cpython/Lib/xmllib.py

753 lines
27 KiB
Python

# A parser for XML, using the derived class as static DTD.
# Author: Sjoerd Mullender.
import re
import string
version = '0.1'
# Regular expressions used for parsing
_S = '[ \t\r\n]+' # white space
_opS = '[ \t\r\n]*' # optional white space
_Name = '[a-zA-Z_:][-a-zA-Z0-9._:]*' # valid XML name
_QStr = "(?:'[^']*'|\"[^\"]*\")" # quoted XML string
illegal = re.compile('[^\t\r\n -\176\240-\377]') # illegal chars in content
interesting = re.compile('[]&<]')
amp = re.compile('&')
ref = re.compile('&(' + _Name + '|#[0-9]+|#x[0-9a-fA-F]+)[^-a-zA-Z0-9._:]')
entityref = re.compile('&(?P<name>' + _Name + ')[^-a-zA-Z0-9._:]')
charref = re.compile('&#(?P<char>[0-9]+[^0-9]|x[0-9a-fA-F]+[^0-9a-fA-F])')
space = re.compile(_S + '$')
newline = re.compile('\n')
attrfind = re.compile(
_S + '(?P<name>' + _Name + ')'
'(' + _opS + '=' + _opS +
'(?P<value>'+_QStr+'|[-a-zA-Z0-9.:+*%?!()_#=~]+))?')
starttagopen = re.compile('<' + _Name)
starttagend = re.compile(_opS + '(?P<slash>/?)>')
starttagmatch = re.compile('<(?P<tagname>'+_Name+')'
'(?P<attrs>(?:'+attrfind.pattern+')*)'+
starttagend.pattern)
endtagopen = re.compile('</')
endbracket = re.compile(_opS + '>')
endbracketfind = re.compile('(?:[^>\'"]|'+_QStr+')*>')
tagfind = re.compile(_Name)
cdataopen = re.compile(r'<!\[CDATA\[')
cdataclose = re.compile(r'\]\]>')
# this matches one of the following:
# SYSTEM SystemLiteral
# PUBLIC PubidLiteral SystemLiteral
_SystemLiteral = '(?P<%s>'+_QStr+')'
_PublicLiteral = '(?P<%s>"[-\'()+,./:=?;!*#@$_%% \n\ra-zA-Z0-9]*"|' \
"'[-()+,./:=?;!*#@$_%% \n\ra-zA-Z0-9]*')"
_ExternalId = '(?:SYSTEM|' \
'PUBLIC'+_S+_PublicLiteral%'pubid'+ \
')'+_S+_SystemLiteral%'syslit'
doctype = re.compile('<!DOCTYPE'+_S+'(?P<name>'+_Name+')'
'(?:'+_S+_ExternalId+')?'+_opS)
xmldecl = re.compile('<\?xml'+_S+
'version'+_opS+'='+_opS+'(?P<version>'+_QStr+')'+
'(?:'+_S+'encoding'+_opS+'='+_opS+
"(?P<encoding>'[A-Za-z][-A-Za-z0-9._]*'|"
'"[A-Za-z][-A-Za-z0-9._]*"))?'
'(?:'+_S+'standalone'+_opS+'='+_opS+
'(?P<standalone>\'(?:yes|no)\'|"(?:yes|no)"))?'+
_opS+'\?>')
procopen = re.compile(r'<\?(?P<proc>' + _Name + ')' + _opS)
procclose = re.compile(_opS + r'\?>')
commentopen = re.compile('<!--')
commentclose = re.compile('-->')
doubledash = re.compile('--')
attrtrans = string.maketrans(' \r\n\t', ' ')
