Yurl is the replacement of built in python urlparse module. Key features of yurl are:
- pythonic api
- better compliance with RFC 3986
- nice performance
- support for python 2.6, 2.7, 3.2, 3.3 and pypy 1.9 with single codebase
Yurl inspired by purl — pythonic interface to urlparse.
To parse url into parts, pass string as first argument to URL() constructor:
>>> from yurl import URL >>> URL('https://www.google.ru/search?q=yurl') URLBase(scheme='https', userinfo=u'', host='www.google.ru', port='', path='/search', query='q=yurl', fragment='', decoded=False)
It also works with relative urls:
>>> URL('search?rls=en&q=yurl&redir_esc=') URLBase(scheme=u'', userinfo=u'', host=u'', port='', path='search', query='rls=en&q=yurl&redir_esc=', fragment='', decoded=False)
Url also can be constructed from known parts:
>>> print URL(host='google.com', path='search', query='q=url') //google.com/search?q=url
Url parsing is always successful, even if some parts have unescaped or not allowed chars. After parsing you can call validate() method:
>>> URL('//google:com').validate() Traceback (most recent call last): File "<stdin>", line 1, in <module> File "yurl.py", line 201, in validate raise InvalidHost() yurl.InvalidHost
Validate() returns object itself or modified version:
>>> URL('//google.com:80').validate() URLBase(scheme=u'', userinfo=u'', host='google.com', port='80', path='', query='', fragment='', decoded=False)
URL() returns named tuple with some additional properties. All properties are strings, even if they don't exists in the url.
- .scheme .authority .path .query .fragment
- Basic parts of url: scheme://authority/path?query#fragment
- .userinfo .host .port
- Parts of authority: userinfo@host:port Port is guaranteed to consist of digits.
- .full_path
- Path, query and fragment joined together: path?query#fragment
- .username .authorization
- Parts of userinfo: username:authorization
Url object has method for checking authority existence:
>>> URL('http://google.com:80').has_authority() True
Also you can check if url is relative:
>>> URL('http://google.com:80').is_relative() False >>> URL('//google.com:80').is_relative() True
Or have relative path:
>>> URL('scheme:path').is_relative_path() False >>> URL('./path').is_relative_path() True
You can also check if url host is an IP address:
>>> URL('//127-0-0-1/').is_host_ip() False >>> URL('//127.0.0.1/').is_host_ip() True >>> URL('//[::ae21:ad12]/').is_host_ip() True >>> URL('//[::ae21:ad12]/').is_host_ipv4() False
Ip is not validated, so it is recommended to use validate() method:
>>> URL('//[+ae21:ad12]/').is_host_ip() True >>> URL('//[+ae21:ad12]/').validate().is_host_ip() Traceback (most recent call last): File "<stdin>", line 1, in <module> File "yurl.py", line 197, in validate raise InvalidHost() yurl.InvalidHost
After parsing, url can be modified in different ways.
You can use replace() method to change whole parts of url:
>>> print URL('http://ya.ru/').replace(scheme='https') https://ya.ru/ >>> print URL('http://ya.ru/?q=yurl').replace(query='') http://ya.ru/
In addition to the usual attributes it takes shortcuts authority and full_path:
>>> print URL('http://[email protected]:80/?q=yurl')\ ... .replace(authority='google.com', full_path='two') http://google.com/two
setdefault() replace parts with given if they don't exists in original url:
>>> print URL('https://google.com').setdefault(scheme='http', path='q') https://google.com/q
Join is analogue to urljoin() function from urlparse module. You can join two urls by adding one to another.
>>> print URL('http://ya.ru/path#chap2') + URL('seqrch?q=some') http://ya.ru/seqrch?q=some
Join for relative urls is also supported:
>>> print URL('path/to/object#chap2') + URL('../from/object') path/from/object
Join is not a commutative operation:
>>> print URL('../from/object') + URL('path/to/object#chap2') from/path/to/object#chap2
And not associative in general:
>>> print (URL('//google/path/to') + URL('../../object')) + URL('path') //google/path >>> print URL('//google/path/to') + (URL('../../object') + URL('path')) //google/path/path
All chars in url are divided in three groups: delimeters, subdelimeters and unreserved chars. Unreserved chars do not affect the parsing and can be encoded or decoded at any time. To decode unreserved chars you can call decode() method. Defaul encoding is utf-8.
