Should basic_*stream::is_open() be const?

[Nathan Myers]

On Fri, Aug 13, 2004 at 06:48:22PM +0200, Gabriel Dos Reis wrote:
> | >"Vladimir Merzliakov" <wanderer@rsu.ru> writes:
> | >
> | >| class Parser {
> | >|    std:ifstream m_in;
> | >|    public: bool isOK() const { return m_in.is_open(); }
> | >| };
>  ...
> Yes, it does not compile, I know and I see it. Similarly, I can post
> many function declarations along the same lines.  The point of the
> question, was in which actual context do you need that thing to be
> const, i.e. is it actually called.

This Parser might be derived from an abstract base that knows
nothing about streams, yet be obliged to implement its interface.
Certainly it is not unreasonable for the base class to have made 
isOK() const.  (Whether it is reasonable to use an ifstream, and 
not a filebuf, in a parser is beside the point.)  

When implementing such a derivation interface, now, you have no 
choice but to say "const_cast<Parser*>(this)->m_in" for access from 
those const members.

This issue came up in committee lots of times, and evoked Gaby's 
reaction (from me, too), several times before examples like this made 
the difference.

Nathan Myers
ncm-nospam@cantrip.org