Re: [boost] STL containers with contracts, any interest?

From: Douglas Gregor (doug.gregor_at_[hidden])
Date: 2005-02-23 19:17:03

On Feb 23, 2005, at 10:34 AM, christopher diggins wrote:
> Christopher Diggins
> Object Oriented Template Library (OOTL)
> http://www.ootl.org
> ----- Original Message ----- From: "Douglas Gregor" 
> <doug.gregor_at_[hidden]>
> To: <boost_at_[hidden]>
> Sent: Wednesday, February 23, 2005 8:27 AM
> Subject: Re: [boost] STL containers with contracts, any interest?
> Hi Doug,
>
> 1) The proposed PwC approach would provide consistent behaviour across 
> STL implementations.

Isn't that more a property of the adaptor approach than PwC? You can 
pass any container to the adaptor, regardless of the underlying 
library, and it will perform the checks when it interfaces with that 
container.

At least two of the debug modes I'd mentioned (GCC libstdc++ and 
STLport) use wrappers/adaptors to implement debug-mode semantics, and I 
suspect the others do as well. Nothing actually ties the debug modes to 
a particular library except a few #defines and/or some namespace 
shuffling.

> 2) The contract verification classes would make it much easier to 
> write new classes which model STL containers and verify that they 
> exhibit correct behaviour.

Sure.

> Also consider the following code:
>
> template<Vector_T>
> void MyAlgorithm(Vector_T& x) {
>  // do stuff ...
> }
>
> If we get a hand-rolled vector type (instead of an STL vector) we have 
> no way of knowing whether or not it conforms to the expected contract 
> even if we can check it conforms to the correct concept. Using 
> contract classes we could instead write (using an imaginary 
> vector_contract_concept):
>
> template<Vector_T>
> void MyAlgorithm(Vector_T& x) {
>  #ifdef DEBUG
>    BOOST_STATIC_ASSERT(vector_contract_concept<Vector_T>::satisfied)
>  #endif
>  // do stuff
> }

I don't understand this part... it looks like a concept check (i.e., 
syntax), but above it sounds like you're saying that this verifies the 
contract (i.e., semantics). How can we check the semantics with a 
static assert?

        Doug