)
        if 'sessionId' not in response:
            response = response['value']
        #获取session_id
        self.session_id = response['sessionId']

        self.capabilities = response.get('value')

        # if capabilities is none we are probably speaking to
        # a W3C endpoint
        if self.capabilities is None:
            self.capabilities = response.get('capabilities')

        # Double check to see if we have a W3C Compliant browser
        self.w3c = response.get('status') is None
        self.command_executor.w3c = self.w3c

   def _make_w3c_caps(caps):
    """Makes a W3C alwaysMatch capabilities object.

    Filters out capability names that are not in the W3C spec. Spec-compliant
    drivers will reject requests containing unknown capability names.

    Moves the Firefox profile, if present, from the old location to the new Firefox
    options object.

    :Args:
     - caps - A dictionary of capabilities requested by the caller.
    """
    #深拷贝
    caps = copy.deepcopy(caps)
    #因为浏览器chrome 所以profile为None
    profile = caps.get('firefox_profile')
    always_match = {}

    if caps.get('proxy') and caps['proxy'].get('proxyType'):
        caps['proxy']['proxyType'] = caps['proxy']['proxyType'].lower()

    for k, v in caps.items():
          #如果caps的key 在_OSS_W3C_CONVERSION key中 而且caps的key对应的值不为空
        if v and k in _OSS_W3C_CONVERSION:
              #always_match的key 为_OSS_W3C_CONVERSION字典的值  value是caps字典的值
            always_match[_OSS_W3C_CONVERSION[k]] = v.lower() if k == 'platform' else v
        if k in _W3C_CAPABILITY_NAMES or ':' in k:
            always_match[k] = v
    if profile:
        moz_opts = always_match.get('moz:firefoxOptions', {})
        # If it's already present, assume the caller did that intentionally.
        if 'profile' not in moz_opts:
            # Don't mutate the original capabilities.
            new_opts = copy.deepcopy(moz_opts)
            new_opts['profile'] = profile
            always_match['moz:firefoxOptions'] = new_opts
    return {"firstMatch": [{}], "alwaysMatch": always_match}

_OSS_W3C_CONVERSION = {
    'acceptSslCerts': 'acceptInsecureCerts',
    'version': 'browserVersion',
    'platform': 'platformName'
}
    #通过self.command_executor.execute发送cmd命令到远程服务器达到控制浏览器的目标。

    def execute(self, driver_command, params=None):
        """
        通过command.CommandExecutor执行driver_command命令
        返回一个字典对象 里面装着JSON response
        """
        if self.session_id is not None:
            if not params:
                params = {'sessionId': self.session_id}
            elif 'sessionId' not in params:
                params['sessionId'] = self.session_id

        #数据封包
        params = self._wrap_value(params)
        #核心代码 执行cmmand_executor实例对象的execute方法
        response = self.command_executor.execute(driver_command, params)
        if response:
            self.error_handler.check_response(response)
            #数据解包
            response['value'] = self._unwrap_value(
                response.get('value', None))
            return response
        # If the server doesn't send a response, assume the command was
        # a success
        return {'success': 0, 'value': None, 'sessionId': self.session_id}

driver.get('https://www.baidu.com')调用的是webdriver/remote/webdriver.py下的get方法
get方法调用了remote_connection.py中execute的方法，remote_connection.py中execute的方法中self.command_executor.execute实际调用的是RemoteConnection.py的execute方法。
实际上是一个HTTP request给监听端口上的Web Service， 在我们的HTTP request的body中，会以WebDriver Wire协议规定的JSON格式的字符串来告诉Selenium我们希望浏览器打开'https://www.baidu.com'页面

#selenium/webdriver/remote/webdriver.py

    def get(self, url):
        """
        Loads a web page in the current browser session.
        """
        #Command.GET: ('POST', '/session/$sessionId/url'),
        self.execute(Command.GET, {'url': url})

总结一下：
首先是webdriver实例化Service 类调用start()方法用subprocess启动chromedriver(带--port参数)驱动。chromedriver启动之后都会在绑定的端口启动Web Service。

接着实例化RemoteConnection获得 command_executor实例化对象 传入给RemoteWebDriver构造方法。

RemoteWebDriver构造方法 start_session()方法启动session并获得唯一的session_id，通过这个session_id来确定找到对方且在多线程并行的时候彼此之间不会有冲突和干扰）

接下来调用WebDriver的任何API，比如get()  都需要借助一个ComandExecutor（remote_connection类的实例对象）调用execute()发送一个命令(这个命令在ComandExecutor实例化时候生成的一个command字典)。

#部分
self._commands = {
        Command.STATUS: ('GET', '/status'),
        Command.NEW_SESSION: ('POST', '/session'),
        Command.GET_ALL_SESSIONS: ('GET', '/sessions'),
        Command.QUIT: ('DELETE', '/session/$sessionId'),
        Command.GET_CURRENT_WINDOW_HANDLE:
            ('GET', '/session/$sessionId/window_handle'),
        Command.W3C_GET_CURRENT_WINDOW_HANDLE:
            ('GET', '/session/$sessionId/window'),
        Command.GET_WINDOW_HANDLES:
            ('GET', '/session/$sessionId/window_handles'),

  #.................省略.....................
}

ComandExecutor中的execute()方法最后返回一个_request()方法，实际上是一个HTTP request给监听端口上的Web Service。

在HTTP request的body中，Wire JSON格式字典来告诉chromedriver接下来做什么事。（通过之前绑定的端口）
实际的执行者是chromedriver驱动，而selenium就相当于一个代理。所以selenium并不是直接操控浏览器而是运行webdriver， 通过webdriver间接操控浏览器。

在现实生活中这类似打出租车，我们告诉司机目的地是哪？走哪条路到达？webdriver就相当于出租车司机。