Ramkrishna Chatterjee, Barbara Ryder, and William Landi. Complexity of concrete type-inference in the presence of excep tions. In LNCS 1381, Proceedings of European Symposium on Programming, April 1998.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....written in languages that have frequently occurring exceptional control flow. The Vortex compiler [8] used a straightforward extension of a traditional control flow graph to enable both intraprocedural and interprocedural analysis of Java, Modula 3 [17] and Cecil [5] programs. Chatterjee et al. [6] have developed an extension of Landi and Ryder s conditional points to [14] that can precisely model exceptional control flow in the ICFG [16, 19, 4, 14, 12] Sinha and Harrold [20] detail the subtleties of correctly capturing the semantics of Java s try catch finally structure in a traditional ....

Ramkrishna Chatterjee, Barbara G. Ryder, and William A. Landi. Complexity of concrete type-inference in the presence of exceptions. In Lecture Notes in Computer Science, 1381, pages 57--74. Springer-Verlag, April 1998. Proceedings of the European Symposium on Programming.

....Reps, Horwitz and Sagiv [RHS95] and Landi and Ryder [LR91] Keywords: Points to Analysis, Java, Exceptions, Complexity. # The research reported here was supported, in part, by NSF grants GER 9023628, CCR97 04703 and the Hewlett Packard Corporation. An abridged version of this paper appeared as [CRL98]. 1 Introduction Compile time analysis of statically typed object oriented programming languages (e.g. Java, C ) is rendered di#cult by several factors. To trace flow of control through call statements which use dynamic binding (e.g. a virtual call in C , a call to a non static Java method) ....

Ramkrishna Chatterjee, Barbara Ryder, and William Landi. Complexity of concrete type-inference in the presence of excep tions. In LNCS 1381, Proceedings of European Symposium on Programming, April 1998.

....dfelms computed by Phase I must be extended to account for the possible controlflow due to exceptions. Under the same assumptions as in Section 3.5, RCI is polynomial time even in the presence of exceptions. A complexity classification of doing analysis in the presence of exceptions is given in [CRL98a] Another important property of RCI is that incomplete programs such as libraries can be analyzed. A unknown initial value at the entry node of a method of a library, that can be directly invoked from a call site outside the library, is called an interface initial value. When RCI is used for ....

Ramkrishna Chatterjee, Barbara Ryder, and William Landi. Complexity of concrete type-inference in the presence of exceptions. In LNCS 1381, Proceedings of European Symposium on Programming, April 1998.

....of exceptions, and discuss the complexity and correctness of the extended algorithm. Finally, we present PSPACE hardness results about CTI in the presence of exceptions. Due to lack of space, we have omitted all proofs. These proofs and further details about the results in this paper are given in [CRL97] 5 . 2 Basic definitions Program representation. Our algorithm operates on an interprocedural control flow graph or ICFG [LR91] An ICFG contains a control flow graph (CFG) for each method in the program. Each statement in a method is represented by a node in the method s CFG. Each call site ....

....by the above routines are defined in Appendix B. Appendix A contains an example which illustrates the basic algorithm. Precision of the basic algorithm. By induction on the number of iterations needed to compute a data flow element and the length of a path associated with a data flow element, in [CRL97] we prove that the basic algorithm computes the initialize worklist. Each worklist node contains a data flow element, which is a conditional points to or reachable, and an ICFG node. create a worklist node containing the entry node of main and reachable, and add it to the worklist; ....

[Article contains additional citation context not shown here]

Ramkrishna Chatterjee, Barbara Ryder, and William Landi. Complexity of concrete type-inference in the presence of exceptions. Technical Report DCS-TR-341, Dept of CS, Rutgers University, September 1997.

....(b) RCI can analyze incomplete programs such as libraries. c) RCI can analyze programs that have exceptions. 2. Chapter 3 presents several new complexity characterizations of points to analysis in the presence of object oriented language constructs such as dynamic dispatch and exceptions [CRL98a] 8 3. Chapter 4 presents a new approach to data flow based testing of object oriented libraries [CR99] 4. Chapter 5 discusses a proof of concept prototype of RCI for points to analysis of C programs [CRL99] The empirical results obtained using this prototype show that RCI is e#ective in ....

....is generated, which belongs to the final points to solution at program point 6. 84 Chapter 3 Complexity Characterizations In this chapter we characterize the complexity of points to analysis in the presence of object oriented language constructs: exceptions and dynamic dispatch. Our results [CRL98a, CRL97] clearly identify the di#cult features and indicate the approximations any e#cient algorithm has to make. Our results are summarized in Table 3.1. The hierarchy of complexity classes is shown in Figure 3.1. It is known that NC is a proper subset of P Space. Apart from this, it is not ....

Ramkrishna Chatterjee, Barbara Ryder, and William Landi. Complexity of concrete type-inference in the presence of exceptions. In LNCS 1381, Proceedings of European Symposium on Programming, April 1998.

