[boost] Re: Contracts and optimizing (was Circular log buffer)

From: David Abrahams (dave_at_[hidden])
Date: 2004-10-31 07:27:25

Darren Cook <darren_at_[hidden]> writes:

>>>I'm hoping Thorsten Ottosen's contract programming will make it into
>>>the standard so the compiler will understand asserts and be able to
>>>optimize away the redundant ones.
>> Compilers are already allowed to do that, and I predict no proposal
>> that requires it to be done will ever be accepted into the standard.
>
> "be able" falls somewhere between "allowed to" and "required to". An
> optimizer will have a hard job analyzing all the code in a function so
> see if any can be skipped. But if it can see a precondition is always
> satisfied then it can jump directly to the body of the function and
> skip the precondition.
>
> E.g.
>    void f(int x){
>    precondition{ x>=0 && x<=99; }	//1
>    ... //2
>    }
>
>    f(atoi(argv[2]));	//Jumps to f, at 1
>    f(34);	//Jumps to f, at 2

If f is inlined, I'm positive that compilers today can do that.

>    void g(int z){
>    precondition{ z>=10 && z<50; }
>    f(z*2);	//Can also jump to f at 2, as range is 20..98
>    }

And I'm pretty sure that compilers today can do that one.

> I don't think Thorsten's proposal requires optimizations like
> this. But I don't know enough about compiler writing to know if the
> above is ever likely to be implemented.

It is just as implementable for assert as it is with Thorsten's
proposal.

-- 
Dave Abrahams
Boost Consulting
http://www.boost-consulting.com