# XML parser base class -- find tags and call handler functions.
# Usage: p = XMLParser(); p.feed(data); ...; p.close().
# The dtd is defined by deriving a class which defines methods with
# special names to handle tags: start_foo and end_foo to handle <foo>
# and </foo>, respectively. The data between tags is passed to the
# parser by calling self.handle_data() with some data as argument (the
# data may be split up in arbutrary chunks). Entity references are
# passed by calling self.handle_entityref() with the entity reference
# as argument.
class XMLParser:
# Interface -- initialize and reset this instance
def __init__(self, verbose=0):
self.verbose = verbose
self.reset()
# Interface -- reset this instance. Loses all unprocessed data
def reset(self):
self.rawdata = ''
self.stack = []
self.nomoretags = 0
self.literal = 0
self.lineno = 1
self.__at_start = 1
self.__seen_doctype = None
self.__seen_starttag = 0
# For derived classes only -- enter literal mode (CDATA) till EOF
def setnomoretags(self):
self.nomoretags = self.literal = 1
# For derived classes only -- enter literal mode (CDATA)
def setliteral(self, *args):
self.literal = 1
# Interface -- feed some data to the parser. Call this as
# often as you want, with as little or as much text as you
# want (may include '\n'). (This just saves the text, all the
# processing is done by goahead().)
def feed(self, data):
self.rawdata = self.rawdata + data
self.goahead(0)
# Interface -- handle the remaining data
def close(self):
self.goahead(1)
# Interface -- translate references
def translate_references(self, data, all = 1):
i = 0
while 1:
res = amp.search(data, i)
if res is None:
return data
res = ref.match(data, res.start(0))
if res is None:
self.syntax_error("bogus `&'")
i =i+1
continue
i = res.end(0)
if data[i - 1] != ';':
self.syntax_error("`;' missing after entity/char reference")
i = i-1
str = res.group(1)
pre = data[:res.start(0)]
post = data[i:]
if str[0] == '#':
if str[1] == 'x':
str = chr(string.atoi(str[2:], 16))
else:
str = chr(string.atoi(str[1:]))
data = pre + str + post
i = res.start(0)+len(str)
elif all:
if self.entitydefs.has_key(str):
data = pre + self.entitydefs[str] + post
i = res.start(0) # rescan substituted text
else:
self.syntax_error('reference to unknown entity')
# can't do it, so keep the entity ref in
data = pre + '&' + str + ';' + post
i = res.start(0) + len(str) + 2
else:
# just translating character references
pass # i is already postioned correctly
# Internal -- handle data as far as reasonable. May leave state
# and data to be processed by a subsequent call. If 'end' is
# true, force handling all data as if followed by EOF marker.
def goahead(self, end):
rawdata = self.rawdata
i = 0
n = len(rawdata)
while i < n:
if i > 0:
self.__at_start = 0
if self.nomoretags:
data = rawdata[i:n]
self.handle_data(data)
self.lineno = self.lineno + string.count(data, '\n')
i = n
break
res = interesting.search(rawdata, i)
if res:
j = res.start(0)
else:
j = n
if i < j:
if self.__at_start:
self.syntax_error('illegal data at start of file')
self.__at_start = 0
data = rawdata[i:j]
if not self.stack and space.match(data) is None:
self.syntax_error('data not in content')
