[libgfortran,patch] Widen default formats for formatted output of floating-point numbers

[FX Coudert]

> I haven't fixed the case of real(kind=16) yet, because it's more  
> complicated there (see, once again, Dominique's testing in the PR).  
> It seemed that there is no reason to wait for the real(kind=16)  
> issue to be resolved to check in that patch, so I submit it here  
> for review.

Further analysis revealed that the real(kind=16) format is wide  
enough (at least on ppc-darwin); the I/O of long doubles in the  
system libc seems completely rotten, though, and I will thus have to  
XFAIL the testcase.


PS: if someone is interested in donating sparc-solaris access, that'd  
be welcome.

FX