XPL, for expert's programming language is a programming language based on PL/I, a portable one-pass compiler written in its own language, and a parser generator tool for easily implementing similar compilers for other languages. XPL was designed in 1967 as a way to teach compiler design principles and as starting point for students to build compilers for their own languages. XPL was designed and implemented by William M. McKeeman, David B. Wortman, James J. Horning and others at Stanford University. XPL was first announced at the 1968 Fall Joint Computer Conference. The methods and compiler are described in detail in the 1971 textbook A Compiler Generator. They called the combined work a 'compiler generator'. But that implies little or no language- or target-specific programming is required to build a compiler for a new language or new target. A better label for XPL is a translator writing system. It helps to write a compiler with less new or changed programming code.
Language The XPL language is a simple, small, efficient dialect of PL/I intended mainly for the task of writing compilers. The XPL language was also used for other purposes once it was available. XPL can be compiled easily to most modern machines by a simple compiler. Compiler internals can be written easily in XPL, and the code is easy to read. The PL/I language was designed by an IBM committee in 1964 as a comprehensive language replacing Fortran, COBOL, and ALGOL, and meeting all customer and internal needs. These ambitious goals made PL/I complex, hard to implement efficiently, and sometimes surprising when used. XPL is a small dialect of the full language. XPL has one added feature not found in PL/I: a STRING datatype with dynamic lengths. String values live in a separate text-only heap memory space with automatic garbage collection of stale values. Much of what a simple compiler does is manipulating input text and output byte streams, so this feature helps simplify XPL-based compilers.
Components
XCOM The XPL compiler, called XCOM, is a one-pass compiler using a table-driven parser and simple code generation techniques. Versions of XCOM exist for different machine architectures, using different hand-written code generation modules for those targets. The original target was IBM System/360, which is a proper subset of IBM System/370, IBM System/390 and IBM System z. XCOM compiles from XPL source code, but since XCOM itself is written in XPL it can compile itself – it is a self-compiling compiler, not reliant on other compilers. Several famous languages have self-compiling compilers, including Burroughs B5000 Algol, PL/I, C, LISP, and Java. Creating such compilers is a chicken-and-egg conundrum. The language is first implemented by a temporary compiler written in some other language, or even by an interpreter (often an interpreter for an intermediate code, as BCPL can do with intcode or O-code). XCOM began as an Algol program running on Burroughs machines, translating XPL source code into System/360 machine code. The XPL team manually turned its Algol source code into XPL source code. That XPL version of XCOM was then compiled on Burroughs, creating a self-compiling XCOM for System/360 machines. The Algol version was then thrown away, and all further improvements happened in the XPL version only. This is called bootstrapping the compiler. The authors of XPL invented the tombstone diagram or T-diagram to document the bootstrapping process. Retargeting the compiler for a new machine architecture is a similar exercise, except only the code generation modules need to be changed. XCOM is a one-pass compiler (but with an emitted code fix-up process for forward branches, loops and other defined situations). It emits machine code for each statement as each grammar rule within a statement is recognized, rather than waiting until it has parsed the entire procedure or entire program. There are no parse trees or other required intermediate program forms, and no loop-wide or procedure-wide optimizations. XCOM does, however, perform peephole optimization. The code generation response to each grammar rule is attached to that rule. This immediate approach can result in inefficient code and inefficient use of machine registers. Such are offset by the efficiency of implementation, namely, the use of dynamic strings mentioned earlier: in processing text during compilation, substring operations are frequently performed. These are as fast as an assignment to an integer; the actual substring is not moved. In short, it is quick, easy to teach in a short course, fits into modest-sized memories, and is easy to change for different languages or different target machines.
ANALYZER The XCOM compiler has a hand-written lexical scanner and a mechanically-generated parser. The syntax of the compiler's input language (in this case, XPL) is described by a simplified BNF grammar. XPL's grammar analyzer tool ANALYZER or XA turns this into a set of large data tables describing all legal combinations of the syntax rules and how to discern them. This table generation step is re-done only when the language is changed. When the compiler runs, those data tables are used by a small, language-independent parsing algorithm to parse and respond to the input language. This style of table-driven parser is generally easier to write than an entirely hand-written recursive descent parser. XCOM uses a bottom-up parsing method, in which the compiler can delay its decision about which syntax rule it has encountered until it has seen the rightmost end of that phrase. This handles a wider range of programming languages than top-down methods, in which the compiler must guess or commit to a specific syntax rule early, when it has only seen the left end of a phrase.
Runtime XPL includes a minimal runtime support library for allocating and garbage-collecting XPL string values. The source code for this library must be included into almost every program written in XPL.
SKELETON The last piece of the XPL compiler writing system is an example compiler named SKELETON. This is just XCOM with parse tables for an example toy grammar instead of XPL's full grammar. It is a starting point for building a compiler for some new language, if that language differs much from XPL.
… excerpt ends here. Continue reading the full article.
