awk (was: Re:)

[David Given <dg@...>]

Miles Bader wrote:
> "steve donovan" <steve.j.donovan@gmail.com> writes:
> > (g)awk is a lovely language. Look at those old books by Jon Bentley and
> > you'll see the oldtimers prototyping things in AWK. Was probably the
> > first scripting language, apart from BASIC and shell ;)
>
> Yes, I agree, awk is wonderful.  In many ways it's a far better
> language, despite its limitations, than most of its purported
> replacements (Lua excepted of course :-).

I think I can definitively pontificate on awk --- I wrote a compiler in 
it once. A full reverse-descent strongly typed bytecode-targeting 
language. Thinking back now, I must have been insane.

The biggest problems with awk: firstly, everything's a string:

1 == "1"     true
1 == "1.0"   false
1.0 == "1.0" false

This makes doing anything involving maths really unpleasant. You keep 
having to do 'n + 0' to force type conversion.

Secondly, the '' operator is string concatenation:

"foo" "bar"  -> "foobar"
1 "bar"      -> "1bar"
1 1.0        -> "11"

...which as you can see interacts nastily with the first problem. This 
can lead to some *ludicrous* typos.

Thirdly, no local variables; you fake them instead by using extra 
function arguments:

foo(arg1 arg2 var1 var2) { ... }

...which can obviously go terribly wrong if you pass too many arguments 
into a function.

Fourthly, no arrays.

This is a big one. There are associative arrays, but both the key and 
the value must be a string. This makes it rather hard to store complex 
data structures. There's syntactic sugar for multidimensional arrays of 
the form array[foo, bar], but that's identical to array[foo SUBSEP bar] 
where SUBSEP is a string containing a non-printing character... which 
makes *enumerating* such arrays highly entertaining!

That said, the entire compiler for my language came out at 1611 lines of 
highly structured and fairly readable code; and it ran astonishingly 
fast, too. When I rewrote the compiler in itself, compiled it with the 
awk bootstrap compiler and tried the result, it turned out to be 
considerably slower than the awk version (at which point I largely lost 
interest).

awk is a great language for things within its problem domain --- for 
text parsing, it's still considerably more elegant than Lua. It's got 
scaling problems, but for things that don't hit those limits, it's well 
worth checking out. To me, it forms a very nice middle ground between 
sed and perl (which I can't stand). The C-like syntax is nice and 
familiar, and the built-in support for things like regexps are 
first-rate. It's an ideal example of why domain-specific languages can 
be good.

But I wouldn't recommend writing a compiler in it.

If anyone actually *cares*, my doomed toy language Mercat and both 
versions of the compiler can be found here:

http://www.cowlark.com/mercat.html

...in the core021.zip package. (core030.zip is a later version where the 
language has acquired features that no longer work with the awk compiler.)

--
David Given
dg@cowlark.com