current cygwin failure analysis

[Benjamin Kosnik]

> In this case it looks like just an argument ordering problem.  It
> would have said "unable to find -liconv" if that was the case, but
> instead what's happening is that the linker processes -liconv before
> it's seen any objects that use functions from it, and so it doesn't
> pull in any symbols from it -- even though -liconv is a shared lib.

Ah ha. Thanks Brian.

> (This is one of the differences between PE and ELF that comes up
> again and again without fail in porting to Windows.)  Simply
> reordering the command ought to work fine.

Hmm. I see some gcc/testsuite/charset files require dg-require-iconv as
well, but they aren't failing on cygwin. Why?

What do those compile lines look like?

Is this an issue with dg-require-iconv, or is libstdc++ using it
incorrectly?

-benjamin