mirror of https://github.com/django/django.git
Fixed #1145 -- Added unit tests for default template filters and fixed two bugs in filters. Thanks, Luke Plant
git-svn-id: http://code.djangoproject.com/svn/django/trunk@1811 bcc190cf-cafb-0310-a4f2-bffc1f526a37
This commit is contained in:
parent
4596b1bbfd
commit
53aef92bf8
|
@ -117,7 +117,8 @@ def urlize(value):
|
|||
|
||||
def urlizetrunc(value, limit):
|
||||
"""
|
||||
Converts URLs into clickable links, truncating URLs to the given character limit
|
||||
Converts URLs into clickable links, truncating URLs to the given character limit,
|
||||
and adding 'rel=nofollow' attribute to discourage spamming.
|
||||
|
||||
Argument: Length to truncate URLs to.
|
||||
"""
|
||||
|
@ -254,7 +255,14 @@ def slice_(value, arg):
|
|||
for an introduction.
|
||||
"""
|
||||
try:
|
||||
return value[slice(*[x and int(x) or None for x in arg.split(':')])]
|
||||
bits = []
|
||||
for x in arg.split(':'):
|
||||
if len(x) == 0:
|
||||
bits.append(None)
|
||||
else:
|
||||
bits.append(int(x))
|
||||
return value[slice(*bits)]
|
||||
|
||||
except (ValueError, TypeError):
|
||||
return value # Fail silently.
|
||||
|
||||
|
|
|
@ -24,14 +24,16 @@ def timesince(d, now=None):
|
|||
else:
|
||||
tz = None
|
||||
now = datetime.datetime(t[0], t[1], t[2], t[3], t[4], t[5], tzinfo=tz)
|
||||
delta = now - d
|
||||
|
||||
# ignore microsecond part of 'd' since we removed it from 'now'
|
||||
delta = now - (d - datetime.timedelta(0, 0, d.microsecond))
|
||||
since = delta.days * 24 * 60 * 60 + delta.seconds
|
||||
for i, (seconds, name) in enumerate(chunks):
|
||||
count = since / seconds
|
||||
if count != 0:
|
||||
break
|
||||
if count < 0:
|
||||
return '%d milliseconds' % math.floor(delta.microseconds / 1000)
|
||||
return '%d milliseconds' % math.floor((now - d).microseconds / 1000)
|
||||
s = '%d %s' % (count, name(count))
|
||||
if i + 1 < len(chunks):
|
||||
# Now get the second item
|
||||
|
|
|
@ -11,9 +11,319 @@
|
|||
'0.0'
|
||||
>>> floatformat(0.0)
|
||||
'0'
|
||||
|
||||
>>> addslashes('"double quotes" and \\'single quotes\\'')
|
||||
'\\\\"double quotes\\\\" and \\\\\\'single quotes\\\\\\''
|
||||
|
||||
>>> capfirst('hello world')
|
||||
'Hello world'
|
||||
|
||||
>>> fix_ampersands('Jack & Jill & Jeroboam')
|
||||
'Jack & Jill & Jeroboam'
|
||||
|
||||
>>> linenumbers('line 1\\nline 2')
|
||||
'1. line 1\\n2. line 2'
|
||||
|
||||
>>> linenumbers('\\n'.join(['x'] * 10))
|
||||
'01. x\\n02. x\\n03. x\\n04. x\\n05. x\\n06. x\\n07. x\\n08. x\\n09. x\\n10. x'
|
||||
|
||||
>>> lower('TEST')
|
||||
'test'
|
||||
|
||||
>>> lower(u'\\xcb') # uppercase E umlaut
|
||||
u'\\xeb'
|
||||
|
||||
>>> make_list('abc')
|
||||
['a', 'b', 'c']
|
||||
|
||||
>>> make_list(1234)
|
||||
['1', '2', '3', '4']
|
||||
|
||||
>>> slugify(' Jack & Jill like numbers 1,2,3 and 4 and silly characters ?%.$!/')
|
||||
'jack-jill-like-numbers-123-and-4-and-silly-characters'
|
||||
|
||||
>>> stringformat(1, '03d')
|
||||
'001'
|
||||
|
||||
>>> stringformat(1, 'z')
|
||||
''
|
||||
|
||||
>>> title('a nice title, isn\\'t it?')
|
||||
"A Nice Title, Isn't It?"
|
||||
|
||||
|
||||
>>> truncatewords('A sentence with a few words in it', 1)
|
||||
'A ...'
|
||||
|
||||
>>> truncatewords('A sentence with a few words in it', 5)
|
||||
'A sentence with a few ...'
