Hello all! This is the off-topic section, right? ;)

I may be wrong but it seems to me most of the compilers at least most of the great ones are written by a very small pool of people. That tells me it's probably pretty hard to do a good job or everybody would be doing it. Anyway it's pretty obvious Walter knows what he is doing.

My question is how to get started with compiler writing. My background is
I'm a guy who has written systems software in assembler for a long time on
large IBM systems. I'm not an academic and I'm pretty crappy at math. Do I
really need to understand finite automata (did I even spell that right?!)
and graph theory and technical topics like that to be able to write a
good compiler? I'm fascinated by this topic but know nothing about it. I
haven't been able to find any general, practical guides to writing compilers
that aren't based on x86 and C or Java, all stuff I really don't know about
and honestly am not all that interested in knowing about. I would like to
use the skills I have and write a compiler for the platform I use in my
job. In other words, I'll never be any competition for Walter or the C/C++
market but you already knew that. ;)

Walter, have you written any books on the topic and can you recommend any practical guides? Would you say a little about your background and how you got started writing compilers? The whole topic is interesting to me.

Thanks,

Mike

On Friday, 11 February 2011 at 10:44:02 UTC, mike@domain.invalid wrote:
> Hello all! This is the off-topic section, right? ;)
>
> I may be wrong but it seems to me most of the compilers at least most of the
> great ones are written by a very small pool of people. That tells me it's
> probably pretty hard to do a good job or everybody would be doing it. Anyway
> it's pretty obvious Walter knows what he is doing.
>
> My question is how to get started with compiler writing. My background is
> I'm a guy who has written systems software in assembler for a long time on
> large IBM systems. I'm not an academic and I'm pretty crappy at math. Do I
> really need to understand finite automata (did I even spell that right?!)
> and graph theory and technical topics like that to be able to write a
> good compiler? I'm fascinated by this topic but know nothing about it. I
> haven't been able to find any general, practical guides to writing compilers
> that aren't based on x86 and C or Java, all stuff I really don't know about
> and honestly am not all that interested in knowing about. I would like to
> use the skills I have and write a compiler for the platform I use in my
> job. In other words, I'll never be any competition for Walter or the C/C++
> market but you already knew that. ;)
>
> Walter, have you written any books on the topic and can you recommend any
> practical guides? Would you say a little about your background and how you
> got started writing compilers? The whole topic is interesting to me.
>
> Thanks,
>
> Mike

http://en.wikipedia.org/wiki/Dragon_Book

On Friday, 11 February 2011 at 10:44:02 UTC, mike@domain.invalid wrote:
> Hello all! This is the off-topic section, right? ;)
>
> I may be wrong but it seems to me most of the compilers at least most of the
> great ones are written by a very small pool of people. That tells me it's
> probably pretty hard to do a good job or everybody would be doing it. Anyway
> it's pretty obvious Walter knows what he is doing.
>
> My question is how to get started with compiler writing. My background is
> I'm a guy who has written systems software in assembler for a long time on
> large IBM systems. I'm not an academic and I'm pretty crappy at math. Do I
> really need to understand finite automata (did I even spell that right?!)
> and graph theory and technical topics like that to be able to write a
> good compiler? I'm fascinated by this topic but know nothing about it. I
> haven't been able to find any general, practical guides to writing compilers
> that aren't based on x86 and C or Java, all stuff I really don't know about
> and honestly am not all that interested in knowing about. I would like to
> use the skills I have and write a compiler for the platform I use in my
> job. In other words, I'll never be any competition for Walter or the C/C++
> market but you already knew that. ;)
>
> Walter, have you written any books on the topic and can you recommend any
> practical guides? Would you say a little about your background and how you
> got started writing compilers? The whole topic is interesting to me.
>
> Thanks,
>
> Mike

this historical S.O question is highly related to the topic.

http://stackoverflow.com/questions/1669/learning-to-write-a-compiler

Hi Mike

First, there are some basic questions like "Do you mean a 'compiler' or something that translates source code and let's you execute it (e.g. interpreter)?", "Is your focus on creating a language and somehow run programs in it or is your focus on writing a compiler (for whatever language, maybe pascal)?", or "What is the platform that it should run on?" and others.

