PATCH: Avoid excessive flushing on lower-layer handle

[Loren James Rittle]

In article <20020423101922.1770f873.bkoz@redhat.com>,
Benjamin Kosnik <bkoz@redhat.com> writes:

>> [If we are willing to entertain a larger patch, other options exist.]

> You might as well post it if you've got it.

I can envision changing the few call sites of _M_really_overflow
(private, used for implementation) to allow complete removal of the
sync/flush to the lower layer within _M_really_overflow.  Callers that
really want sync instead may call sync() to get the heavier weight
behavior (it internally calls _M_really_overflow but also arranges for
the physical flush).

I am still tracking down an elusive linker problem with libjava on my
platform (with binutils 2.12 prerelease) thus I can't commit to doing
much more with libstdc++-v3 until after 3.1 release.

> I'm in the middle of testing this: I'll let you know linux results as I
> finish.