Re: To get things started...

Bart Lateur wrote:
> 
> On Tue, 21 Nov 2000 17:24:49 +0000, Simon Cozens wrote:
> 
> >I really *would* recommend Aho, Sethi, Ullman, "Compilers: Principles,
> >Techniques and Tools".
> 
> AKA "The Dragon Book". You're not the only one to mention this book on
> this list.
> 
> IMO, this book is really thick to crawl through, and in the end, it's
> all just theory, you still won't be able to build a compiler, or an
> optimizer in case you already have a compiler. Heh. In short: are there
> any more *practical* "how do I build my own compiler" books, that people
> can wholeheartedly recommend?

There's the LCC book. Practical Compiler Construction, or something?
Should be easy to search for. It uses literate programming to describe a
complete C compiler. But on the scale from theory to practice, it's too
far on the practice end to be "how do I build my own compiler" -- the
stuff they talk about often doesn't translate to other compilers. It's
really a case study.

I've written a complete compiler using the Dragon Book and the lectures
from the class it was for. It was a complete compiler, from input text
to MIPS assembly code output. But I wouldn't recommend the Dragon Book
for practical purposes either. It's mostly useful to see what problems
exist, the general approach with which they're traditionally addressed,
and what terminology to use in talking about them. (And my experience
isn't all that relevant, considering it was a class project with
provided skeleton code and a toy language.)

I've also implemented my own LALR parser from scratch, using only the
Dragon Book as a reference, and I'd say it was more of a hindrance than
a help. Too theoretical, too much magic hidden in the math. Sends you
off on the wrong direction implementation-wise. I got more benefit from
my memories of the yacc source code that I read through about five years
ago, even though I intentionally avoided copying yacc's algorithm.

For other recommendations, there's always gcc's source code. But it's
too hairy to be comprehensible. I've heard very good things about a
compiler written at Stanford. I forget the name now. But apparently many
people prefer to modify it rather than dealing with gcc's complexity.
Might be an informative read.