if illegal.search(data):
self.syntax_error('illegal character in content')
self.handle_data(data)
self.lineno = self.lineno + string.count(data, '\n')
i = j
if i == n: break
if rawdata[i] == '<':
if starttagopen.match(rawdata, i):
if self.literal:
data = rawdata[i]
self.handle_data(data)
self.lineno = self.lineno + string.count(data, '\n')
i = i+1
continue
k = self.parse_starttag(i)
if k < 0: break
self.__seen_starttag = 1
self.lineno = self.lineno + string.count(rawdata[i:k], '\n')
i = k
continue
if endtagopen.match(rawdata, i):
k = self.parse_endtag(i)
if k < 0: break
self.lineno = self.lineno + string.count(rawdata[i:k], '\n')
i = k
self.literal = 0
continue
if commentopen.match(rawdata, i):
if self.literal:
data = rawdata[i]
self.handle_data(data)
self.lineno = self.lineno + string.count(data, '\n')
i = i+1
continue
k = self.parse_comment(i)
if k < 0: break
self.lineno = self.lineno + string.count(rawdata[i:k], '\n')
i = k
continue
if cdataopen.match(rawdata, i):
k = self.parse_cdata(i)
if k < 0: break
self.lineno = self.lineno + string.count(rawdata[i:i], '\n')
i = k
continue
res = xmldecl.match(rawdata, i)
if res:
if not self.__at_start:
self.syntax_error("<?xml?> declaration not at start of document")
version, encoding, standalone = res.group('version',
'encoding',
'standalone')
if version[1:-1] != '1.0':
raise RuntimeError, 'only XML version 1.0 supported'
if encoding: encoding = encoding[1:-1]
if standalone: standalone = standalone[1:-1]
self.handle_xml(encoding, standalone)
i = res.end(0)
continue
res = procopen.match(rawdata, i)
if res:
k = self.parse_proc(i)
if k < 0: break
self.lineno = self.lineno + string.count(rawdata[i:k], '\n')
i = k
continue
res = doctype.match(rawdata, i)
if res:
if self.literal:
data = rawdata[i]
self.handle_data(data)
self.lineno = self.lineno + string.count(data, '\n')
i = i+1
continue
if self.__seen_doctype:
self.syntax_error('multiple DOCTYPE elements')
if self.__seen_starttag:
self.syntax_error('DOCTYPE not at beginning of document')
k = self.parse_doctype(res)
if k < 0: break
self.__seen_doctype = res.group('name')
self.lineno = self.lineno + string.count(rawdata[i:k], '\n')
i = k
continue
elif rawdata[i] == '&':
res = charref.match(rawdata, i)
if res is not None:
i = res.end(0)
if rawdata[i-1] != ';':
self.syntax_error("`;' missing in charref")
i = i-1
if not self.stack:
self.syntax_error('data not in content')
self.handle_charref(res.group('char')[:-1])
self.lineno = self.lineno + string.count(res.group(0), '\n')
continue
res = entityref.match(rawdata, i)
if res is not None:
i = res.end(0)
if rawdata[i-1] != ';':
self.syntax_error("`;' missing in entityref")
i = i-1
name = res.group('name')
if self.entitydefs.has_key(name):
self.rawdata = rawdata = rawdata[:res.start(0)] + self.entitydefs[name] + rawdata[i:]
n = len(rawdata)
i = res.start(0)
else:
self.syntax_error('reference to unknown entity')
self.unknown_entityref(name)
self.lineno = self.lineno + string.count(res.group(0), '\n')
continue
elif rawdata[i] == ']':
if n-i < 3:
break
if cdataclose.match(rawdata, i):
self.syntax_error("bogus `]]>'")
self.handle_data(rawdata[i])
i = i+1
continue
else:
raise RuntimeError, 'neither < nor & ??'
