Side-effect free containers for E

[Mark S. Miller]

At 05:46 AM 8/10/00 , Tyler Close wrote:
>For associative containers:
>
>         container += (key, value)
>
>expands to
>
>         container := container + (key, value)
>
>expands to
>
>         container := container with(key, value)
>
>I think this is a clean and accurate syntax.

You raise two sets of issues here.  The first has to do with our original 
discussion -- how shall containers behave and what syntax shall we use to 
manipulate them?  

The second you raise implicitly:  Since our overloaded operators are merely 
syntactic sugar for message names, why not allow the right hand argument of 
an overloaded operator to be an n-ary argument list? After all, if the right 
argument is a non-parenthesized expression, that already means the same 
thing as a 1-ary argument list.  Interesting.  I think this generalization 
is simple and seems simple.  I just tried it, and it does result in an 
unambiguous grammar.  From my latest Elmer session:

   ? e`x += (2, 3)`
   # value: e`x := x add(2, 3)`

   ? e` 2 + ()`
   # value: e`2 add`

And checking an edge condition:

   ? e`x ** y %% z`
   # value: e`x modPow(y, z)`

   ? e`x ** (y, z) %% w`
   # value: e`x pow(y, z) mod(w)`

   ? e`x ** y %% (z, w)`
   # value: e`x pow(y) mod(z, w)`

Only the double 1-ary case of ** combined with %% gets optimized into modPow.

Does anyone object to this generalization of the syntax rules?


         Cheers,
         --MarkM