Re: Pointers to "why C behaves like that ?"

Roman Shaposhnick wrote:
>
> So what happens when all the signatures have the most general types in
> them ? My concern is -- how big the runtime hit will be.


If you leave the most general type in, the code will not be very efficient.
Example: if you have a function that's of type
      foo (Numeric): Numeric
then you'll probably have an inefficient foo() function.
However, if you call
      foo (5)
then the compiler can always generate a specialized, efficient variant
of foo (Integer).


I know that this doesn't work well with the standard division of labor
between compiler and linker. But you have similar problems even with
C++, so that's not a large factor in my eyes (YMMV).


> I still, can't believe that there won't be any, but that might be due
> to the lack of detailed information on Hindley-Milner typing.


Hindley-Milner typing is just a way of collecting type information. It
doesn't imply anything about the efficiency of the resulting program;
it's just a theory how far you can get without type information, if the
type system satisfies some basic properties (such as providing a common
supertype for any combination of types).


  > I will appreciate it if someone
> can give me any pointers to the basic papers on this topic.


Take a look at:
Michael I Schwartzbach: Polymorphic Type Inference.
It's available on the web (probably at Citeseer).


> P.S. And I still can't get a hold of the original Milner article,
> but I'll try the biggest library in the town today.


Original articles tend to bee disappointing; too much theory, little
practical value. (Which is OK, since seminal papers serve the purpose
of proving that the idea works; explanation and exploring the
practical consequences is what follows in the months and years after
the original article.) I found the Schwartzbach article very
readable.


Regards,
Joachim

Return to the comp.compilers page.
Search the comp.compilers archives again.