Re: reworked long_call heuristic patch.

["Dmitri Makarov" <dim at windriver dot com>]

Hello Nick,

> type.  Instead change the name of the function!  This is how other
> ports solve this problem.  (For example see the PowerPC or the V850
> ports).  What they do is encode attribute like information in the
> function name by appending an at sign '@' followed by some special
> codes.  There are macros that can then be defined to strip this
> encoding off the name before it used by the compiler, so that they do
> not affect the rest of the compilaiton process.

I'll do it the way you suggest doing this.  It's no problem.  I still think
the situation with multiple use of the same type nodes is an open
issue.  My assumption __attribute__ ((x_call)) will not work correctly
given the type nodes are used by more than one decl nodes.  It would
help if someone could comment on why the type nodes are reused.
Is it just not to waste heap memory or there are other reasons..?

Another way of implementing the heuristic might be to use decl_attributes
rather than type_attributes...

Thank you for the feedback.  I'm reworking the patch in accordance with
you comments.

Regards,
Dmitri.