# This program is free software; you can redistribute it and/or modify it under # the terms of the (LGPL) GNU Lesser General Public License as published by the # Free Software Foundation; either version 3 of the License, or (at your # option) any later version. # # This program is distributed in the hope that it will be useful, but WITHOUT # ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS # FOR A PARTICULAR PURPOSE. See the GNU Library Lesser General Public License # for more details at ( http://www.gnu.org/licenses/lgpl.html ). # # You should have received a copy of the GNU Lesser General Public License # along with this program; if not, write to the Free Software Foundation, Inc., # 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. # written by: Jeff Ortel ( jortel@redhat.com ) """I{Base} classes representing XSD schema objects.""" from suds import * from suds.xsd import * from suds.sax.element import Element from suds.sax import Namespace from logging import getLogger log = getLogger(__name__) class SchemaObject(UnicodeMixin): """ A schema object is an extension to object with schema awareness. @ivar root: The XML root element. @type root: L{Element} @ivar schema: The schema containing this object. @type schema: L{schema.Schema} @ivar form_qualified: A flag indicating that @elementFormDefault has a value of I{qualified}. @type form_qualified: boolean @ivar nillable: A flag indicating that @nillable has a value of I{true}. @type nillable: boolean @ivar default: The default value. @type default: object @ivar rawchildren: A list raw of all children. @type rawchildren: [L{SchemaObject},...] """ @classmethod def prepend(cls, d, s, filter=Filter()): """ Prepend B{s}ource XSD schema objects to the B{d}estination list. B{filter} is used to decide which objects to prepend and which to skip. @param d: The destination list. @type d: list @param s: The source list. @type s: list @param filter: A filter allowing items to be prepended. @type filter: L{Filter} """ i = 0 for x in s: if x in filter: d.insert(i, x) i += 1 @classmethod def append(cls, d, s, filter=Filter()): """ Append B{s}ource XSD schema objects to the B{d}estination list. B{filter} is used to decide which objects to append and which to skip. @param d: The destination list. @type d: list @param s: The source list. @type s: list @param filter: A filter that allows items to be appended. @type filter: L{Filter} """ for item in s: if item in filter: d.append(item) def __init__(self, schema, root): """ @param schema: The containing schema. @type schema: L{schema.Schema} @param root: The XML root node. @type root: L{Element} """ self.schema = schema self.root = root self.id = objid(self) self.name = root.get("name") self.qname = (self.name, schema.tns[1]) self.min = root.get("minOccurs") self.max = root.get("maxOccurs") self.type = root.get("type") self.ref = root.get("ref") self.form_qualified = schema.form_qualified self.nillable = False self.default = root.get("default") self.rawchildren = [] def attributes(self, filter=Filter()): """ Get only the attribute content. @param filter: A filter to constrain the result. @type filter: L{Filter} @return: A list of (attr, ancestry) tuples. @rtype: [(L{SchemaObject}, [L{SchemaObject},..]),..] """ result = [] for child, ancestry in self: if child.isattr() and child in filter: result.append((child, ancestry)) return result def children(self, filter=Filter()): """ Get only the I{direct} or non-attribute content. @param filter: A filter to constrain the result. @type filter: L{Filter} @return: A list tuples: (child, ancestry) @rtype: [(L{SchemaObject}, [L{SchemaObject},..]),..] """ result = [] for child, ancestry in self: if not child.isattr() and child in filter: result.append((child, ancestry)) return result def get_attribute(self, name): """ Get (find) an attribute by name. @param name: A attribute name. @type name: str @return: A tuple: the requested (attribute, ancestry). @rtype: (L{SchemaObject}, [L{SchemaObject},..]) """ for child, ancestry in self.attributes(): if child.name == name: return child, ancestry return None, [] def get_child(self, name): """ Get (find) a I{non-attribute} child by name. @param name: A child name. @type name: str @return: A tuple: the requested (child, ancestry). @rtype: (L{SchemaObject}, [L{SchemaObject},..]) """ for child, ancestry in self.children(): if child.any() or child.name == name: return child, ancestry return None, [] def namespace(self, prefix=None): """ Get this property's namespace. @param prefix: The default prefix. @type prefix: str @return: The schema's target namespace. @rtype: (I{prefix}, I{URI}) """ ns = self.schema.tns if ns[0] is None: ns = (prefix, ns[1]) return ns def default_namespace(self): return self.root.defaultNamespace() def multi_occurrence(self): """ Get whether the node has multiple occurrences, i.e. is a I{collection}. @return: True if it has, False if it has at most 1 occurrence. @rtype: boolean """ max = self.max if max is None: return False if max.isdigit(): return int(max) > 1 return max == "unbounded" def optional(self): """ Get whether this type is optional. @return: