bug in library standard 23.2.1.3/4

[Martin Sebor]

AlisdairM wrote:
> Steve LoBasso wrote:
> 
>> The problem is the standard doesn't say that end() will become
>> invalid. This is probably why the debug implementation doesn't catch
>> the problem [ie. compiling with -D_GLIBCXX_DEBUG].
> 
> <delurk>
> 
> If I understand correctly, the standard says very little about end
> iterators, wrt invalidation.  THe weasel-wording is that most of the
> wording describing when an iterator is invalidated talk about iterators
> in the sequence, or similar, but end iterators are
> one-past-the-end-of-the-sequence, but not actually in the sequence, and
> so are not covered.

There are three kinds of iterators: dereferenceable, valid, and
singular. A past-the-end iterator is a valid non-dereferenceable
iterator into the sequence. (I couldn't find any text that says
otherwise.)

> 
> An implementation is usually free to invalidate an end iterator as it
> feels, including never if it so chooses.

Since end() is a valid iterator into the container and the text
clearly says that erasing at either end of a deque invalidates
only iterators pointing to the erased elements, the requirement
is that all other valid iterators must remain valid after the
call. I'm not saying that's what we want or that it's even
implementable, just how I read it.

Martin