bundled/webpy/web/browser.py @ 001f5ecfd9bc

Merge the webui skipped line counts -- too important to leave in the branch.
author Steve Losh <steve@stevelosh.com>
date Wed, 21 Oct 2009 19:58:13 -0400
parents 7c9d165e0cd1
children (none)
"""Browser to test web applications.
(from web.py)
"""
from utils import re_compile
from net import htmlunquote

import httplib, urllib, urllib2
import copy
from StringIO import StringIO

DEBUG = False

__all__ = [
    "BrowserError",
    "Browser", "AppBrowser",
    "AppHandler"
]

class BrowserError(Exception):
    pass

class Browser:
    def __init__(self):
        import cookielib
        self.cookiejar = cookielib.CookieJar()
        self._cookie_processor = urllib2.HTTPCookieProcessor(self.cookiejar)
        self.form = None

        self.url = "http://0.0.0.0:8080/"
        self.path = "/"
        
        self.status = None
        self.data = None
        self._response = None
        self._forms = None

    def reset(self):
        """Clears all cookies and history."""
        self.cookiejar.clear()

    def build_opener(self):
        """Builds the opener using urllib2.build_opener. 
        Subclasses can override this function to prodive custom openers.
        """
        return urllib2.build_opener()

    def do_request(self, req):
        if DEBUG:
            print 'requesting', req.get_method(), req.get_full_url()
        opener = self.build_opener()
        opener.add_handler(self._cookie_processor)
        try:
            self._response = opener.open(req)
        except urllib2.HTTPError, e:
            self._response = e

        self.url = self._response.geturl()
        self.path = urllib2.Request(self.url).get_selector()
        self.data = self._response.read()
        self.status = self._response.code
        self._forms = None
        self.form = None
        return self.get_response()

    def open(self, url, data=None, headers={}):
        """Opens the specified url."""
        url = urllib.basejoin(self.url, url)
        req = urllib2.Request(url, data, headers)
        return self.do_request(req)

    def show(self):
        """Opens the current page in real web browser."""
        f = open('page.html', 'w')
        f.write(self.data)
        f.close()

        import webbrowser, os
        url = 'file://' + os.path.abspath('page.html')
        webbrowser.open(url)

    def get_response(self):
        """Returns a copy of the current response."""
        return urllib.addinfourl(StringIO(self.data), self._response.info(), self._response.geturl())

    def get_soup(self):
        """Returns beautiful soup of the current document."""
        import BeautifulSoup
        return BeautifulSoup.BeautifulSoup(self.data)

    def get_text(self, e=None):
        """Returns content of e or the current document as plain text."""
        e = e or self.get_soup()
        return ''.join([htmlunquote(c) for c in e.recursiveChildGenerator() if isinstance(c, unicode)])

    def _get_links(self):
        soup = self.get_soup()
        return [a for a in soup.findAll(name='a')]
        
    def get_links(self, text=None, text_regex=None, url=None, url_regex=None, predicate=None):
        """Returns all links in the document."""
        return self._filter_links(self._get_links(),
            text=text, text_regex=text_regex, url=url, url_regex=url_regex, predicate=predicate)

    def follow_link(self, link=None, text=None, text_regex=None, url=None, url_regex=None, predicate=None):
        if link is None:
            links = self._filter_links(self.get_links(),
                text=text, text_regex=text_regex, url=url, url_regex=url_regex, predicate=predicate)
            link = links and links[0]
            
        if link:
            return self.open(link['href'])
        else:
            raise BrowserError("No link found")
            
    def find_link(self, text=None, text_regex=None, url=None, url_regex=None, predicate=None):
        links = self._filter_links(self.get_links(), 
            text=text, text_regex=text_regex, url=url, url_regex=url_regex, predicate=predicate)
        return links and links[0] or None
            
    def _filter_links(self, links, 
            text=None, text_regex=None,
            url=None, url_regex=None,
            predicate=None):
        predicates = []
        if text is not None:
            predicates.append(lambda link: link.string == text)
        if text_regex is not None:
            predicates.append(lambda link: re_compile(text_regex).search(link.string or ''))
        if url is not None:
            predicates.append(lambda link: link.get('href') == url)
        if url_regex is not None:
            predicates.append(lambda link: re_compile(url_regex).search(link.get('href', '')))
        if predicate:
            predicate.append(predicate)

        def f(link):
            for p in predicates:
                if not p(link):
                    return False
            return True

        return [link for link in links if f(link)]

    def get_forms(self):
        """Returns all forms in the current document.
        The returned form objects implement the ClientForm.HTMLForm interface.
        """
        if self._forms is None:
            import ClientForm
            self._forms = ClientForm.ParseResponse(self.get_response(), backwards_compat=False)
        return self._forms

    def select_form(self, name=None, predicate=None, index=0):
        """Selects the specified form."""
        forms = self.get_forms()

        if name is not None:
            forms = [f for f in forms if f.name == name]
        if predicate:
            forms = [f for f in forms if predicate(f)]
            
        if forms:
            self.form = forms[index]
            return self.form
        else:
            raise BrowserError("No form selected.")
        
    def submit(self, **kw):
        """submits the currently selected form."""
        if self.form is None:
            raise BrowserError("No form selected.")
        req = self.form.click(**kw)
        return self.do_request(req)

    def __getitem__(self, key):
        return self.form[key]

    def __setitem__(self, key, value):
        self.form[key] = value

class AppBrowser(Browser):
    """Browser interface to test web.py apps.
    
        b = AppBrowser(app)
        b.open('/')
        b.follow_link(text='Login')
        
        b.select_form(name='login')
        b['username'] = 'joe'
        b['password'] = 'secret'
        b.submit()

        assert b.path == '/'
        assert 'Welcome joe' in b.get_text()
    """
    def __init__(self, app):
        Browser.__init__(self)
        self.app = app

    def build_opener(self):
        return urllib2.build_opener(AppHandler(self.app))

class AppHandler(urllib2.HTTPHandler):
    """urllib2 handler to handle requests using web.py application."""
    handler_order = 100

    def __init__(self, app):
        self.app = app

    def http_open(self, req):
        result = self.app.request(
            localpart=req.get_selector(),
            method=req.get_method(),
            host=req.get_host(),
            data=req.get_data(),
            headers=dict(req.header_items()),
            https=req.get_type() == "https"
        )
        return self._make_response(result, req.get_full_url())

    def https_open(self, req):
        return self.http_open(req)
    
    try:
        https_request = urllib2.HTTPHandler.do_request_
    except AttributeError:
        # for python 2.3
        pass

    def _make_response(self, result, url):
        data = "\r\n".join(["%s: %s" % (k, v) for k, v in result.header_items])
        headers = httplib.HTTPMessage(StringIO(data))
        response = urllib.addinfourl(StringIO(result.data), headers, url)
        code, msg = result.status.split(None, 1)
        response.code, response.msg = int(code), msg
        return response