I think that introducing MultiPassInputIterator isn't the right solution. Do you also want to define MultiPassBidirectionnalIterator and MultiPassRandomAccessIterator ? I don't, definitly. It only confuses the issue. The problem lies into the existing hierarchy of iterators, which mixes movabillity, modifiabillity and lvalue-ness, and these are clearly independant.
The terms Forward, Bidirectionnal and RandomAccess are about movabillity and shouldn't be used to mean anything else. In a completly orthogonal way, iterators can be immutable, mutable, or neither. Lvalueness of iterators is also orthogonal with immutabillity. With these clean concepts, your MultiPassInputIterator is just called a ForwardIterator.
Other translations are:
std::ForwardIterator -> ForwardIterator & LvalueIterator
std::BidirectionnalIterator -> BidirectionnalIterator & LvalueIterator
std::RandomAccessIterator -> RandomAccessIterator & LvalueIterator
Note that in practice the only operation not allowed on my ForwardIterator which is allowed on std::ForwardIterator is &*it. I think that &* is rarely needed in generic code.
reply by Jeremy Siek:
The above analysis by Valentin is right on. Of course, there is
the problem with backward compatibility. The current STL implementations
are based on the old definition of ForwardIterator. The right course
of action is to get ForwardIterator, etc. changed in the C++ standard.
Once that is done we can drop MultiPassInputIterator.
Copyright © 2000 | Jeremy Siek, Univ.of Notre Dame (jsiek@lsc.nd.edu) |