Enhanced type guard syntax]

[Martin J. Logan]

On Thu, 2003-09-18 at 17:24, Chris Pressey wrote:
> On Thu, 18 Sep 2003 22:02:26 +0200
> Joachim Durchholz <joachim.durchholz@REDACTED> wrote:
> 
> > Thomas Arts wrote:
> > > [...] there was a philosophy that all guards should be executable in
> > > constant time.
> > 
> > This philosophy doesn't work out IMHO.
> > Either, the guard is used as a selector. Then, if the guard is
> > disallowed, people are going to write the exactly same
> > non-constant-time code into the condition of an "if" statement, and
> > nothing is won. Or the guard is used as a precondition. In that case,
> > it shouldn't be a guard (but an assert statement or something -
> > particularly, something that can be switched off for production code).
> > 
> > Just my 2c.
> > 
> > Regards,
> > Jo
> 
> Sing it, brother!
> 
> A much nicer world would be:
> 
> 1) the programmer writes whatever code they want in the "when" test
> 2) the compiler analyzes that code
> 3) whatever parts of it that have side-effects are rejected

How could this be done at compile time while still using run time
function call binding?
If I say erlc modfoo.erl and it uses a guard that references code in
modbar.erl  erlc must be made aware of modbar.erl so it can look for a
"!" in that module. 

> 4) whatever parts of it are constant-time are used as the guard
> 5) whatever parts of it are non-constant-time are factored out into a
> "case" statement
> 
> But that sort of requires that a) compiler-writers are well paid and b)
> the language is optimized for maintainability over efficiency... which
> only ever seem to hold true in my private little world.
> 
> -Chris