>>> url = '%D1%81%D1%85%D0%B5%D0%BC%D0%B0%3A%D0%BF%D1%83%D1%82%D1%8C' >>> print URL(url).decode() схема%3Aпуть
If you want decode all chars, you should apply decode_url_component() function to url component:
>>> from yurl import decode_url_component >>> print decode_url_component(URL(url).decode().path) схема:путь
You can also omit decode method if you pass encoding in decode_url_component():
>>> print decode_url_component(url, 'utf-8') схема:путь
If you do not pass encoding, only reserved chars will be decoded:
>>> print decode_url_component(url) %D1%81%D1%85%D0%B5%D0%BC%D0%B0:%D0%BF%D1%83%D1%82%D1%8C
Original urlparse() cache every parsed url. In most cases this is unnecessary. But if you parse the same link again and again you can use CachedURL:
>>> CachedURL('http://host') is CachedURL('http://host') True
Rfc define format of valid url and ways to interact with it. But sometimes we need to interact with invalid urls. And RFC's not much help with it. So this library took many decisions.
Many libraries do not allow scheme or authority with invalid chars. Rfc unambiguously define format of this parts. So we can say 'sche_me:path' can not be scheme because of underscore and should be parsed as path:
>>> urlsplit('sche_me:path')[:] ('', '', 'sche_me:path', '', '')The problem is rfc also defines that the first segment of the path can not contain colon. I believe the right way is to split url as is and then validate if necessary.
>>> urlsplit('sche_me:path')[:] ('sche_me', '', 'path', '', '')Rfc define two operations against url: parse and join. As long as we can construct url from parts and replace parts we should sometimes fix this parts. For example url with authority can not be relative. And relative url can not starts with // or contain : in first path segment. These fixes can be done while url constructing or while recomposition. First way may be wrong because we can apply unnecessary in future fix:
>>> # This is example of wrong behavior. >>> print URL("//host") + URL(path="//path") //host////path # now path have four slashes
Second way is wrong when we replace some parts:
>>> # This is example of wrong behavior. >>> print URL("rel/path").replace(host='host').path rel/path # path is relative even if host there
So I divide all fixes to real fixes:
>>> # path can not be relative when host present >>> print URL("rel/path").replace(host='host').path /rel/path
And escapes which should be applied on recomposition:
>>> # url starts with path can not contain ':' in first path segment >>> print URL(path="rel:path") ./rel:path >>> print URL(path="rel:path").path rel:path
The short answer is urlparse is broken. If you're interested, here's detailed response.
urlparse module have two functions: urlparse() and urlsplit(). In addition to urlsplit(), urlparse() separates params from path. Params is not part of most schemas and in last rfc is not part of url at all. Instead of this each path segment can have own params. The problem is that most programmers use urlparse() and ignore params when extracting path:
>>> import purl >>> print purl.URL('/path;with?semicolon') /path?semicolon
urlsplit() has strange parameters. It takes default addressing scheme. But scheme can only have default value in urlsplit().
Another parameter allow_fragments can be used to prevent splitting #fragment from path. The problem is that we can't say «I do not want fragment in this url». If url contatin '#', it contatin frаgment. If scheme can not contatin fragment, '#' still can not be used in another parts. Caller has a choise: he can ignore fragment or raise. But url can not be parsed with ignoring '#':
>>> urlparse('/path#frag:ment?query').query '' >>> urlparse('/path#frag:ment?query', allow_fragments=False).query 'query'
Module makes no difference between parsing and validating. For example urlsplit() check allowed chars in scheme and raise on invalid IP URL:
>>> urlsplit('not_scheme://google.com').path 'not_scheme://google.com' >>> urlsplit('//ho[st/') Traceback (most recent call last): File "<stdin>", line 1, in <module> File "/usr/lib/python2.7/urlparse.py", line 211, in urlsplit raise ValueError("Invalid IPv6 URL") ValueError: Invalid IPv6 URL
But ignores other errors:
>>> urlsplit('//host@with@butterflies').username 'host@with' >>> urlsplit('//butterflies[]:80').port 80
It doesn't understand my favorite scheme:
>>> urlsplit('lucky-number:33')[:] ('', '', 'lucky-number:33', '', '')It loses path with two slashes:
>>> urlsplit('////path')[:] ('', '', '//path', '', '') >>> urlsplit(urlsplit('////path').geturl())[:] ('', 'path', '', '', '')
Function urljoin() is sometimes broken:
>>> urljoin('http://host/', '../') 'http://host/../' >>> print URL('http://host/') + URL('../') http://host
I'm sure the list is not complete.
