import pprint
import re
import unittest
from test import test_support
sgmllib = test_support.import_module('sgmllib', deprecated=True)
class EventCollector(sgmllib.SGMLParser):
def __init__(self):
self.events = []
self.append = self.events.append
sgmllib.SGMLParser.__init__(self)
def get_events(self):
# Normalize the list of events so that buffer artefacts don't
# separate runs of contiguous characters.
L = []
prevtype = None
for event in self.events:
type = event[0]
if type == prevtype == "data":
L[-1] = ("data", L[-1][1] + event[1])
else:
L.append(event)
prevtype = type
self.events = L
return L
# structure markup
def unknown_starttag(self, tag, attrs):
self.append(("starttag", tag, attrs))
def unknown_endtag(self, tag):
self.append(("endtag", tag))
# all other markup
def handle_comment(self, data):
self.append(("comment", data))
def handle_charref(self, data):
self.append(("charref", data))
def handle_data(self, data):
self.append(("data", data))
def handle_decl(self, decl):
self.append(("decl", decl))
def handle_entityref(self, data):
self.append(("entityref", data))
def handle_pi(self, data):
self.append(("pi", data))
def unknown_decl(self, decl):
self.append(("unknown decl", decl))
class CDATAEventCollector(EventCollector):
def start_cdata(self, attrs):
self.append(("starttag", "cdata", attrs))
self.setliteral()
class HTMLEntityCollector(EventCollector):
entity_or_charref = re.compile('(?:&([a-zA-Z][-.a-zA-Z0-9]*)'
'|&#(x[0-9a-zA-Z]+|[0-9]+))(;?)')
def convert_charref(self, name):
self.append(("charref", "convert", name))
if name[0] != "x":
return EventCollector.convert_charref(self, name)
def convert_codepoint(self, codepoint):
self.append(("codepoint", "convert", codepoint))
EventCollector.convert_codepoint(self, codepoint)
def convert_entityref(self, name):
self.append(("entityref", "convert", name))
return EventCollector.convert_entityref(self, name)
# These to record that they were called, then pass the call along
# to the default implementation so that it's actions can be
# recorded.
def handle_charref(self, data):
self.append(("charref", data))
sgmllib.SGMLParser.handle_charref(self, data)
def handle_entityref(self, data):
self.append(("entityref", data))
sgmllib.SGMLParser.handle_entityref(self, data)
class SGMLParserTestCase(unittest.TestCase):
collector = EventCollector
def get_events(self, source):
parser = self.collector()
try:
for s in source:
parser.feed(s)
parser.close()
except:
#self.events = parser.events
raise
return parser.get_events()
def check_events(self, source, expected_events):
try:
events = self.get_events(source)
except:
#import sys
#print >>sys.stderr, pprint.pformat(self.events)
raise
if events != expected_events:
self.fail("received events did not match expected events\n"
"Expected:\n" + pprint.pformat(expected_events) +
"\nReceived:\n" + pprint.pformat(events))
def check_parse_error(self, source):
parser = EventCollector()
try:
parser.feed(source)
parser.close()
except sgmllib.SGMLParseError:
pass
else:
self.fail("expected SGMLParseError for %r\nReceived:\n%s"
% (source, pprint.pformat(parser.get_events())))
def test_doctype_decl_internal(self):
inside = """\
DOCTYPE html PUBLIC '-//W3C//DTD HTML 4.01//EN'
SYSTEM 'http://www.w3.org/TR/html401/strict.dtd' [
<!ELEMENT html - O EMPTY>
<!ATTLIST html
version CDATA #IMPLIED
profile CDATA 'DublinCore'>
<!NOTATION datatype SYSTEM 'http://xml.python.org/notations/python-module'>
<!ENTITY myEntity 'internal parsed entity'>
<!ENTITY anEntity SYSTEM 'http://xml.python.org/entities/something.xml'>
<!ENTITY % paramEntity 'name|name|name'>
%paramEntity;
<!-- comment -->
]"""
self.check_events(["<!%s>" % inside], [
("decl", inside),
])
def test_doctype_decl_external(self):
inside = "DOCTYPE html PUBLIC '-//W3C//DTD HTML 4.01//EN'"
self.check_events("<!%s>" % inside, [
("decl", inside),
])
def test_underscore_in_attrname(self):
# SF bug #436621
"""Make sure attribute names with underscores are accepted"""
self.check_events("<a has_under _under>", [