|
||||
|
||||
>>> truncatewords('A sentence with a few words in it', 100)
|
||||
'A sentence with a few words in it'
|
||||
|
||||
>>> truncatewords('A sentence with a few words in it', 'not a number')
|
||||
'A sentence with a few words in it'
|
||||
|
||||
|
||||
>>> upper('Mixed case input')
|
||||
'MIXED CASE INPUT'
|
||||
|
||||
>>> upper(u'\\xeb') # lowercase e umlaut
|
||||
u'\\xcb'
|
||||
|
||||
|
||||
>>> urlencode('jack & jill')
|
||||
'jack%20%26%20jill'
|
||||
|
||||
|
||||
>>> urlizetrunc('http://short.com/', 20)
|
||||
'<a href="http://short.com/" rel="nofollow">http://short.com/</a>'
|
||||
|
||||
>>> urlizetrunc('http://www.google.co.uk/search?hl=en&q=some+long+url&btnG=Search&meta=', 20)
|
||||
'<a href="http://www.google.co.uk/search?hl=en&q=some+long+url&btnG=Search&meta=" rel="nofollow">http://www.google.co...</a>'
|
||||
|
||||
>>> wordcount('')
|
||||
0
|
||||
|
||||
>>> wordcount('oneword')
|
||||
1
|
||||
|
||||
>>> wordcount('lots of words')
|
||||
3
|
||||
|
||||
>>> wordwrap('this is a long paragraph of text that really needs to be wrapped I\\'m afraid', 14)
|
||||
"this is a long\\nparagraph of\\ntext that\\nreally needs\\nto be wrapped\\nI'm afraid"
|
||||
|
||||
>>> ljust('test', 10)
|
||||
'test '
|
||||
|
||||
>>> ljust('test', 3)
|
||||
'test'
|
||||
|
||||
>>> rjust('test', 10)
|
||||
' test'
|
||||
|
||||
>>> rjust('test', 3)
|
||||
'test'
|
||||
|
||||
>>> center('test', 6)
|
||||
' test '
|
||||
|
||||
>>> cut('a string to be mangled', 'a')
|
||||
' string to be mngled'
|
||||
|
||||
>>> cut('a string to be mangled', 'ng')
|
||||
'a stri to be maled'
|
||||
|
||||
>>> cut('a string to be mangled', 'strings')
|
||||
'a string to be mangled'
|
||||
|
||||
>>> escape('<some html & special characters > here')
|
||||
'<some html & special characters > here'
|
||||
|
||||
>>> linebreaks('line 1')
|
||||
'<p>line 1</p>'
|
||||
|
||||
>>> linebreaks('line 1\\nline 2')
|
||||
'<p>line 1<br />line 2</p>'
|
||||
|
||||
>>> removetags('some <b>html</b> with <script>alert("You smell")</script> disallowed <img /> tags', 'script img')
|
||||
'some <b>html</b> with alert("You smell") disallowed tags'
|
||||
|
||||
>>> striptags('some <b>html</b> with <script>alert("You smell")</script> disallowed <img /> tags')
|
||||
'some html with alert("You smell") disallowed tags'
|
||||
|
||||
>>> dictsort([{'age': 23, 'name': 'Barbara-Ann'},\
|
||||
{'age': 63, 'name': 'Ra Ra Rasputin'},\
|
||||
{'name': 'Jonny B Goode', 'age': 18}], 'age')
|
||||
[{'age': 18, 'name': 'Jonny B Goode'},\
|
||||
{'age': 23, 'name': 'Barbara-Ann'},\
|
||||
{'age': 63, 'name': 'Ra Ra Rasputin'}]
|
||||
|
||||
>>> dictsortreversed([{'age': 23, 'name': 'Barbara-Ann'},\
|
||||
{'age': 63, 'name': 'Ra Ra Rasputin'},\
|
||||
{'name': 'Jonny B Goode', 'age': 18}], 'age')
|
||||
[{'age': 63, 'name': 'Ra Ra Rasputin'},\
|
||||
{'age': 23, 'name': 'Barbara-Ann'},\
|
||||
{'age': 18, 'name': 'Jonny B Goode'}]
|
||||
|
||||
>>> first([0,1,2])
|
||||
0
|
||||
|
||||
>>> first('')
|
||||
''
|
||||
|
||||
>>> first('test')
|
||||
't'
|
||||
|
||||
>>> join([0,1,2], 'glue')
|
||||
'0glue1glue2'
|
||||
|
||||
>>> length('1234')
|
||||
4
|
||||
|
||||
>>> length([1,2,3,4])
|
||||
4
|
||||
|
||||
>>> length_is([], 0)
|
||||
True
|
||||
|
||||
>>> length_is([], 1)
|
||||
False
|
||||
|
||||
>>> length_is('a', 1)
|
||||
True
|
||||
|
||||
>>> length_is('a', 10)
|
||||
False
|
||||
|
||||
>>> slice_('abcdefg', '0')
|
||||
''
|
||||
|
||||
>>> slice_('abcdefg', '1')
|
||||
'a'
|
||||
|
||||
>>> slice_('abcdefg', '-1')
|
||||
'abcdef'
|
||||
|
||||
>>> slice_('abcdefg', '1:2')
|
||||
'b'
|
||||
|
||||
>>> slice_('abcdefg', '1:3')
|
||||
'bc'
|
||||
|
||||
>>> slice_('abcdefg', '0::2')
|
||||