Purl built on top of urlparse() and include almost all problems listed above. And some other:
Purl parsing is about 2 times slower then urlparse(), while yurl parsing is about 2 times faster then urlparse().
Purl manipulations is about 20 times slower then yurl:
>>> timeit("url.scheme('https')", "import purl; url = purl.URL('http://google.com/')", number=10000) 0.4427049160003662 >>> timeit("url.replace(scheme='https')", "import yurl; url = yurl.URL('http://google.com/')", number=10000) 0.020306110382080078
Purl have ugly jquery-like api, when one method may return different objects depending on the arguments.
Purl parsing is dangerous:
>>> purl.URL('//@host') ValueError: need more than 1 value to unpack >>> purl.URL('//host:/') ValueError: invalid literal for int() with base 10: '' >>> purl.URL('//user:pass:word@host') ValueError: too many values to unpack
Purl loses path after ';'. While ';' is valid char in url:
>>> print purl.URL('/path;with?semicolon') /path?semicolon
Purl loses host in relative urls:
>>> print purl.URL('//google.com/path?query') google.com/path?query
Purl loses username with empty password and password with empty username:
>>> print purl.URL('http://user:@google.com/') http://google.com/
Yurl comes with bunch of performance tests. Results may vary depending on the Python version and the CPU:
$ python2.7 ./test.py -bench === Test as string === yurl usplit uparse purl 12.01 9.783 11.94 27.08 !worse https://user:[email protected]:8080/path/to+the=ar?gum=ent#s 8.533 21.89 23.82 18.88 scheme:8080/path/to;the=ar?gum=ent#s 10.12 3.879 9.007 12.21 !worse re/ative:path;with?query 5.268 2.39 4.043 10.26 !worse lucky-number:3456 4.806 3.662 5.349 13.73 !worse //host:80 4.953 3.342 4.885 13.2 !worse #frag === Manipulations speed === noop yurl 0.0751 178.9 https://habrahabr.ru:80/a/b/c?d=f#h === Test join === = result is string = yurl ujoin 111.6 127.2 u'http://ya.ru/user/photos/id12324/photo3' + u'../../../mikhail/photos/id6543/photo99?param' 85.87 71.06 !worse u'http://ya.ru/user/photos/id12324' + u'#fragment' 82.12 100.8 u'http://ya.ru/' + u'https://google.com/?q=yurl' = result is parsed = yurl ujoin 102.6 181.3 u'http://ya.ru/user/photos/id12324/photo3' + u'../../../mikhail/photos/id6543/photo99?param' 73.15 125.7 u'http://ya.ru/user/photos/id12324' + u'#fragment' 76.26 184.3 u'http://ya.ru/' + u'https://google.com/?q=yurl' === Test parse === = dupass cache = yurl usplit uparse purl 36.25 73.31 85.91 166.5 https://user:[email protected]:8080/path/to+the=ar?gum=ent#s 20.34 58.84 77.29 138.9 scheme:8080/path/to;the=ar?gum=ent#s 18.25 33.21 48.72 109.3 re/ative:path;with?query 19.3 66.77 76.16 135.5 lucky-number:3456 24.0 35.57 43.36 119.2 //host:80 18.0 25.57 37.78 114.4 #frag = with cache = yurl usplit uparse purl 9.902 14.43 24.04 95.92 https://user:[email protected]:8080/path/to+the=ar?gum=ent#s 5.726 7.211 23.14 79.94 scheme:8080/path/to;the=ar?gum=ent#s 5.497 6.804 22.86 80.93 re/ative:path;with?query 5.357 6.521 14.72 72.0 lucky-number:3456 5.076 6.763 14.12 87.39 //host:80 5.824 7.993 26.78 73.03 #frag
In tests where any of the other libraries beats yurl you can see "!worse" marker.
- fixed installation on not utf-8 systems
- added URLError exception on top of ValueError
- decode() method
- username and authorization properties
- order of tuple members now same as url parts: scheme, userinfo, host, port, path, query, fragment
- raw url parsing was moved to split_url() function of utils module
- method replace_from() removed
- concatenation with string no longer aliasd with join
- join always remove dots segments (as defined in rfc)
First release.