True if optional, else False. @rtype: boolean """ return self.min == "0" def required(self): """ Get whether this type is required. @return: True if required, else False. @rtype: boolean """ return not self.optional() def resolve(self, nobuiltin=False): """ Resolve the node's type reference and return the referenced type node. Only XSD schema objects that actually support 'having a type' custom implement this interface while others simply resolve as themselves. @param nobuiltin: Flag indicating whether resolving to an external XSD built-in type should not be allowed. @return: The resolved (true) type. @rtype: L{SchemaObject} """ return self def sequence(self): """ Get whether this is an . @return: True if , else False. @rtype: boolean """ return False def xslist(self): """ Get whether this is an . @return: True if , else False. @rtype: boolean """ return False def all(self): """ Get whether this is an . @return: True if , else False. @rtype: boolean """ return False def choice(self): """ Get whether this is an . @return: True if , else False. @rtype: boolean """ return False def any(self): """ Get whether this is an . @return: True if , else False. @rtype: boolean """ return False def builtin(self): """ Get whether this is a built-in schema-instance XSD type. @return: True if a built-in type, else False. @rtype: boolean """ return False def enum(self): """ Get whether this is a simple-type containing an enumeration. @return: True if enumeration, else False. @rtype: boolean """ return False def isattr(self): """ Get whether the object is a schema I{attribute} definition. @return: True if an attribute, else False. @rtype: boolean """ return False def extension(self): """ Get whether the object is an extension of another type. @return: True if an extension, else False. @rtype: boolean """ return False def restriction(self): """ Get whether the object is an restriction of another type. @return: True if a restriction, else False. @rtype: boolean """ return False def mixed(self): """Get whether the object has I{mixed} content.""" return False def find(self, qref, classes=[], ignore=None): """ Find a referenced type in self or children. Return None if not found. Qualified references for all schema objects checked in this search will be added to the set of ignored qualified references to avoid the find operation going into an infinite loop in case of recursively defined structures. @param qref: A qualified reference. @type qref: qref @param classes: A collection of classes used to qualify the match. @type classes: Collection(I{class},...), e.g. [I(class),...] @param ignore: A set of qualified references to ignore in this search. @type ignore: {qref,...} @return: The referenced type. @rtype: L{SchemaObject} @see: L{qualify()} """ if not len(classes): classes = (self.__class__,) if ignore is None: ignore = set() if self.qname in ignore: return ignore.add(self.qname) if self.qname == qref and self.__class__ in classes: return self for c in self.rawchildren: p = c.find(qref, classes, ignore=ignore) if p is not None: return p def translate(self, value, topython=True): """ Translate between an XSD type values and Python objects. When converting a Python object to an XSD type value the operation may return any Python object whose string representation matches the desired XSD type value. @param value: A value to translate. @type value: str if topython is True; any Python object otherwise @param topython: Flag indicating the translation direction. @type topython: bool @return: The converted I{language} type. """ return value def childtags(self): """ Get a list of valid child tag names. @return: A list of child tag names. @rtype: [str,...] """ return () def dependencies(self): """ Get a list of dependencies for dereferencing. @return: A merge dependency index and a list of dependencies. @rtype: (int, [L{SchemaObject},...]) """ return None, [] def autoqualified(self): """ The list of I{auto} qualified attribute values. Qualification means to convert values into I{qref}. @return: A list of attribute names. @rtype: list """ return ["type", "ref"] def qualify(self): """ Convert reference attribute values into a I{qref}. Constructed I{qref} uses the default document namespace. Since many WSDL schemas are written improperly: when the document does not define its default namespace, the schema target namespace is used to qualify references. """ defns = self.root.defaultNamespace() if Namespace.none(defns): defns = self.schema.tns for a in self.autoqualified(): ref = getattr(self, a) if ref is None: continue if isqref(ref): continue qref = qualify(ref, self.root, defns) log.debug("%s, convert %s='%s' to %s", self.id, a, ref, qref) setattr(self, a, qref) def merge(self, other): """Merge another object as needed.""" other.qualify() for n in ("default", "max", "min", "name", "nillable", "qname", "type"): if getattr(self, n) is not None: continue v = getattr(other, n) if v is None: continue setattr(self, n, v) def content(self, collection=None, filter=Filter(), history=None): """ Get a I{flattened} list of this node's contents. @param collection: A list to