Web Sites Verification: An Abductive Logic Programming Tool

We present the CIFFWEB system, an innovative tool for the verification of web sites, relying upon abductive logic programming. The system allows the user to define rules that a web site should fulfill by using (a fragment of) the query language Xcerpt. The rules are translated into abductive logic programs with constraints and their fulfillment is checked through the CIFF abductive proof procedure. 1 Web Checking Rules The exponential growth of the WWW raises the question of maintaining and repairing automatically web pages at both structural and data level. Our web checking rules are characterized by using (a fragment of) the query language Xcerpt [2] for expressing complex queries in a natural syntax. The following is an example of a XML page [P1] about a theater company (left) and a rule expressing that no director should occur twice in the director list [R1] (right). %%% directorindex.xml GOAL all error [var D,"double director"] FROM dir1 in {resource{"file:directorindex.xml"}, dir2 dirlist {{ dir2 dir {{var D}}, dir {{var D}} }} } END Web checking rules are specified by a condition part (starting with FROM) and an error part (starting with GOAL). The intuitive meaning of a rule is that for each instance in the XML resource(s) matching the condition part, an error needs to be returned. Due to lack of space, the example does not cover the whole expressiveness of our rules: in particular the absence of XML data (without construct) and (arithmetical) constraints over variables can also be expressed. However, to the best of our knowledge, Xcerpt lacks of both a clear semantics for negation constructs (without) and a concrete tool for evaluating Xcerpt queries. Hence, we map web checking rules into abductive logic programs with constraints (ALPCs) that can be fed as input to the CIFF System 4.0, an implementation of the general-purpose CIFF abductive proof procedure [3] which is sound with respect to the 3-valued completion semantics. Using CIFF for determining the fulfillment of the rules, we inherit its formal properties, thus obtaining a sound concrete tool for web sites verification. 2 Mapping rules to abductive logic programs The CIFF proof procedure gets as input an ALPC 〈P,A, IC〉< where P is a normal logic program (with constraints), A is a set of abducible predicates and IC is a set of integrity constraints of the form L1 ∧ · · · ∧ Lm → H where each Li is a literal and H is a disjunction of atoms. Constraint atoms are evaluated wrt an underlying structure <, as in constraint logic programming [3]. Since CIFF is not designed to handle directly XML resources, we translate XML pages into sets of atoms of the form pg el(ID,Tag,IDF). Each XML Tag element is associated to both a unique ID and to its father’s id (IDF) in order to represent the page structure. The same is done for the data inside a XML tag. Similarly, we translate Xcerpt rules into ALPCs, where abducibles are associated with errors. In the absence of negation (without construct) each web checking rule is translated into a single integrity constraint whose head is an abducible atom abd err(Args,Msg) representing the error. As an example, the translation of both [P1] and [R1] is the following. [P1]: pg_el(1,dirlist,_). pg_el(2,dir,1). data_el(3,’dir1’,2). pg_el(4,dir,1). data_el(5,’dir2’,4). pg_el(6,dir,1). data_el(7,’dir2’,6). [R1]: [pg_el(ID1,dirlist,_), pg_el(ID2,dir,ID1), data_el(ID3,D,ID2), pg_el(ID4,dir,ID1), data_el(ID5,D,ID4), ID2 #\= ID4] implies [abd_err([D],"double director")]. Running the CIFF System 4.0 with the above input, the abductive answer [abd err([’dir2’],’double director’)] is correctly produced.