'aceg'
|
||||
|
||||
>>> unordered_list(['item 1', []])
|
||||
'\\t<li>item 1</li>'
|
||||
|
||||
>>> unordered_list(['item 1', [['item 1.1', []]]])
|
||||
'\\t<li>item 1\\n\\t<ul>\\n\\t\\t<li>item 1.1</li>\\n\\t</ul>\\n\\t</li>'
|
||||
|
||||
>>> unordered_list(['item 1', [['item 1.1', []], ['item 1.2', []]]])
|
||||
'\\t<li>item 1\\n\\t<ul>\\n\\t\\t<li>item 1.1</li>\\n\\t\\t<li>item 1.2</li>\\n\\t</ul>\\n\\t</li>'
|
||||
|
||||
>>> add('1', '2')
|
||||
3
|
||||
|
||||
>>> get_digit(123, 1)
|
||||
3
|
||||
|
||||
>>> get_digit(123, 2)
|
||||
2
|
||||
|
||||
>>> get_digit(123, 3)
|
||||
1
|
||||
|
||||
>>> get_digit(123, 4)
|
||||
0
|
||||
|
||||
>>> get_digit(123, 0)
|
||||
123
|
||||
|
||||
>>> get_digit('xyz', 0)
|
||||
'xyz'
|
||||
|
||||
# real testing of date() is in dateformat.py
|
||||
>>> date(datetime.datetime(2005, 12, 29), "d F Y")
|
||||
'29 December 2005'
|
||||
|
||||
# real testing of time() is done in dateformat.py
|
||||
>>> time(datetime.time(13), "h")
|
||||
'01'
|
||||
|
||||
# real testing is done in timesince.py, where we can provide our own 'now'
|
||||
>>> timesince(datetime.datetime.now() - datetime.timedelta(1))
|
||||
'1 day'
|
||||
|
||||
>>> default("val", "default")
|
||||
'val'
|
||||
|
||||
>>> default(None, "default")
|
||||
'default'
|
||||
|
||||
>>> default('', "default")
|
||||
'default'
|
||||
|
||||
>>> default_if_none("val", "default")
|
||||
'val'
|
||||
|
||||
>>> default_if_none(None, "default")
|
||||
'default'
|
||||
|
||||
>>> default_if_none('', "default")
|
||||
''
|
||||
|
||||
>>> divisibleby(4, 2)
|
||||
True
|
||||
|
||||
>>> divisibleby(4, 3)
|
||||
False
|
||||
|
||||
>>> yesno(True)
|
||||
'yes'
|
||||
|
||||
>>> yesno(False)
|
||||
'no'
|
||||
|
||||
>>> yesno(None)
|
||||
'maybe'
|
||||
|
||||
>>> yesno(True, 'certainly,get out of town,perhaps')
|
||||
'certainly'
|
||||
|
||||
>>> yesno(False, 'certainly,get out of town,perhaps')
|
||||
'get out of town'
|
||||
|
||||
>>> yesno(None, 'certainly,get out of town,perhaps')
|
||||
'perhaps'
|
||||
|
||||
>>> yesno(None, 'certainly,get out of town')
|
||||
'get out of town'
|
||||
|
||||
>>> filesizeformat(1023)
|
||||
'1023 bytes'
|
||||
|
||||
>>> filesizeformat(1024)
|
||||
'1.0 KB'
|
||||
|
||||
>>> filesizeformat(10*1024)
|
||||
'10.0 KB'
|
||||
|
||||
>>> filesizeformat(1024*1024-1)
|
||||
'1024.0 KB'
|
||||
|
||||
>>> filesizeformat(1024*1024)
|
||||
'1.0 MB'
|
||||
|
||||
>>> filesizeformat(1024*1024*50)
|
||||
'50.0 MB'
|
||||
|
||||
>>> filesizeformat(1024*1024*1024-1)
|
||||
'1024.0 MB'
|
||||
|
||||
>>> filesizeformat(1024*1024*1024)
|
||||
'1.0 GB'
|
||||
|
||||
>>> pluralize(1)
|
||||
''
|
||||
|
||||
>>> pluralize(0)
|
||||
's'
|
||||
|
||||
>>> pluralize(2)
|
||||
's'
|
||||
|
||||
>>> phone2numeric('0800 flowers')
|
||||
'0800 3569377'
|
||||
|
||||
|
||||
|
||||
"""
|
||||
|
||||
from django.core.template.defaultfilters import *
|
||||
import datetime
|
||||
|
||||
if __name__ == '__main__':
|
||||
import doctest
|
||||
|
|
|
@ -206,7 +206,7 @@ if __name__ == "__main__":
|
|||
parser.add_option('-v', help='How verbose should the output be? Choices are 0, 1 and 2, where 2 is most verbose. Default is 0.',
|
||||
type='choice', choices=['0', '1', '2'])
|
||||
parser.add_option('--settings',
|
||||
help='Python path to settings module, e.g. "myproject.settings.main". If this isn\'t provided, the DJANGO_SETTINGS_MODULE environment variable will be used.')
|
||||
help='Python path to settings module, e.g. "myproject.settings". If this isn\'t provided, the DJANGO_SETTINGS_MODULE environment variable will be used.')
|
||||
options, args = parser.parse_args()
|
||||
verbosity_level = 0
|
||||
if options.v:
|
||||
|
|
Loading…
Reference in New Issue