Large file iostreams

[Martin Sebor]

Nathan Myers wrote:

> On Mon, Sep 01, 2003 at 09:42:40AM -0000, P?tur Run?lfsson wrote:
> 
>>Martin Sebor wrote:
>>
>>>>27.4.3.1 p2: The initial value of state() is not specified.
>>>
>>>That's not necessarily a defect, is it?
>>
>>Probably not, it's just annoying.
> 
> 
> No, it's a defect.  At least at one time there was text in the
> standard that required that stateT (and by extension mbstate_t)
> could be initialized with stateT(), and that matched the state
> at the beginning of a sequence.  I don't find that text now, 
> and it's essential.  Without it, all the multibyte stuff is 
> meaningless.

What I meant was that if the state of an fpos object that's not
explicitly initialized is not specified, it doesn't make objects
of the type impossible to use correctly with multibyte-encoded
streams. I.e., portable code must explicitly initialize the
state before using the fpos object:

     fpos<mbstate_t> p (0);    // state may be uninitialized
     p.state (mbstate_t ());   // initialize it here

That said, I can't think of any advantages of a fpos default
ctor with these tricky semantics, so tightening up the spec
to require that the default ctor default-initialize the state
member would be fine by me.

Martin