[merd-devel] Re: welcome to merd-devel

[Pixel <pi...@ma...>]

"Tommy Olesen" <to...@so...> writes:

> > soundness i don't know yet. Maybe some things will have to rely on
> > runtime checks a la Java. Hopefully it will happen only in weird
> > cases.
> 
> This was the compromise I was willing to take too, when I was 
> looking at this last (though I would perhaps prefer to force 
> the user to write the type check in the code rather than insert
> it automatically, e.g.,
> 
>   x : bool | int  (sorry, it has been a while since I read about
>                    merd, so I have forgotten the syntax)
>   y : int = cast<int>(x) + 42

for this example, this is a down-cast, and must be explicit.


one day it could be:

  dynamic_cast(t) =
      (x !! t) -> x
      _ -> raise(Dynamic_cast)
  
  #=> dynamic_cast !! t -> x -> x !> Val(t); Val(t)
  
  to_Uint = dynamic_cast(Uint)
  #=> to_Uint !! x -> x !> Uint ; Uint

though i'm not sure about the syntax of (x !! t) yet

>   
> This way the user knows where he is paying a run-time cost).
>  
> > Since the goal is not performance (well Perl/Python/Ruby performance
> > is targeted), keeping many tags at runtime is no big deal.
> 
> I agree.
> 
> > > I saw some of your code (did not read it in detail).  Have you
> > > considered writing some type inference rules for the type system?
> > 
> > you mean a more formal version than what's done in the code? Well not
> > yet. The type inference rules are much based on the
> > make-it-work-as-expected goal. I hope to consolidate things in the
> > code, since currently they are many duplicated rewriting rules.
> 
> Well, I suppose there is nothing more formal than code :-)  Inference
> rules give you a way of specifying what the rules should do and thus
> some hope of finding bugs in the code, but of course it only shifts
> the problem to the specification: how do you ensure there are no bugs
> in the inference rules?  

i don't :-(

for example I had a(b) | a(c)  ==> a(b | c)  which is not true
(http://merd.net/types.html#co_and_contravariance)

there may be other bugs awaiting :)