("starttag", "a", [("has_under", "has_under"),
("_under", "_under")]),
])
def test_underscore_in_tagname(self):
# SF bug #436621
"""Make sure tag names with underscores are accepted"""
self.check_events("<has_under></has_under>", [
("starttag", "has_under", []),
("endtag", "has_under"),
])
def test_quotes_in_unquoted_attrs(self):
# SF bug #436621
"""Be sure quotes in unquoted attributes are made part of the value"""
self.check_events("<a href=foo'bar\"baz>", [
("starttag", "a", [("href", "foo'bar\"baz")]),
])
def test_xhtml_empty_tag(self):
"""Handling of XHTML-style empty start tags"""
self.check_events("<br />text<i></i>", [
("starttag", "br", []),
("data", "text"),
("starttag", "i", []),
("endtag", "i"),
])
def test_processing_instruction_only(self):
self.check_events("<?processing instruction>", [
("pi", "processing instruction"),
])
def test_bad_nesting(self):
self.check_events("<a><b></a></b>", [
("starttag", "a", []),
("starttag", "b", []),
("endtag", "a"),
("endtag", "b"),
])
def test_bare_ampersands(self):
self.check_events("this text & contains & ampersands &", [
("data", "this text & contains & ampersands &"),
])
def test_bare_pointy_brackets(self):
self.check_events("this < text > contains < bare>pointy< brackets", [
("data", "this < text > contains < bare>pointy< brackets"),
])
def test_attr_syntax(self):
output = [
("starttag", "a", [("b", "v"), ("c", "v"), ("d", "v"), ("e", "e")])
]
self.check_events("""<a b='v' c="v" d=v e>""", output)
self.check_events("""<a b = 'v' c = "v" d = v e>""", output)
self.check_events("""<a\nb\n=\n'v'\nc\n=\n"v"\nd\n=\nv\ne>""", output)
self.check_events("""<a\tb\t=\t'v'\tc\t=\t"v"\td\t=\tv\te>""", output)
def test_attr_values(self):
self.check_events("""<a b='xxx\n\txxx' c="yyy\t\nyyy" d='\txyz\n'>""",
[("starttag", "a", [("b", "xxx\n\txxx"),
("c", "yyy\t\nyyy"),
("d", "\txyz\n")])
])
self.check_events("""<a b='' c="">""", [
("starttag", "a", [("b", ""), ("c", "")]),
])
# URL construction stuff from RFC 1808:
safe = "$-_.+"
extra = "!*'(),"
reserved = ";/?:@&="
url = "http://example.com:8080/path/to/file?%s%s%s" % (
safe, extra, reserved)
self.check_events("""<e a=%s>""" % url, [
("starttag", "e", [("a", url)]),
])
# Regression test for SF patch #669683.
self.check_events("<e a=rgb(1,2,3)>", [
("starttag", "e", [("a", "rgb(1,2,3)")]),
])
def test_attr_values_entities(self):
"""Substitution of entities and charrefs in attribute values"""
# SF bug #1452246
self.check_events("""<a b=< c=<> d=<-> e='< '
f="&xxx;" g=' !' h='Ǵ'
i='x?a=b&c=d;'
j='&#42;' k='&#42;'>""",
[("starttag", "a", [("b", "<"),
("c", "<>"),
("d", "<->"),
("e", "< "),
("f", "&xxx;"),
("g", " !"),
("h", "Ǵ"),
("i", "x?a=b&c=d;"),
("j", "*"),
("k", "*"),
])])
def test_convert_overrides(self):
# This checks that the character and entity reference
# conversion helpers are called at the documented times. No
# attempt is made to really change what the parser accepts.
#
self.collector = HTMLEntityCollector
self.check_events(('<a title="“test”">foo</a>'
'&foobar;*'), [
('entityref', 'convert', 'ldquo'),
('charref', 'convert', 'x201d'),
('starttag', 'a', [('title', '“test”')]),
('data', 'foo'),
('endtag', 'a'),
('entityref', 'foobar'),
('entityref', 'convert', 'foobar'),
('charref', '42'),
('charref', 'convert', '42'),
('codepoint', 'convert', 42),
])
def test_attr_funky_names(self):
self.check_events("""<a a.b='v' c:d=v e-f=v>""", [
("starttag", "a", [("a.b", "v"), ("c:d", "v"), ("e-f", "v")]),
])
def test_attr_value_ip6_url(self):
# http://www.python.org/sf/853506
self.check_events(("<a href='http://[1080::8:800:200C:417A]/'>"
"<a href=http://[1080::8:800:200C:417A]/>"), [
("starttag", "a", [("href", "http://[1080::8:800:200C:417A]/")]),
("starttag", "a", [("href", "http://[1080::8:800:200C:417A]/")]),
])
def test_weird_starttags(self):
self.check_events("<a<a>", [
("starttag", "a", []),
("starttag", "a", []),
])
self.check_events("</a<a>", [
("endtag", "a"),
("starttag", "a", []),
])
def test_declaration_junk_chars(self):
self.check_parse_error("<!DOCTYPE foo $ >")
