Apparatus and method enabling a compiled program to exactly recreate its source code5249275Abstract A computer system capable of compiling PROLOG programs appends to the compiled code information necessary to recreate the source which generated the compiled code. This allows program statements which use the original source code to be compiled. Each program statement is compiled into a separate block of code, with all statements in a procedure linked together in lists. Program statements can be compiled as they are created, and statements can be added or deleted by adding or deleting blocks of compiled code from the appropriate lists. Claims What is claimed is: Description BACKGROUND OF THE INVENTION
TABLE I
______________________________________
a(1):- a1(X,Y), a2(Q,Z)
a(2):- a2(X,Z), b2(2)
a(X):- a2(1,Y), b2(Z)
a(X):- a1(g(X),3), b3, g2(4)
b2 :- C3(X), C2(2)
b2 :- g(4)
b2 :- g(X), g1(X,Y),g2(X)
a1(1,1) :- g(2), g2(2)
a1(1,2) :- C2(X),g(1)
a1(2,X) :- b3, b4
______________________________________
No particular meaning is intended to be associated with the clauses shown in Table I. The first four clauses can be considered to be a single procedure. Taken together, they determine the truth of the assertion a, which has a single argument. The next three clauses are considered to be a procedure which determines the truth of the assertion b2 and the last three determine the truth of the assertion a1, which has two arguments. The remaining clauses used in Table I are assumed to be defined elsewhere. The first two clauses in the procedure for a have arguments which are constants, while the last two have arguments which are variables. These two types of clauses will generally execute under different circumstances, and it is convenient to group them separately during compilation in order to improve execution speed. FIG. 1 shows a segment of compiled code 10 which corresponds to a PROLOG procedure, for example procedure a of Table I. The first item in the procedure is a jump table 12 which is consulted upon entering the procedure. The jump table 12 contains address locations for the beginning of different execution code blocks. These code blocks contain the compiled code for the procedure. As described above, different portions of the procedure are entered depending upon the type of argument passed into it. If a constant or variable is passed to the procedure as described above in connection with Table I, the corresponding code block for constants 14 or variables 16 is jumped to and executed. The argument passed to a procedure can also be a list or a structure, which types were not shown in Table I, and arguments of these types cause execution of the procedure to be transferred to the corresponding lists 18 or structures 20 code blocks. The compiled procedure code does not contain information sufficient to recreate the source code from which it was generated. Thus, this type of compilation procedure cannot be used to compile clauses which will be used by a CLAUSE command. The CLAUSE command has the syntax CLAUSE(head, body), where head is the procedure identifier, and body is a value to be matched. Clauses which match the current value of body will be formed into a data structure, which is returned by CLAUSE. The matching is done at the source code level, and source code of the matching clauses is used to form the data structure. The RETRACT command has the syntax RETRACT(head, body), where head is the procedure identifier, and body is a value to be matched. As in the CLAUSE command, the matching is done of source code. Clauses which match body must be removed from procedure head at execution time. Since the source code cannot be matched, RETRACT cannot be implemented by prior art compilation schemes. Assert has the syntax ASSERT (clause), where clause will be inserted into the appropriate procedure. There is no extra room left in the various code blocks 14, 16, 18, 20, so that the ASSERT command cannot be used to add additional compiled clauses to the procedure at execution time. A system which allows the CLAUSE, ASSERT and RETRACT commands to be compiled is illustrated in FIG. 2. Instead of compiling a procedure into separate code blocks and a jump table, a procedure is compiled into a procedure execution frame 30 (PEF). The PEF consists of four pointers 32, 34, 36, 38, each pointer corresponding to one of the argument types which can be passed to the procedure. Thus, the four pointers correspond to parameters which are constants, variables, lists and structures. Each pointer is the head of a linked list. Each clause is compiled separately into a clause execution frame 40 (CEF). The details of the CEF will be shown in FIG. 3, but each CEF contains compiled code corresponding to a separate clause. All of the CEF's for each argument type are coupled together into a linked list as shown. When a procedure is called, the computer system determines the type of the argument. The correct pointer is selected based upon that argument type, and the individual CEF's in the corresponding linked list are executed until either a true value is returned or the end of the list is reached. In some ways, the PEF 30 functions in a manner similar to the jump table 12 of FIG. 1. The procedure a of Table I would have two CEF's in the list pointed to by the constants pointer 32, and two CEF's in the list pointed to by the variables pointer 34. The lists and structures pointers 36, 38 would not point to any list, and would typically contain a standard empty pointer known as NIL. Many systems will contain features to further restrict the number of clauses, CEFs in the preferred embodiment, which are executed during a procedure call. Because PROLOG operates by executing all relevant clauses within a procedure, the performance of large systems degrades considerably unless steps are taken to minimize the number of clauses which are relevant. One typical method is known as indexing, which involves the use of hash tables. A separate hash table is kept for each pointer in the PEF. Once the argument type has been determined (i.e., constant, structure, etc.), the first one or two arguments are hashed and a hash table (not shown) is consulted in a known