# We get here only if incomplete matches but
# nothing else
break
# end while
if i > 0:
self.__at_start = 0
if end and i < n:
data = rawdata[i]
self.syntax_error("bogus `%s'" % data)
if illegal.search(data):
self.syntax_error('illegal character in content')
self.handle_data(data)
self.lineno = self.lineno + string.count(data, '\n')
self.rawdata = rawdata[i+1:]
return self.goahead(end)
self.rawdata = rawdata[i:]
if end:
if not self.__seen_starttag:
self.syntax_error('no elements in file')
if self.stack:
self.syntax_error('missing end tags')
while self.stack:
self.finish_endtag(self.stack[-1])
# Internal -- parse comment, return length or -1 if not terminated
def parse_comment(self, i):
rawdata = self.rawdata
if rawdata[i:i+4] <> '<!--':
raise RuntimeError, 'unexpected call to handle_comment'
res = commentclose.search(rawdata, i+4)
if res is None:
return -1
if doubledash.search(rawdata, i+4, res.start(0)):
self.syntax_error("`--' inside comment")
if rawdata[res.start(0)-1] == '-':
self.syntax_error('comment cannot end in three dashes')
if illegal.search(rawdata, i+4, res.start(0)):
self.syntax_error('illegal character in comment')
self.handle_comment(rawdata[i+4: res.start(0)])
return res.end(0)
# Internal -- handle DOCTYPE tag, return length or -1 if not terminated
def parse_doctype(self, res):
rawdata = self.rawdata
n = len(rawdata)
name = res.group('name')
pubid, syslit = res.group('pubid', 'syslit')
if pubid is not None:
pubid = pubid[1:-1] # remove quotes
pubid = string.join(string.split(pubid)) # normalize
if syslit is not None: syslit = syslit[1:-1] # remove quotes
j = k = res.end(0)
if k >= n:
return -1
if rawdata[k] == '[':
level = 0
k = k+1
dq = sq = 0
while k < n:
c = rawdata[k]
if not sq and c == '"':
dq = not dq
elif not dq and c == "'":
sq = not sq
elif sq or dq:
pass
elif level <= 0 and c == ']':
res = endbracket.match(rawdata, k+1)
if res is None:
return -1
self.handle_doctype(name, pubid, syslit, rawdata[j+1:k])
return res.end(0)
elif c == '<':
level = level + 1
elif c == '>':
level = level - 1
if level < 0:
self.syntax_error("bogus `>' in DOCTYPE")
k = k+1
res = endbracketfind.match(rawdata, k)
if res is None:
return -1
if endbracket.match(rawdata, k) is None:
self.syntax_error('garbage in DOCTYPE')
self.handle_doctype(name, pubid, syslit, None)
return res.end(0)
# Internal -- handle CDATA tag, return length or -1 if not terminated
def parse_cdata(self, i):
rawdata = self.rawdata
if rawdata[i:i+9] <> '<![CDATA[':
raise RuntimeError, 'unexpected call to parse_cdata'
res = cdataclose.search(rawdata, i+9)
if res is None:
return -1
if illegal.search(rawdata, i+9, res.start(0)):
self.syntax_error('illegal character in CDATA')
if not self.stack:
self.syntax_error('CDATA not in content')
self.handle_cdata(rawdata[i+9:res.start(0)])
return res.end(0)
__xml_attributes = {'version': '1.0', 'standalone': 'no', 'encoding': None}
# Internal -- handle a processing instruction tag
def parse_proc(self, i):
rawdata = self.rawdata
end = procclose.search(rawdata, i)
if end is None:
return -1
j = end.start(0)
if illegal.search(rawdata, i+2, j):
self.syntax_error('illegal character in processing instruction')
res = tagfind.match(rawdata, i+2)
if res is None:
raise RuntimeError, 'unexpected call to parse_proc'
k = res.end(0)
name = res.group(0)
if string.find(string.lower(name), 'xml') >= 0:
self.syntax_error('illegal processing instruction target name')
self.handle_proc(name, rawdata[k:j])
return end.end(0)
# Internal -- parse attributes between i and j
def parse_attributes(self, tag, i, j, attributes = None):
rawdata = self.rawdata
# Now parse the data between i and j into a tag and attrs
attrdict = {}
try:
# convert attributes list to dictionary
d = {}
