############################################################################## # # Copyright (c) 2002 Zope Corporation and Contributors. All Rights Reserved. # # This software is subject to the provisions of the Zope Public License, # Version 2.1 (ZPL). A copy of the ZPL should accompany this distribution. # THIS SOFTWARE IS PROVIDED "AS IS" AND ANY AND ALL EXPRESS OR IMPLIED # WARRANTIES ARE DISCLAIMED, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED # WARRANTIES OF TITLE, MERCHANTABILITY, AGAINST INFRINGEMENT, AND FITNESS # FOR A PARTICULAR PURPOSE # ############################################################################## """DTML Utilities $Id: DT_Util.py 40320 2005-11-22 16:08:56Z tseaver $""" import re from html_quote import html_quote, ustr # for import by other modules, dont remove! from RestrictedPython.Guards import safe_builtins from RestrictedPython.Utilities import utility_builtins from RestrictedPython.Eval import RestrictionCapableEval from cDocumentTemplate import InstanceDict, TemplateDict, \ render_blocks, safe_callable, join_unicode test = utility_builtins['test'] # for backwards compatibility, dont remove! LIMITED_BUILTINS = 1 str=__builtins__['str'] # Waaaaa, waaaaaaaa needed for pickling waaaaa ParseError='Document Template Parse Error' from zExceptions import Unauthorized as ValidationError def int_param(params,md,name,default=0, st=type('')): v = params.get(name, default) if v: try: v = int(v) except: v = md[v] if isinstance(v, str): v = int(v) return v or 0 functype = type(int_param) class NotBindable: # Used to prevent TemplateDict from trying to bind to functions. def __init__(self, f): self.__call__ = f for name, f in safe_builtins.items() + utility_builtins.items(): if type(f) is functype: f = NotBindable(f) setattr(TemplateDict, name, f) if LIMITED_BUILTINS: # Replace certain builtins with limited versions. from RestrictedPython.Limits import limited_builtins for name, f in limited_builtins.items(): if type(f) is functype: f = NotBindable(f) setattr(TemplateDict, name, f) try: # Wrap the string module so it can deal with TaintedString strings. from ZPublisher.TaintedString import TaintedString from types import FunctionType, BuiltinFunctionType, StringType import string class StringModuleWrapper: def __getattr__(self, key): attr = getattr(string, key) if (isinstance(attr, FunctionType) or isinstance(attr, BuiltinFunctionType)): return StringFunctionWrapper(attr) else: return attr class StringFunctionWrapper: def __init__(self, method): self._method = method def __call__(self, *args, **kw): tainted = 0 args = list(args) for i in range(len(args)): if isinstance(args[i], TaintedString): tainted = 1 args[i] = str(args[i]) for k, v in kw.items(): if isinstance(v, TaintedString): tainted = 1 kw[k] = str(v) args = tuple(args) retval = self._method(*args, **kw) if tainted and isinstance(retval, StringType) and '<' in retval: retval = TaintedString(retval) return retval TemplateDict.string = StringModuleWrapper() except ImportError: # Use the string module already defined in RestrictedPython.Utilities pass # The functions below are meant to bind to the TemplateDict. _marker = [] # Create a new marker object. def careful_getattr(md, inst, name, default=_marker): get = md.guarded_getattr if get is None: get = getattr try: return get(inst, name) except AttributeError: if default is _marker: raise return default def careful_hasattr(md, inst, name): get = md.guarded_getattr if get is None: get = getattr try: get(inst, name) except (AttributeError, ValidationError, KeyError): return 0 else: return 1 TemplateDict.getattr = careful_getattr TemplateDict.hasattr = careful_hasattr def namespace(self, **kw): """Create a tuple consisting of a single instance whose attributes are provided as keyword arguments.""" if not (getattr(self, '__class__', None) == TemplateDict or isinstance(self, TemplateDict)): raise TypeError,'''A call was made to DT_Util.namespace() with an incorrect "self" argument. It could be caused by a product which is not yet compatible with this version of Zope. The traceback information may contain more details.)''' return self(**kw) TemplateDict.namespace = namespace def render(self, v): "Render an object in the way done by the 'name' attribute" if hasattr(v, '__render_with_namespace__'): v = v.__render_with_namespace__(self) else: vbase = getattr(v, 'aq_base', v) if safe_callable(vbase): if getattr(vbase, 'isDocTemp', 0): v = v(None, self) else: v = v() return v TemplateDict.render = render class Eval(RestrictionCapableEval): def eval(self, md): gattr = getattr(md, 'guarded_getattr', None) if gattr is not None: gitem = getattr(md, 'guarded_getitem', None) self.prepRestrictedCode() code = self.rcode d = {'_': md, '_vars': md, '_getattr_': gattr, '_getitem_': gitem, '__builtins__': None} else: self.prepUnrestrictedCode() code = self.ucode d = {'_': md, '_vars': md} d.update(self.globals) has_key = d.has_key for name in self.used: __traceback_info__ = name try: if not has_key(name): d[name] = md.getitem(name, 0) except KeyError: # Swallow KeyErrors since the expression # might not actually need the name. If it # does need the name, a NameError will occur. pass return eval(code, d) def __call__(self, **kw): # Never used? md = TemplateDict() md._push(kw) return self.eval(md) simple_name = re.compile('[a-z][a-z0-9_]*', re.I).match class Add_with_prefix: def __init__(self, map, defprefix, prefix): self.map = map self.defprefix = defprefix self.prefix = prefix def __setitem__(self, name, value): map = self.map map[name] = value dp = self.defprefix if name.startswith(dp + '-'): map[self.prefix + name[len(dp):].replace('-', '_')] = value else: map['%s_%s' % (self.prefix, name)] = value def add_with_prefix(map, defprefix, prefix): if not prefix: return map return Add_with_prefix(map, defprefix, prefix) def name_param(params,tag='',expr=0, attr='name', default_unnamed=1): used=params.has_key __traceback_info__=params, tag, expr, attr #if expr and used('expr') and used('') and not used(params['']): # # Fix up something like: # params[params['']]=default_unnamed # del params[''] if used(''): v=params[''] if v[:1]=='"' and v[-1:]=='"' and len(v) > 1: # expr shorthand if used(attr): raise ParseError, ('%s and expr given' % attr, tag) if expr: if used('expr'): raise ParseError, ('two exprs given', tag) v=v[1:-1] try: expr=Eval(v) except SyntaxError, v: raise ParseError, ( 'Expression (Python) Syntax error:' '\n
