Re: FIX_REGISTER_CHECK

[Gavin Romig-Koch <gavin at cygnus dot com>]

Jim Wilson writes:
 > The patch seems incomplete, in two different ways.
 > 1) You are adding an error message for invalid -fcall* options, but you
 >    only do so for two registers.  You don't give errors for all invalid
 >    options.
 > 2) You are adding an error message for invalid -fcall* options, but you
 >    only do so for one target.  You don't give errors for all targets.
 > Both of these problems seem undesirable to me.
 > 
 > If it is really only the stack pointer and frame pointer that you
 > are worried about at this time, then why make the patch MIPS specific? 

I didn't write versions of FIX_REGISTER_CHECK for all the other
ports because I don't know the list of invalid registers for
each port.  I only added fp and sp to the mips version of 
FIX_REGISTER_CHECK because they were the only ones I was sure
of.  My intent was to provide a place maintainers who knew 
the list for a particular port could encode it.  

It was not obvious to me that fp and sp were invalid for -fcall*
for all ports.  If they are, then yes, this error message
could be added in a machine independant way, but I think the
ability for a port to list the invalid registers is worth the
cost of the new target macro.

I considard using CONDITIONAL_REGISTER_USAGE for this as Jeff
suggests, but at that point we no longer have the name for the
register that the user specified in the -fcall* option.  There
does not seem to be a way to generate the name of a register
suitable for use in error messages, and no way to ensure that
the generated name clearly points back to the erronious -fcall*
option.  These problems are solveable, but adding a new target
macro still seems to be a better solution.

After rethinking all this, I still believe my patch is the 
best solution, but I don't feel strongly about this, and
if the above comments don't change your mind, I'll rewrite
the patch whatever way you think is best.  

                                          -gavin...