Re: [boost] [utility] prior(it, n) for n being an unsigned type

Subject: Re: [boost] [utility] prior(it, n) for n being an unsigned type
From: Andrey Semashev (andrey.semashev_at_[hidden])
Date: 2014-06-23 04:29:27

On Monday 23 June 2014 08:40:29 Adam Romanek wrote:
> Hi!
> 
> This is a re-post from boost-users mailing list, as I didn't get any
> response there.
> 
> I've just finished investigating an issue in my code which led it to an
> assert/abort in certain cases. It turned out it was caused by the use of
> boost::prior(it, n) where n was unsigned int with some non-zero value.
> 
> I found a question [1] on the mailing list back in 2004 which describes
> this issue. Additionally a solution to this problem was also posted in
> one of the replies [2].
> 
> My question is - why hasn't it made into the repository yet?

Probably because it got forgotten.

> The current
> behavior of boost::prior(it, n) when n is unsigned type is far from the
> one expected by its potential users so I think it should either be
> properly documented (warned) or fixed immediately.

I agree this should be fixed. Although it would probably be better to follow 
the standard interface:

template <class BidirectionalIterator>
BidirectionalIterator prev(BidirectionalIterator x,
typename std::iterator_traits<BidirectionalIterator>::difference_type n = 1);

I.e. the second argument should be accepted as a signed value and not as a 
template parameter.

I'll apply the change later if noone objects.