manner. This table will point to one of a set of linked CEFs, and these are the only ones that must be executed to execute the procedure. Thus, there are a larger number of shorter lists of CEFs, rather than only four longer lists. It will be appreciated that, in the general case, a shorter list will be executed to completion faster than a longer list. Thus, the pointers 32, 34, 36, 38 in the PEF can point to structures other than simple linked lists. Any structure interconnected by pointers may be used. Consider now FIG. 6 which shows a flow chart of a method according to the present invention for compiling a procedure having a plurality of clauses, comprising the steps of: (a) creating a procedure execution frame having a plurality of pointers, each pointer corresponding to a preselected subset of clauses of the procedure having similar characteristics (Block 80); separately compiling each clause into an object code block (Block 82); attaching to each object code block information sufficient to recreate the clause from which that object code block was compiled, to create a clause execution frame (Block 84); and inserting each clause execution frame into a data structure which is pointed to by one of the procedure execution frame pointers (Block 86). Referring to FIG. 3, detailed construction of the CEF 40 is shown. A header 42 contains typical data required by the system, such as length of the CEF code block 40, data-typing information, and so forth. Next-pointer 42 is the pointer to the next CEF in the list, and will be NIL if a particular CEF is the last element of the linked list. The compiled code portion 46 of the CEF 40 is identical to that code which is compiled into a code block of the compiled procedure 10 shown in FIG. 1. Note that the compiled code blocks 14, 16, 18, 20 of FIG. 1 each contain the compiled code for all clauses of the appropriate type, while the compiled code portion 46 of a CEF contains code for only a single clause. The last segment 48 of the CEF 40 contains clause mode commands, which are those additional commands necessary for recreation of the source code from which that CEF 40 was compiled. When the system is executing in a normal mode, that is, not in clause mode, the compiled code 46 is executed directly when the CEF 40 is entered. If the compiled code 46 indicates that this particular clause has failed, the system follows the next-pointer 44 to the next CEF on the list and repeats the process. When the processor is operating in the clause mode, upon entry into the CEF 40 the system executes the first of the clause mode commands 48. Upon exit from the CEF 40, a data structure has been created, usually in the system heap space, which is identical to the source code from which the CEF 40 was originally compiled. The clause mode commands within the CEF may take any form which is suitable for recreating the source code. For example, the clause mode commands could simply be a copy of the original source code. Upon entering the CEF 40 in clause mode, the source code statements in the clause code commands section 48 would simply be copied to the heap, and the CEF 40 exited. Such a solution is wasteful of space, inasmuch as there is redundant information contained between the compiled code 46 and the retained source code 48. In the preferred embodiment, special clause mode commands are made available to the processor which uses the information in the compiled code 46 to assist in the recreation of the original source code. FIG. 4 shows a CEF 50 which corresponds to the first clause of Table I. The clause is recreated as part of FIG. 4 for convenience. The sequence of events which occurs when the CEF 50 is called in a normal mode will now be described. To determine whether the goal a(1) is true, the two subterms a2 (X,Y) and a2 (Q,Z) must be evaluated. The compiled code for doing so is shown in the compiled code portion 52 of the CEF 50. The current values for X and Y are put into argument registers (not shown) from a frame on a local system stack (not shown), and a call made to the procedure a1. If a1 fails, the PROLOG system failure mechanism terminates execution of this CEF, removing the stock frame and following next-pointer 56 if it is not NIL. If as succeeds, then the current values of the variables Z and Q are placed into argument registers and a call made to procedure a2. If a2 succeeds, this clause succeeds and procedure a will return successfully. The clause has succeeded, if execution reaches the proceed command. If a2 fails, a1 will be retried beginning from the point where it left off. This will be the next CEF in the list in a1 after that CEF which last returned successfully. If there is no pair for which a1 and a2 both succeed, head unification on this clause will fail. If this is the last clause in the list, the procedure a will have failed, and the system failure mechanism will unsuccessfully return from procedure a. This is done by removing from the system stack the most recent frame holding a choice point, which choice-point frame will be pointed to by a backtrack pointer as known in the art. Execution of procedure a is then complete. If the system is executing instructions in clause mode, upon entry to the CEF 50 control is transferred to the first clause mode command 58. This command 58 puts a conjunction command, shown as a comma in PROLOG, onto the system heap 70 as the beginning of this portion of the data structure to be created there. The system heap 70 built by executing the CEF 50 of FIG. 4 is shown in FIG. 5. The command Put-Heap(,) is accompanied by a number of unify-var() commands equal to the number of subterms in this clause. Since this clause has 2 subterms a1 and a2, unify-var(A) and unify-var(B) are used to put two pointer positions A and B onto the heap 70. The pointers are initially NIL pointers, but will be later reassigned as described below. The next clause mode command 60, Put-Heap(a1, A), puts the procedure identifier a1 onto the heap, and changes the value of the pointer A to point to a1, as shown in FIG. 5. Execution then branches to the point labeled (1) in the compiled code portion 52 of the CEF 50. When this branch is made, a location