OUTPUT = 'This is a line';
It has inline machine code. "Programmers are given all the rope they ask for. Novices tend to hang themselves fairly frequently." XPL has been implemented on IBM 360, Univac 1100, ICL System 4, CDC 6000 and Cyber series, XDS Sigma-5 and Sigma-7, DEC PDP-10, IA32, FreeBSD and Linux.
An optimising XPL compiler (version 1) by Robin Vowels <firstname.lastname@example.org> is a standard implementation of XPL and is based on McKeeman, Horning, and Wortman's improved XCOM (which employs hashed symbol table generation). It includes the extra built-in function COREHALFWORD.
The following areas have been optimised: procedures calls when the argument and corresponding parameter are of the same type, and when the argument is a constant; constant subscripts; use of CORELHALFWORD and COREWORD; string constants of length one; iterative DO statements by transferring code to the end of the loop.
String constants of length one do not require a descriptor, hence more descriptors are available for string variables. Comparison operations are treated as commutative, and an improved Commute algorithm is used. Halfword instructions are generated for BIT(16) variables.
These areas have been improved or re-written: calls on OUTPUT, catenation, integer-to-string conversion, multiply, divide, and MOD. An emitter for SS-type instructions has been added. The compiler achieves an 11% reduction in object code compiling itself, an 11% increase in compilation rate, a 55% increase in compilation speed when the $E toggle is set. Special treatment for catenating a string to an integer substantially decreases consumption of the free string area, and decreases string moves. The latter improvement is most noticeable on small core machines.
Core requirements: less than the improved XCOM on which it is based (approx. 98000 bytes). Symbol table size is 468. Ported to IBM System 370. The compiler is written in XPL. The code generators are machine-specific.
[JCC, AFIPS 1968].