[v3] use AM_ICONV

[Benjamin Kosnik]

> Won't you need to apply this to the 4.1 branch as well?  I thought the fix
> for 28125 went in there too.

Hi Kaveh. Thanks for your patience with this.

I reverted the patch for 28125 on gcc-4_1-branch and gcc-4_2-branch. I'm 
waiting for your regression testers to get back to normal before trying 
anything tricky. Given the size of this patch I'm not super interested 
in putting it in gcc-4_1-branch, ever.

I'm not convinced that this patch, as-is, is going to be sufficient. 
Conceptually, it seems the right approach but I don't understand how 
AM_ICONV can be used for crosses, even though it clearly is being used 
that way in both libjava and gcc.

I see some breakage on the solaris10 results you posted. Any chance you 
could investigate? Also awaiting solaris8 and 9 results as that seems to 
be the real issue, not solaris10. I'm still not convinced I have the 
--disable-iconv-prefix thing fixed to your satisfaction.

http://gcc.gnu.org/ml/gcc-testresults/2007-01/msg01157.html

Does something need to be wrapped with _GLIBCXX_HAVE_ICONV to get these 
to work? Or are the testcase assumptions incorrect? Not quite sure where 
the breakage is: more details would be helpful. Also I see the darwin 
tester has new issues.

http://gcc.gnu.org/ml/gcc-testresults/2007-01/msg01153.html

best,
benjamin