pointer to the next clause mode command 62 is retained. Since the processor is executing in the clause mode, the two commands put(X) and put(Y) put X and Y onto the heap to be included in the data structure, instead of putting them into the argument registers to be executed. The call (a1) statement, instead of executing a procedure call, returns to the location pointer in the clause code commands 54. The next clause mode command 62 is then executed, which puts the procedure identifier a2 onto the heap and changes the value of the pointer B to point to a2. A branch to location (2) is made, and the variables Z and Q are put onto the heap as described above. The call(a2) command transfers control back to the clause mode commands 54. When execution reaches proceed command 64, the CEF has been completed. The source code for the clause has now been re-created on the heap 70, and can be compared to the body of the CLAUSE or RETRACT command, whichever is operating. If a match is found, procedure a terminates; if not, the heap pointer is reset to delete the data which was put there during execution of this CEF, and the CEF pointed to by next-pointer 56 is executed. Alternatively, the comparison can be made as each item is added to the heap, with the failure mechanism jumping to the next CEF as soon as a difference is detected. Using this method, simply reaching the proceed command 64 indicates that this CEF was successfully matched with the pattern, and successfully returns from the procedure. If the argument to a goal is a list or structure, rather than a constant or variable, it cannot generally be copied to a single location on the heap as can the variables X, Y, Z and Q. Instead, the list or structure can be created on the heap first, and a pointer to it put onto the heap in a position corresponding to that of X, Y, Z or Q. TECHNICAL ADVANTAGES When the source code is compiled into CEFs and PEFs, and the processor is modified to accept the existence of the clause mode, preferably by simply changing a flag setting, the commands CLAUSE, ASSERT and RETRACT are easily compiled for execution. The CLAUSE command is executed by changing the state of the processor to the clause mode and operating on the appropriate procedures in the manner described in connection with FIGS. 3 and 4. The ASSERT mode, which adds a clause to a procedure, is easily implemented. When a clause is ASSERTed, it is compiled into a CEF 40 and inserted into the appropriate linked list coming from the appropriate PEF 30. The RETRACT statement, which deletes clauses which pattern match with its argument, is accomplished almost as easily. Since the pattern match must be made to the source code of the clause, the processor switches into the clause mode and recreates the source code for the clause in each relevant CEF 40. If a pattern match is then found, that CEF 40 is simply deleted from the linked list in a known manner. The generated code is completely compiled. That is, the clause mode commands as well as the compiled code located in the CEF's are machine language instructions. No source code remains to be interpreted at execution time. This causes the execution speed of the PROLOG program to be much faster than that of prior art systems, which are partially compiled and partially interpreted. Many typical applications will run 5-10 times faster when fully compiled, and applications which operate in the clause mode for any significant portion of their execution time, such as meta-interpreters, can execute up to 100 times faster. The current invention allows the PROLOG development environment to also be greatly improved over previous systems. In prior art PROLOG systems, a complete source code file must be created, and then compiled in a single step. In actual practice, this usually means that during the early debugging phases of program creation, the code is run completely interpreted. Once the program has been debugged, those portions of it which can be compiled are then compiled. Before compilation, the programmer must carefully delineate which portions of the code are to be compiled and which are to be interpreted. Since PROLOG programs are not deterministic at compile time, there is a high likelihood of mistakes, leading to a run-time crash as the system tries to CLAUSE a segment of compiled code. In addition, it is not unusual for the compiled code to execute differently than the interpreted code, which is clearly not desirable. The current system allows a PROLOG program to be incrementally compiled as it is created. As each clause is entered, it is compiled into a CEF 40. That CEF 40 is then linked into the appropriate PEF 30. Incomplete programs can be run at any time in order to aid debugging, and such execution is always done on fully compiled code. Thus, the execution time of programs in the development phase is also greatly improved over the prior art. Since the generated code is completely compiled, efficiency techniques, such as the use of hash tables for indexing described in connection with FIG. 2, can be used in the clause mode as well as in normal mode. This could not be done using the prior art, which runs interpretively in the clause mode. An important restriction on the use of programs compiled using prior art methods is that the program cannot be changed on the fly (using CLAUSE, RETRACT and ASSERT) unless the programmer has previously determined which portions of the code might be changed, and caused those portions to remain interpreted. This severely restricts the ability of the user to change the data base and rules to fit his particular circumstances, with the result that current applications which require data to be added and modified by the user are generally interpreted. The present invention removes all such restrictions, as all changes can be immediately compiled. The present invention has been illustrated by the system described above, and it will become apparent to those skilled in the art that various modifications and alterations may be made thereto. Such variations fall within the spirit of the present invention, the scope of which is defined by the appended claims.
|
Same subclass Same class Consider this |
||||||||||