Functional-imperative programming

[0rthodontist]

If a function written in an imperative style does not set any variables other than local variables, and has access to no input other than its declared inputs (no keyboard input for example), then the function has no side effects. For example

Input n
int sum = 0
for i <-- 1 to n
  sum <-- sum + i
return sum

is consistent with functional programming--it has no side effects even though it is written in an imperative style.

Are there functional languages that make use of this--so that individual functions may be written in an imperative style while the whole program is functional in nature? And I don't mean that the language is primarily functional but allows some imperative programming as a special case that you should try to avoid unless absolutely necessary, I mean a language that checks basically imperative functions and verifies they do not have side effects. It seems like this would retain the advantages of functional programming while allowing many things to be implemented more simply.

 

[0rthodontist]

Also--physicsforums doesn't seem to be very active on computer science. Is there someplace that talks about computer science--as opposed to technical details of any particular product--and is very active/contains experts?

 

[Hurkyl]

What about doing it the other way? Take an imperative programming language, and write in a functional style.

In fact, the C++ standard template library sort of encourages that!

 

[0rthodontist]

That's basically what I had in mind: adapting an imperative language so that sections of code can be called functional.

From what I've read about it, the STL looks like it emulates the first-class-functions part of functional programming. What I'm really looking for is some language or compiler that has a built in mechanism that can certify functions as "no side effects" and "well defined."

 

[Hurkyl]

I doubt it will suit your purposes, but I think you can't have any side effects from C++'s template mechanisms.

 

[0rthodontist]

Could you explain how that works? (are we using the same definition of "side effect"? I mean it in the functional-programming sense of not producing any effect other than evaluating to the return value)

 

[Hurkyl]

A template metaprogram "runs" at compile-time. The only statements available to such a program are declarations. The only values accessible to a template metaprogram are those that can be determined at compile-time.

So, in that sense, there cannot be any side effects.


Of course, the whole point of a template metaprogram is to write C++ code; everything declared is subsequently accessible to the C++ program in which it resides... though I don't think that counts as a "side-effect", as you meant it.

 

[jim mcnamara]

Hurkyl -

I think he means idempotent functions. And in C++/C it certainly is possible to write code that has side effects. Opening a file is one example.
Because the file "metadata", like file pointers, are globally modifiable.

If I got what he meant.

 

[Hurkyl]

I was talking specifically about the template metaprogramming aspect of C++. You can't open a file with a template metaprogram... you can only use it to generate code that, when executed, will open a file.

(The template metaprogram executes at compile time)