Brutally simplifying it, there are two major aspects. First, "reading the source code" (of your language) and creating an abstract representation, typically an AST tree. Second would be the generation of some kind of code. Depending on what you want/need you might get away without a lot of math.

If it's just for the fun of giving in to a fascination you might prefer to go for an interpreter, possibly using an existing VM.

If you're dead serious you might have a look at the LLVM project. They have a lot to offer, e.g. a very well designed "meta assembler language" that their engine groks and then, another big plus, your language can be both compiled and interpreted on a (increasingly large) variety of architectures.

Walking the whole chain all alone will quite probably not worth it unless one has (like Walter Bright) loads and loads of know how and experience and wants to create a full fledged product for a competitive environment (Which, btw, does not mean LLVM is just for hobbyists; it's definitely not).

One hint I'd like to give is that you shouldn't enthusiastically come up with a lot of desirables for your language but always keep in mind that - and how - those can be implemented and what trade offs are involved.

If you can and want to describe in more detail what you have in mind, you might get more, more concrete and more useful tips.

Good luck

A+ - R

On Friday, 11 February 2011 at 10:44:02 UTC, mike wrote:

> My question is how to get started with compiler writing.


I remember  way back, late 1970s, toying with an interpreter from scratch based on a short article in Byte.  Might have been early 1980s. I don't recall what it was called.  It was probably in a regular section of the magazine called Programming Quickies.  Google isn't coming up with much, but there is this index to old Byte issues:  http://www.devili.iki.fi/library/publication/10.en.html

This, I think is different, but also a very tiny interpreter - WADUZITDO appearing in Byte in 1978. Someone put it up on a website - http://waduzitdo.org/

Working with tiny languages and tiny compilers or interpreters is a great way to start, to figure out the fundamental workings.  Then add on new features.

When it grows and gets messy, it's time to read the famous Dragon Book http://en.wikipedia.org/wiki/Principles_of_Compiler_Design  or study source for a compiler for any language that isn't C++.  Languages that don't make a big deal about Objects, and don't try very hard to be cross-platform, so you can concentrate on the compiler workings of language parsing and code generation.  Oberon is good, Modula-2 or Modula-3, Pascal, and other languages which typically aren't popular today.   Peek into Java or Ruby if you want OO - these have nicer compiler innards than gcc or clang or python, but do note there are big differences between languages with just runtime libraries (like C) and those with complete runtime engines like .net, java's JVM.  Hoo boy, that's a whole 'nother nest of complexity to become familiar with.

An interpreted language with a very simple working principle is TCL. It gains power through modules or extension or whatever - it's been a while. But the inner core is very clean and simple. (Or was in older versions.)

One other language that's worth mentioning and studying to learn about interpreters, old but still in use in a sense: Forth.  It has a unique way of working. Though different most most other languages, it's educational for understanding interpreters in general. Byte had a great special issue covering it.  Forth itself isn't seen much these days, but it's the basis of Postscript, and therefore used by PDF documents.

On Friday, 11 February 2011 at 10:44:02 UTC, mike wrote:
> Hello all! This is the off-topic section, right? ;)
>
> I may be wrong but it seems to me most of the compilers at least most of the
> great ones are written by a very small pool of people. That tells me it's
> probably pretty hard to do a good job or everybody would be doing it. Anyway
> it's pretty obvious Walter knows what he is doing.
>
> My question is how to get started with compiler writing. My background is
.....
> Mike

Since you are the "Assembly guy", you could look at:
http://www.nand2tetris.org/course.php

Their projects:
High-Level Language Project 9
Compiler I: Syntax Analysis Project 10
Compiler II: Code Generation Project 11
could be of interest as they cover it from the ground up.

Also, being "close to the metal" you could learn Forth which has
an interpreter and assembler and these fit in under 8k RAM i.e.
very small and very tightly coupled to the underlying hardware.

JonesForth.S and JonesForth.F are very well documented text files
(which you could assemble for a working system or just read and
understand them).

Download locations are various (google search), eg:
http://www.eecs.wsu.edu/~hauser/teaching/Arch-F07/handouts/jonesforth.s.txt
and for Jonesforth.F
https://code.google.com/p/subtle-stack/downloads/detail?name=jonesforth.f.txt
will do.

Happy compiler writing,
Mark Barton