Severe regression in autoconf 2.61

There seems to have been a bit of a brain cramp upstream :-(.
Previously, AC_FUNC_FSEEKO did this to test if fseeko was available:

return !fseeko;

Now it does this:

return fseeko (stdin, 0, 0) && (fseeko) (stdin, 0, 0);

Unfortunately, that gives the compiler enough of a syntactic clue
to guess that fseeko is probably an undeclared function, and therefore
*it will not error out*, only generate a warning, if it's not seen
a declaration for fseeko.

The proximate result of this in our HEAD is that configure fails to
detect that _LARGEFILE_SOURCE is needed, resulting in a rather broken
build on platforms where that really is needed. I had mis-blamed this
on Bruce's recent NetBSD/BSDi hack, but I think it's been there since
we installed 2.61 autoconf. I suspect that in fact the problem Bruce
was seeing was due to this very bug, and that what we need to do is
revert his BSD-specific patch and find a different solution.

regards, tom lane