Re: C++ API backwards compatibility

[Andrew Johnson <anj@aps.anl.gov>]

Chris Larrieu wrote:
> 
> The comparison between an iterator and a const_iterator is not
> defined by the standard.  The only portable solution is to
> cast my_list.end() to a const iterator:
> 
>   if (it != (list<t>::const_iterator)my_list.end())
>     cout << "groovy" << endl;

The following should be portable:

  if (my_list.end() != it)

I suspect the Standard requires that a const_iterator can be implicitly
created from an iterator, although I don't know for sure (I can't find
very much about const_iterator in it, but I may be looking in the wrong
places).  In the above case, if no suitable comparison operator (cit !=
it) exists then the compiler will automatically convert the RHS to cit and
use (cit != cit).

> But, both g++ and MS VC++ quietly permit the first comparison,
> while the HP aCC compiler specifically disallows it, specifically
> because it's undefined.  You might take this in one of two ways:
> either as evidence that HP is behind the times, or that HP avoids
> supporting de facto standards.

My current list implementation allows either order in comparisons, but I
could take out the (it == cit) operations if you really think this is
necessary.  I'm not sure that we'll be switching over to vendor's list
classes for quite a long time though - Tornado 2 is the worst performer,
and I can't see that being updated anytime soon.

- Andrew
-- 
Every great idea appears crazy to start with.