PATCH for libstdc++/2211 (some discussion held on gcc-bugs)

[Benjamin Kosnik]

Ok folks, I'll be honest: I've not really been following this too 
closely. Loren's patch looks fine, and if it doesn't cause any new 
regressions should be checked in. 

-benjamin

> I concur except the C++ IO buffer size is set "(much) less than optimally"
> for most hosts even when C++ IO is to be unsynchronized with C IO.

Yes this is a vestige from older, more significant problems. Your patch 
seems fine. Here's a couple of comments:

>     ios_base::Init::_S_ios_create(bool __sync)
>     {
>       int __bufsize = __sync ? 0 : static_cast<int>(BUFSIZ);
> +     int __in_bufsize = __sync ? 1 : static_cast<int>(BUFSIZ);

you might want to just re-name to __out_bufsize and __in_bufsize.

> + #if _GLIBCPP_AVOID_FSEEK
> +     __in_bufsize = 1;
> + #endif
> + 

should probably add a comment as to why this is the case