opless / oldpas

prototype for retro pascal, based on p5 pascal compler

Introduction

This repository contains oldpas, a compiler for ISO 7185 standard pascal, which will target a small forth-style dual-stack virtual machine.

The purpose of this project is to prototype retpas, a planned compiler for a new pascal dialect called retro pascal.

The philosophy here is twofold:

1. Separate the work of designing the retro pascal language from the design of its compiler and virtual machine.
2. Start and end with a tested, working system, keeping it working at each step along the way.

Therefore, oldpas is not really a development project but rather a refactoring project.

Origins

The prototype is derived from the public domain P5 compiler.

P5 is a member of the pascal-p series of reference compilers, created originally by Niklaus Wirth and his students at ETH Zurich, documented in detail by Steven Pemberton and Martin Daniels, and updated to the ISO 7185 standard by Scott A. Moore.

P5 performs single pass compilation of pascal source code, meaning that input files are scanned from top to bottom and code is generated immediately for each language construct immediately as it parsed. The generated code is an idealized assembly language called p-code, which is then assembled and run on a small interpreter (virtual machine).

(Planned) Differences from P5

Although this prototype is derived from P5, the two compilers will bear little resemblence to each other.

The word derived here is meant in the mathematical sense: our approach is to start with the working P5 compiler and, through series of meaning-producing transformations (or refactorings), produce a completely different, yet logically equivalent compiler.

In contrast to P5, oldpas will:

• [ ] Separate lexical analysis, parsing, code generation, assembly, and execution into logically distinct modules.
• [ ] Introduce an intermediate abstract syntax tree on which optimizations may be performed.
• [ ] Replace the hand-coded recursive descent parser with a generic, data-driven parser.
• [ ] Replace the traditional single-stack virtual machine with a dual-stack architecture inspired by forth.
• [ ] Introduce a pre-processing step for literate programming, described in the Literalization section below.

Virtual Machines

In addition to the derived dual-stack virtual machine written in pascal and explained above, this project will produce two virtual machines capable of running inside a web browser:

• [ ] p5vm.cf - a direct port of the original P5 virtual machine, written in the coffeescript language for clarity.
• [ ] rpvm.js - an implementation of the target dual-stack machine, optimized for maximum performance in modern web browsers.

Literalization

To perform the above transformations in a safe and step-by-step manner, we will use WEB, the original literate programming system, created in 1976 by Donald E. Knuth.

WEB is programming language that closely resembles pascal, but provides a flexible macro facility and the ability to mix the pascal with richly formatted documentation, written in TeX.

The key feature of a literate programming tool is that it allows you to present the source code in whatever order makes it easiest for humans to understand. Thus, WEB consists of two programs:

• TANGLE translates a WEB file and generates raw pascal source code in the order the pascal compiler requires.
• WEAVE translates a WEB file to pure TeX, adding section numbers, cross references, pretty printing (a kind of syntax highlighting), an index and table of contents, etc. TeX can then translate this into a device independent representation for publication or display.

Since both of these programs are themselves written in WEB, a pre-generated TANGLE.PAS is included for bootstrapping.

(Planned) Differences from WEB

WEB predates the ISO pascal standard, was intended to compile on a wide variety of mutually incompatible pascal dialects, some of which had not yet even adopted the ASCII character set.

• [ ] Remove code for non-ASCII text encodings (with the possible exception of UTF-8)
• [ ] Update the syntax to ISO 7185 standard pascal.
• [ ] Remove code to support nonstandard pascal dialects.
• [ ] Consolidate the parsing system with the new generic parser from oldpas.

Relationship to Other Pascal Implementations

Any compiler that supports the ISO standard should be able to compile and run oldpas.

In particular, oldpas is compatible with fpc, the excellent free pascal compiler, which can produce native code for a wide variety of processors.

Although oldpas cannot run code written for delphi, fpc can, and thus Borland-style pascal and ISO pascal can be combined in the same program.

The planned retro pascal language will introduce a variety of syntactic changes, but is intended to remain compatable with code written for delphi by way of free pascal’s ISO mode.

Therefore it is likely that retpas will simply add a new parser and a series of tree transformations to produce oldpas-compatable abstract syntax trees. If this is the case, then retpas will simply become a frontend for oldpas, and a pretty-printer will be added to generate ISO pascal code from the internal representation.

Finally, the refactored virtual machine is designed to be extensible in free pascal, so that both ISO and retro pascal can be used as a scripting language in free pascal applications.