On Fri, May 14, 2010 at 18:12, Jerry DeLisle<jvdelisle@verizon.net> wrote:
On 05/14/2010 07:57 AM, Steve Kargl wrote:
Well, there are plans to break ABI with the array
desrciptor work.
I wonder, would it be feasible to avoid breaking the ABI by using
access macros for descriptors and then generate both old and new
descriptor functions?
Based on a quick look, the only function in the I/O library that uses
the descriptor is tranfer_array, so the problem is restricted to the
intrinsics, most of the intrinsics with array arguments are generated
from the m4 macros.
There's also the desire to reduce
the number of functions in the library with the
use of type casting in the front-end.
Yes. I think this would be one good candidate for that. I donÂt see
much need for having stop codes> 2**31, and on POSIX the return code
is between 0 and 256 in any case.
So, if you break the ABI, you need to bump the
library version number. Janne, is the right?
The patch bumps the library version when adding the new error_stop_numeric
function.
No, it adds a new version node to the current major version. If you
change the behavior or ABI of an existing function you need to bump
the .so major version (in libgfortran/libtool-version IIRC).