License: Apache 2.0

• tested with Java 8, developed with IntelliJ and cmd line

• it assumes SWI-Prolog 7.x for compiling from .pl to .pl.nl files

usage:

• go.sh add
• go.sh perms
• go.sh queens

or

• go.sh < any pure Prolog program in directory prog >

with Prolog files assumed having suffix .pl

see

• doc/paper.pdf (A Hitchhiker's Guide to Reinventing a Prolog Machine)

and

• tutorial at VMSS'16

for some motivations for this and justification of implementation choices

a swi prolog script first compiles the code to a pl.nl file, than Main calls stuff in Engine which loads it in memory and runs it

Notes

primitive types:

• int
• ground
• var (U+V)
• ref
• array

see main code (~1000lines) in Engine.java

• pl2nl.pl compiles a .pl file to its .nl equivalent, ready to run by the java based runtime system

• natint.pl emulates its work by compiling back .nl files to Prolog clauses

todo:

• design a self-contained compiler
• convert to C for more effective benchmarking against C-based Prolog systems
• faster runtime: memory efficiency - eg recursive loop, LCO code to be mem_copied with ptrs to var/ref cells to be relocated?

some out of the box thoughts:

• no symbol tables: a symbol is just a (small) ground array of ints, and instead of a symbol table we would have a "ground cache" - that helps with better memory usege and also speed

• when a non-ground compound tries to unify with a ground, the ground term is expanded to the heap

• when a ground unifies with a ground - it's just pointer equality and when a var unifies with a ground, it just points to it - as if it were a constant

Enjoy,

Paul Tarau

August 2017