\n%s\n\n' % v[0], tag) return v, expr else: raise ParseError, ( 'The "..." shorthand for expr was used in a tag ' 'that doesn\'t support expr attributes.', tag) else: # name shorthand if used(attr): raise ParseError, ('Two %s values were given' % attr, tag) if expr: if used('expr'): # raise 'Waaaaaa', 'waaa' raise ParseError, ('%s and expr given' % attr, tag) return params[''],None return params[''] elif used(attr): if expr: if used('expr'): raise ParseError, ('%s and expr given' % attr, tag) return params[attr],None return params[attr] elif expr and used('expr'): name=params['expr'] expr=Eval(name) return name, expr raise ParseError, ('No %s given' % attr, tag) Expr_doc=""" Python expression support Several document template tags, including 'var', 'in', 'if', 'else', and 'elif' provide support for using Python expressions via an 'expr' tag attribute. Expressions may be used where a simple variable value is inadequate. For example, an expression might be used to test whether a variable is greater than some amount:: or to transform some basic data:: Objects available in the document templates namespace may be used. Subobjects of these objects may be used as well, although subobject access is restricted by the optional validation method. In addition, a special additional name, '_', is available. The '_' variable provides access to the document template namespace as a mapping object. This variable can be useful for accessing objects in a document template namespace that have names that are not legal Python variable names:: This variable also has attributes that provide access to standard utility objects. These attributes include: - The objects: 'None', 'abs', 'chr', 'divmod', 'float', 'hash', 'hex', 'int', 'len', 'max', 'min', 'oct', 'ord', 'pow', 'round', and 'str' from the standard Python builtin module. - Special security-aware versions of 'getattr' and 'hasattr', - The Python 'string', 'math', and 'random' modules, and - A special function, 'test', that supports if-then expressions. The 'test' function accepts any number of arguments. If the first argument is true, then the second argument is returned, otherwise if the third argument is true, then the fourth argument is returned, and so on. If there is an odd number of arguments, then the last argument is returned in the case that none of the tested arguments is true, otherwise None is returned. For example, to convert a value to lower case:: """ ListType=type([]) def parse_params(text, result=None, tag='', unparmre=re.compile('([\000- ]*([^\000- ="]+))'), qunparmre=re.compile('([\000- ]*("[^"]*"))'), parmre=re.compile('([\000- ]*([^\000- ="]+)=([^\000- ="]+))'), qparmre=re.compile('([\000- ]*([^\000- ="]+)="([^"]*)")'), **parms): """Parse tag parameters The format of tag parameters consists of 1 or more parameter specifications separated by whitespace. Each specification consists of an unnamed and unquoted value, a valueless name, or a name-value pair. A name-value pair consists of a name and a quoted or unquoted value separated by an '='. The input parameter, text, gives the text to be parsed. The keyword parameters give valid parameter names and default values. If a specification is not a name-value pair and it is not the first specification and it is a valid parameter name, then it is treated as a name-value pair with a value as given in the keyword argument. Otherwise, if it is not a name-value pair, it is treated as an unnamed value. The data are parsed into a dictionary mapping names to values. Unnamed values are mapped from the name '""'. Only one value may be given for a name and there may be only one unnamed value. """ result=result or {} # HACK - we precalculate all matches. Maybe we don't need them # all. This should be fixed for performance issues mo_p = parmre.match(text) mo_q = qparmre.match(text) mo_unp = unparmre.match(text) mo_unq = qunparmre.match(text) if mo_p: name=mo_p.group(2).lower() value=mo_p.group(3) l=len(mo_p.group(1)) elif mo_q: name=mo_q.group(2).lower() value=mo_q.group(3) l=len(mo_q.group(1)) elif mo_unp: name=mo_unp.group(2) l=len(mo_unp.group(1)) if result: if parms.has_key(name): if parms[name] is None: raise ParseError, ( 'Attribute %s requires a value' % name, tag) result[name]=parms[name] else: raise ParseError, ( 'Invalid attribute name, "%s"' % name, tag) else: result['']=name return parse_params(text[l:],result,**parms) elif mo_unq: name=mo_unq.group(2) l=len(mo_unq.group(1)) if result: raise ParseError, ( 'Invalid attribute name, "%s"' % name, tag) else: result['']=name return parse_params(text[l:],result,**parms) else: if not text or not text.strip(): return result raise ParseError, ('invalid parameter: "%s"' % text, tag) if not parms.has_key(name): raise ParseError, ( 'Invalid attribute name, "%s"' % name, tag) if result.has_key(name): p=parms[name] if type(p) is not ListType or p: raise ParseError, ( 'Duplicate values for attribute "%s"' % name, tag) result[name]=value text=text[l:].strip() if text: return parse_params(text,result,**parms) else: return result