fill. @type collection: list @param filter: A filter used to constrain the result. @type filter: L{Filter} @param history: The history list used to prevent cyclic dependency. @type history: list @return: The filled list. @rtype: list """ if collection is None: collection = [] if history is None: history = [] if self in history: return collection history.append(self) if self in filter: collection.append(self) for c in self.rawchildren: c.content(collection, filter, history) history.pop() return collection def str(self, indent=0, history=None): """ Get a string representation of this object. @param indent: The indent. @type indent: int @return: A string. @rtype: str """ if history is None: history = [] if self in history: return "%s ..." % Repr(self) history.append(self) tab = "%*s" % (indent * 3, "") result = ["%s<%s" % (tab, self.id)] for n in self.description(): if not hasattr(self, n): continue v = getattr(self, n) if v is None: continue result.append(' %s="%s"' % (n, v)) if len(self): result.append(">") for c in self.rawchildren: result.append("\n") result.append(c.str(indent+1, history[:])) if c.isattr(): result.append("@") result.append("\n%s" % (tab,)) result.append("" % (self.__class__.__name__,)) else: result.append(" />") return "".join(result) def description(self): """ Get the names used for repr() and str() description. @return: A dictionary of relevant attributes. @rtype: [str,...] """ return () def __unicode__(self): return str(self.str()) def __repr__(self): s = [] s.append("<%s" % (self.id,)) for n in self.description(): if not hasattr(self, n): continue v = getattr(self, n) if v is None: continue s.append(' %s="%s"' % (n, v)) s.append(" />") return "".join(s) def __len__(self): n = 0 for x in self: n += 1 return n def __iter__(self): return Iter(self) def __getitem__(self, index): """ Returns a contained schema object referenced by its 0-based index. Returns None if such an object does not exist. """ i = 0 for c in self: if i == index: return c i += 1 class Iter: """ The content iterator - used to iterate the L{Content} children. The iterator provides a I{view} of the children that is free of container elements such as , or . @ivar stack: A stack used to control nesting. @type stack: list """ class Frame: """A content iterator frame.""" def __init__(self, sx): """ @param sx: A schema object. @type sx: L{SchemaObject} """ self.sx = sx self.items = sx.rawchildren self.index = 0 def __next__(self): """ Get the I{next} item in the frame's collection. @return: The next item or None @rtype: L{SchemaObject} """ if self.index < len(self.items): result = self.items[self.index] self.index += 1 return result def __iter__(self): return self def __init__(self, sx): """ @param sx: A schema object. @type sx: L{SchemaObject} """ self.stack = [] self.push(sx) def push(self, sx): """ Create a frame and push the specified object. @param sx: A schema object to push. @type sx: L{SchemaObject} """ self.stack.append(Iter.Frame(sx)) def pop(self): """ Pop the I{top} frame. @return: The popped frame. @rtype: L{Frame} @raise StopIteration: when stack is empty. """ if self.stack: return self.stack.pop() raise StopIteration() def top(self): """ Get the I{top} frame. @return: The top frame. @rtype: L{Frame} @raise StopIteration: when stack is empty. """ if self.stack: return self.stack[-1] raise StopIteration() def __next__(self): """ Get the next item. @return: A tuple: the next (child, ancestry). @rtype: (L{SchemaObject}, [L{SchemaObject},..]) @raise StopIteration: A the end. """ frame = self.top() while True: result = next(frame) if result is None: self.pop() return next(self) if isinstance(result, Content): ancestry = [f.sx for f in self.stack] return result, ancestry self.push(result) return next(self) def __iter__(self): return self class XBuiltin(SchemaObject): """Represents a built-in XSD schema node.""" def __init__(self, schema, name): """ @param schema: The containing schema. @type schema: L{schema.Schema} """ root = Element(name) SchemaObject.__init__(self, schema, root) self.name = name self.nillable = True def namespace(self, prefix=None): return Namespace.xsdns def builtin(self): return True class Content(SchemaObject): """XSD schema objects representing real XML document content.""" pass class NodeFinder: """ Find nodes based on flexable criteria. I{matcher} may be any object implementing a match(n) method. @ivar matcher: An object used as criteria for match. @type matcher: I{any}.match(n) @ivar limit: Limit the number of matches. 0=unlimited. @type limit: int """ def __init__(self, matcher, limit=0): """ @param matcher: An object used as criteria for match. @type matcher: I{any}.match(n) @param limit: Limit the number of matches. 0=unlimited. @type limit: int """ self.matcher = matcher self.limit = limit def find(self, node, list): """ Traverse the tree looking for matches. @param node: A node to match on. @type node: L{SchemaObject} @param list: A list to fill. @type list: list """ if self.matcher.match(node): list.append(node) self.limit -= 1 if self.limit == 0: return for c in node.rawchildren: self.find(c, list) return self