def test_get_starttag_text(self):
s = """<foobar \n one="1"\ttwo=2 >"""
self.check_events(s, [
("starttag", "foobar", [("one", "1"), ("two", "2")]),
])
def test_cdata_content(self):
s = ("<cdata> <!-- not a comment --> ¬-an-entity-ref; </cdata>"
"<notcdata> <!-- comment --> </notcdata>")
self.collector = CDATAEventCollector
self.check_events(s, [
("starttag", "cdata", []),
("data", " <!-- not a comment --> ¬-an-entity-ref; "),
("endtag", "cdata"),
("starttag", "notcdata", []),
("data", " "),
("comment", " comment "),
("data", " "),
("endtag", "notcdata"),
])
s = """<cdata> <not a='start tag'> </cdata>"""
self.check_events(s, [
("starttag", "cdata", []),
("data", " <not a='start tag'> "),
("endtag", "cdata"),
])
def test_illegal_declarations(self):
s = 'abc<!spacer type="block" height="25">def'
self.check_events(s, [
("data", "abc"),
("unknown decl", 'spacer type="block" height="25"'),
("data", "def"),
])
def test_enumerated_attr_type(self):
s = "<!DOCTYPE doc [<!ATTLIST doc attr (a | b) >]>"
self.check_events(s, [
('decl', 'DOCTYPE doc [<!ATTLIST doc attr (a | b) >]'),
])
def test_read_chunks(self):
# SF bug #1541697, this caused sgml parser to hang
# Just verify this code doesn't cause a hang.
CHUNK = 1024 # increasing this to 8212 makes the problem go away
f = open(test_support.findfile('sgml_input.html'))
fp = sgmllib.SGMLParser()
while 1:
data = f.read(CHUNK)
fp.feed(data)
if len(data) != CHUNK:
break
def test_only_decode_ascii(self):
# SF bug #1651995, make sure non-ascii character references are not decoded
s = '<signs exclamation="!" copyright="©" quoteleft="‘">'
self.check_events(s, [
('starttag', 'signs',
[('exclamation', '!'), ('copyright', '©'),
('quoteleft', '‘')]),
])
# XXX These tests have been disabled by prefixing their names with
# an underscore. The first two exercise outstanding bugs in the
# sgmllib module, and the third exhibits questionable behavior
# that needs to be carefully considered before changing it.
def _test_starttag_end_boundary(self):
self.check_events("<a b='<'>", [("starttag", "a", [("b", "<")])])
self.check_events("<a b='>'>", [("starttag", "a", [("b", ">")])])
def _test_buffer_artefacts(self):
output = [("starttag", "a", [("b", "<")])]
self.check_events(["<a b='<'>"], output)
self.check_events(["<a ", "b='<'>"], output)
self.check_events(["<a b", "='<'>"], output)
self.check_events(["<a b=", "'<'>"], output)
self.check_events(["<a b='<", "'>"], output)
self.check_events(["<a b='<'", ">"], output)
output = [("starttag", "a", [("b", ">")])]
self.check_events(["<a b='>'>"], output)
self.check_events(["<a ", "b='>'>"], output)
self.check_events(["<a b", "='>'>"], output)
self.check_events(["<a b=", "'>'>"], output)
self.check_events(["<a b='>", "'>"], output)
self.check_events(["<a b='>'", ">"], output)
output = [("comment", "abc")]
self.check_events(["", "<!--abc-->"], output)
self.check_events(["<", "!--abc-->"], output)
self.check_events(["<!", "--abc-->"], output)
self.check_events(["<!-", "-abc-->"], output)
self.check_events(["<!--", "abc-->"], output)
self.check_events(["<!--a", "bc-->"], output)
self.check_events(["<!--ab", "c-->"], output)
self.check_events(["<!--abc", "-->"], output)
self.check_events(["<!--abc-", "->"], output)
self.check_events(["<!--abc--", ">"], output)
self.check_events(["<!--abc-->", ""], output)
def _test_starttag_junk_chars(self):
self.check_parse_error("<")
self.check_parse_error("<>")
self.check_parse_error("</$>")
self.check_parse_error("</")
self.check_parse_error("</a")
self.check_parse_error("<$")
self.check_parse_error("<$>")
self.check_parse_error("<!")
self.check_parse_error("<a $>")
self.check_parse_error("<a")
self.check_parse_error("<a foo='bar'")
self.check_parse_error("<a foo='bar")
self.check_parse_error("<a foo='>'")
self.check_parse_error("<a foo='>")
self.check_parse_error("<a foo=>")
def test_main():
test_support.run_unittest(SGMLParserTestCase)
if __name__ == "__main__":
test_main()
您还没有登录,登录后您可以:
-
收藏Android系统代码
-
收藏喜欢的文章
-
多个平台共享账号
去登录
首次使用?从这里 注册