....which belongs to the final points to solution at program point 6. 84 Chapter 3 Complexity Characterizations In this chapter we characterize the complexity of points to analysis in the presence of object oriented language constructs: exceptions and dynamic dispatch. Our results [CRL98a, CRL97] clearly identify the di#cult features and indicate the approximations any e#cient algorithm has to make. Our results are summarized in Table 3.1. The hierarchy of complexity classes is shown in Figure 3.1. It is known that NC is a proper subset of P Space. Apart from this, it is not known if ....

Ramkrishna Chatterjee, Barbara Ryder, and William Landi. Complexity of concrete type-inference in the presence of exceptions. Technical Report DCS-TR-341, Dept of CS, Rutgers University, September 1997.

.... er = N 2k h k, ep = N 2 h . Here er is the worst case cost of evaluating the relevant context of the dfelm and ep is the worst case cost of evaluating the points to of the dfelm. Hence the worst case cost of this phase is O(N 2k 2 h k # Nnodes # Nd fe ) A single level type [CRL98] is either a class all of whose fields are of primitive types or it is an array of a primitive type. For programs with only single level types and without dynamic dispatch, CR97] shows that an algorithm similar to RCI computes the precise solution of points to analysis in O(n 4 ) worst case ....

....assumption of data flow analysis: all realizable paths are executable. This is an improvement over the O(n 7 ) worst case bound achievable by applying previous techniques [LR91, RHS95] to this case. If dynamic dispatch or fields of pointer type are allowed, points to analysis is P space hard [CRL98, Lan92] 4 Exceptions In this section, we extend RCI for points to analysis in the presence of exceptions because, unlike C , exceptions are frequently used in Java programs. In [CRL98, CRL97] we showed how to do points to analysis of whole programs in the presence of exceptions. The ....

[Article contains additional citation context not shown here]

Ramkrishna Chatterjee, Barbara Ryder, and William Landi. Complexity of concrete type-inference in the presence of exceptions. In LNCS 1381, Proceedings of European Symposium on Programming, April 1998.

....case. If dynamic dispatch or fields of pointer type are allowed, points to analysis is P space hard [CRL98, Lan92] 4 Exceptions In this section, we extend RCI for points to analysis in the presence of exceptions because, unlike C , exceptions are frequently used in Java programs. In [CRL98, CRL97] we showed how to do points to analysis of whole programs in the presence of exceptions. The algorithm in [CRL98, CRL97] needs to keep the whole program in memory and cannot analyze incomplete programs. Here, we extend the ideas presented in [CRL98, CRL97] for modular analysis. The semantics of ....

....In this section, we extend RCI for points to analysis in the presence of exceptions because, unlike C , exceptions are frequently used in Java programs. In [CRL98, CRL97] we showed how to do points to analysis of whole programs in the presence of exceptions. The algorithm in [CRL98, CRL97] needs to keep the whole program in memory and cannot analyze incomplete programs. Here, we extend the ideas presented in [CRL98, CRL97] for modular analysis. The semantics of exceptions in Java is discussed in [GJS96, CRL97] Data Representations. In the presence of exceptions, each dfelm ....

[Article contains additional citation context not shown here]

Ramkrishna Chatterjee, Barbara Ryder, and William Landi. Complexity of concrete type-inference in the presence of exceptions. Technical Report DCS-TR-341, Dept of CS, Rutgers University, September 1997.

....with it; this must be executed [GJS96] no matter how the try statement terminates: with an exception or without an exception. As stated earlier, for the ease of presentation we have ignored finally statements in this paper. These can be easily accommodated using the approach described in [CRL98, CRL97]. Intuitively an additional kind of ECFInfo is needed to capture control flow when a try statement terminates normally or when it terminates due to a labelled break or continue. A call or a try statement nested inside a finally can cause exceptions to stack up. However, singleton ECFInfo is still ....

Ramkrishna Chatterjee, Barbara Ryder, and William Landi. Complexity of concrete type-inference in the presence of exceptions. Technical Report DCS-TR-341, Dept of CS, Rutgers University, September 1997.

....concrete type inference of Java and C 4 ; Sections 4.3 and 4. 4) a proof that this technique computes the precise solution for programs with only single level types and without dynamic dispatch, and has the worst case complexity of O(n 4 ) which improves upon the O(n 7 ) 3 In [CRL97] we show how to do concrete type inference in the presence of exceptions. 4 In this paper, we present our algorithm only for Java. worst case bound achievable by applying previous techniques of [RHS95] and [LR91] to this case; Section 6) a new method for demand driven computation using ....

....1 deriv2 574 44 40 3 chess 437 47 47 1 deltablue 1258 102 102 1 richards 836 83 83 1 FeynDiagram 3976 237 225 4 vvector 1780 128 128 1 vmatrix1 1223 100 100 1 Table 1. Call graph decomposition and without dynamic dispatch (the only known natural polynomial time solvable special case [LR91, PR96, CRL97] if exceptions and threads are excluded) Related work of the second kind include [DGS95] and [HRS95] In both of these, each data flow element is treated uniformly and computed from scratch on demand, although later queries may be answered more efficiently because of caching. However, ....

Ramkrishna Chatterjee, Barbara Ryder, and William Landi. Complexity of concrete type-inference in the presence of exceptions. Technical Report DCS-TR-341, Dept of CS, Rutgers University, September 1997.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC