🚀 go-pugleaf

The generated-code quality of LSP is, on the whole, very similar to that 
of TML Pascal.  Both are single-pass compilers which are incapable of 
performing the kinds of optimizations possible in MPW Pascal (e.g. 
assigning local variables to registers).

However, I can think of one area in which LSP would be (significantly)
slower than TML:  SET OF integer-subrange operations.  One of the design
goals of LSP (and one that I sometimes regret) was the need to retain full
compatibility with the Macintosh Pascal interpreter.  In this case, the
interpreter provided for SET OF INTEGER as the largest set type.  Operations
on sets which may contain negative elements are cumbersome, since the
requirement of ToolBox compatibility constrains the choice of representation.  
Furthermore, because the size of intermediate values can be large
(up to 8K!) they are allocated from the heap (typically a slow operation). 
Sets of base-type other than integer or integer subrange are *much* more
efficient.

My guess is that the program that you are benchmarking uses SETs of integer
subranges, and that you have struck one of the few areas in which
compatibility with Macintosh Pascal has cost us some performance
(text I/O is another).

It is unfortunate that Lightspeed Pascal's performance is inferior in this
case, but this is not indicative of its performance on typical programs.

                                    John McEnerney
                                    THINK Technologies, Inc.