Re: Why doesn't -fno-common really work???

[Jim Wilson <wilson at specifixinc dot com>]

Jack Howarth wrote:
> Is this the presence of the common symbols in object code
> compiled in g77 with -fno-common really non-broken behavior?

-fcommon/-fno-common are C front end options.  They are not used by the 
C++ or Fortran front ends, and probably not by other front ends either.

-fno-common changes the behavior of C programs if you have the same 
uninitialized global variable defined in multiple files.  So if you have
tmp.c:
  int i;
tmp2.c:
  int i;
then the default (-fcommon) is to accept this without complaint, but if 
you use -fno-common, you get a linker error for a multiply defined 
variable i.  Some C compilers give -fno-common behavior by default.

These options are not appropriate for C++.  I forget the details, but it 
violates the C++ language standard.  Hence the C++ front end has its own 
option -fconserve-space which does something similar to -fcommon, but is 
not guaranteed to work.  The default for C++ is effectively -fno-common.

I don't know enough about Fortran standards to say whether these options 
are appropriate.  However, it seems to me that if you specified 
something is in a common block, then surely you meant for it to be 
common.  How else you would implement common blocks otherwise?  If 
common doesn't work with the Apple Mach-O linker, then that seems like 
an Apple problem.  But since Apple doesn't support fortran, and binutils 
doesn't support Mach-O, you might be out of luck, and may have to find a 
work around.
--
Jim Wilson, GNU Tools Support, http://www.SpecifixInc.com