for a in attributes:
d[a] = None
attributes = d
except TypeError:
pass
while i < j:
res = attrfind.match(rawdata, i)
if res is None:
break
attrname, attrvalue = res.group('name', 'value')
if attrvalue is None:
self.syntax_error("no value specified for attribute `%s'" % attrname)
attrvalue = attrname
elif attrvalue[:1] == "'" == attrvalue[-1:] or \
attrvalue[:1] == '"' == attrvalue[-1:]:
attrvalue = attrvalue[1:-1]
else:
self.syntax_error("attribute `%s' value not quoted" % attrname)
if '<' in attrvalue:
self.syntax_error("`<' illegal in attribute value")
if attributes is not None and not attributes.has_key(attrname):
self.syntax_error("unknown attribute `%s' of element `%s'" %
(attrname, tag))
if attrdict.has_key(attrname):
self.syntax_error("attribute `%s' specified twice" % attrname)
attrvalue = string.translate(attrvalue, attrtrans)
attrdict[attrname] = self.translate_references(attrvalue)
i = res.end(0)
if attributes is not None:
# fill in with default attributes
for key, val in attributes.items():
if val is not None and not attrdict.has_key(key):
attrdict[key] = val
return attrdict, i
# Internal -- handle starttag, return length or -1 if not terminated
def parse_starttag(self, i):
rawdata = self.rawdata
# i points to start of tag
end = endbracketfind.match(rawdata, i+1)
if end is None:
return -1
tag = starttagmatch.match(rawdata, i)
if tag is None or tag.end(0) != end.end(0):
self.syntax_error('garbage in starttag')
return end.end(0)
tagname = tag.group('tagname')
if not self.__seen_starttag and self.__seen_doctype and \
tagname != self.__seen_doctype:
self.syntax_error('starttag does not match DOCTYPE')
if self.__seen_starttag and not self.stack:
self.syntax_error('multiple elements on top level')
if hasattr(self, tagname + '_attributes'):
attributes = getattr(self, tagname + '_attributes')
else:
attributes = None
k, j = tag.span('attrs')
attrdict, k = self.parse_attributes(tagname, k, j, attributes)
self.finish_starttag(tagname, attrdict)
if tag.group('slash') == '/':
self.finish_endtag(tagname)
return tag.end(0)
# Internal -- parse endtag
def parse_endtag(self, i):
rawdata = self.rawdata
end = endbracketfind.match(rawdata, i+1)
if end is None:
return -1
res = tagfind.match(rawdata, i+2)
if res is None:
self.syntax_error('no name specified in end tag')
tag = ''
k = i+2
else:
tag = res.group(0)
k = res.end(0)
if endbracket.match(rawdata, k) is None:
self.syntax_error('garbage in end tag')
self.finish_endtag(tag)
return end.end(0)
# Internal -- finish processing of start tag
# Return -1 for unknown tag, 1 for balanced tag
def finish_starttag(self, tag, attrs):
self.stack.append(tag)
methodname = 'start_' + tag
if hasattr(self, methodname):
method = getattr(self, methodname)
self.handle_starttag(tag, method, attrs)
return 1
else:
self.unknown_starttag(tag, attrs)
return -1
# Internal -- finish processing of end tag
def finish_endtag(self, tag):
methodname = 'end_' + tag
if not tag:
self.syntax_error('name-less end tag')
found = len(self.stack) - 1
if found < 0:
self.unknown_endtag(tag)
return
else:
if tag not in self.stack:
self.syntax_error('unopened end tag')
if hasattr(self, methodname):
method = getattr(self, methodname)
self.handle_endtag(tag, method)
else:
self.unknown_endtag(tag)
return
found = len(self.stack)
for i in range(found):
if self.stack[i] == tag:
found = i
while len(self.stack) > found:
if found < len(self.stack) - 1:
self.syntax_error('missing close tag for %s' % self.stack[-1])
tag = self.stack[-1]
if hasattr(self, methodname):
method = getattr(self, methodname)
self.handle_endtag(tag, method)
else:
self.unknown_endtag(tag)
del self.stack[-1]
# Overridable -- handle xml processing instruction
def handle_xml(self, encoding, standalone):
pass
