AIX I/O failures

[benjamin kosnik]

> My goal is very simple:
> 
>   Get V3 working on AIX.
> 
[snip]
> If my "opaque" hacks cause people to be sufficiently appalled that
> they do something cleaner, and thereby satisfy the goal, I will be not
> all embarssed -- even if not one line of the code I wrote remains! :-)

Right.

That's why I'm for this, if it doesn't break other platforms. It
doesn't really matter, we can clean it up later. We can't do the
istream/ostream hacks suggested in the orignal ILM post because this
will not be standards-conformant, initializing with a null buffer
actually sets bits and can't be ignored. And we have to do something
to actually be correct and not just lucky.

It's my hope that at some point we can do this cleanly, in a way
that other people using static data members in c++ can emulate and
understand. That's kind of one of the points of a standard library
IMHO but for now we can gloss over it.....

FWIW, I think g++ needs a FAQ entry or explict user-level
documentation on how to deal with static initialization portably
across platforms. 

-benjamin