# Overridable -- handle DOCTYPE
def handle_doctype(self, tag, pubid, syslit, data):
pass
# Overridable -- handle start tag
def handle_starttag(self, tag, method, attrs):
method(attrs)
# Overridable -- handle end tag
def handle_endtag(self, tag, method):
method()
# Example -- handle character reference, no need to override
def handle_charref(self, name):
try:
if name[0] == 'x':
n = string.atoi(name[1:], 16)
else:
n = string.atoi(name)
except string.atoi_error:
self.unknown_charref(name)
return
if not 0 <= n <= 255:
self.unknown_charref(name)
return
self.handle_data(chr(n))
# Definition of entities -- derived classes may override
entitydefs = {'lt': '&#60;', # must use charref
'gt': '&#62;',
'amp': '&#38;', # must use charref
'quot': '&#34;',
'apos': '&#39;',
}
# Example -- handle entity reference, no need to override
def handle_entityref(self, name):
table = self.entitydefs
if table.has_key(name):
self.handle_data(table[name])
else:
self.unknown_entityref(name)
return
# Example -- handle data, should be overridden
def handle_data(self, data):
pass
# Example -- handle cdata, could be overridden
def handle_cdata(self, data):
pass
# Example -- handle comment, could be overridden
def handle_comment(self, data):
pass
# Example -- handle processing instructions, could be overridden
def handle_proc(self, name, data):
pass
# Example -- handle relatively harmless syntax errors, could be overridden
def syntax_error(self, message):
raise RuntimeError, 'Syntax error at line %d: %s' % (self.lineno, message)
# To be overridden -- handlers for unknown objects
def unknown_starttag(self, tag, attrs): pass
def unknown_endtag(self, tag): pass
def unknown_charref(self, ref): pass
def unknown_entityref(self, ref): pass
class TestXMLParser(XMLParser):
def __init__(self, verbose=0):
self.testdata = ""
XMLParser.__init__(self, verbose)
def handle_xml(self, encoding, standalone):
self.flush()
print 'xml: encoding =',encoding,'standalone =',standalone
def handle_doctype(self, tag, pubid, syslit, data):
self.flush()
print 'DOCTYPE:',tag, `data`
def handle_entity(self, name, strval, pubid, syslit, ndata):
self.flush()
print 'ENTITY:',`data`
def handle_data(self, data):
self.testdata = self.testdata + data
if len(`self.testdata`) >= 70:
self.flush()
def flush(self):
data = self.testdata
if data:
self.testdata = ""
print 'data:', `data`
def handle_cdata(self, data):
self.flush()
print 'cdata:', `data`
def handle_proc(self, name, data):
self.flush()
print 'processing:',name,`data`
def handle_comment(self, data):
self.flush()
r = `data`
if len(r) > 68:
r = r[:32] + '...' + r[-32:]
print 'comment:', r
def syntax_error(self, message):
print 'error at line %d:' % self.lineno, message
def unknown_starttag(self, tag, attrs):
self.flush()
if not attrs:
print 'start tag: <' + tag + '>'
else:
print 'start tag: <' + tag,
for name, value in attrs.items():
print name + '=' + '"' + value + '"',
print '>'
def unknown_endtag(self, tag):
self.flush()
print 'end tag: </' + tag + '>'
def unknown_entityref(self, ref):
self.flush()
print '*** unknown entity ref: &' + ref + ';'
def unknown_charref(self, ref):
self.flush()
print '*** unknown char ref: &#' + ref + ';'
def close(self):
XMLParser.close(self)
self.flush()
def test(args = None):
import sys
if not args:
args = sys.argv[1:]
if args and args[0] == '-s':
args = args[1:]
klass = XMLParser
else:
klass = TestXMLParser
if args:
file = args[0]
else:
file = 'test.xml'
if file == '-':
f = sys.stdin
else:
try:
f = open(file, 'r')
except IOError, msg:
print file, ":", msg
sys.exit(1)
data = f.read()
if f is not sys.stdin:
f.close()
x = klass()
try:
for c in data:
x.feed(c)
x.close()
except RuntimeError, msg:
print msg
sys.exit(1)
if __name__ == '__main__':
test()