[m-dev.] bug fix to frameopt

[Fergus Henderson]

Zoltan Somogyi, you wrote:
> 
> David and I just looked at the template in pragma_c_gen, and we
> couldn't quite figure out why for may_call_mercury pragma C code,
> the restore_registers was conditional on #ifndef CONSERVATIVE_GC
> but the corresponding save_registers wasn't.  We thought it was
> to prepare for the restore_registers that is done when calling
> Mercury from C, but we weren't sure. Whether we were correct or not,
> the reason should be documented in pragma_c_gen.m.

Hi,

You were correct.

I had a look at the documentation, and it seemed clear enough.
The reason for the save_registers() and restore_registers() _is_
documented in pragma_c_gen.m, and it does explain why the
restore_registers() need only be done #ifndef CONSERVATIVE_GC.

% <save live variables onto the stack> /* see note (1) below */
% {
%	<declaration of one local variable for each arg>
%	<assignment of input values from registers to local variables>
%	save_registers(); /* see notes (1) and (2) below */
%	{ <the c code itself> }
%	#ifndef CONSERVATIVE_GC
%	  restore_registers(); /* see notes (1) and (3) below */
%	#endif
%	<assignment of the output values from local variables to registers>
% }
...
% Notes:
%
% (1)	These parts are only emitted if the C code may call Mercury.
%	If a pragma c_code(will_not_call_mercury, ...) declaration was used,
%	they will not be emitted.
%
% (2)	The call to save_registers() is needed so that if the
%	C code calls Mercury code, we can call restore_registers()
%	on entry to the Mercury code (see export.m) to get the
%	right values of `sp', `hp', `curfr' and `maxfr' for the
%	recursive invocation of Mercury.
%
% (3)	The call to restore_registers() is needed in case the
%	C code calls Mercury code which allocates some data
%	on the heap, and this data is returned from Mercury
%	through C back to Mercury.  In that case, we need to
%	keep the value of `hp' that was set by the recursive
%	invocation of Mercury.  The Mercury calling convention
%	guarantees that the values of `sp', `curfr', and `maxfr'
%	will be preserved, so if we're using conservative gc,
%	there is nothing that needs restoring.

Of course if you have some concrete ideas about improving
this documentation, you're welcome to do so.

-- 
Fergus Henderson <fjh at cs.mu.oz.au>   |  "I have always known that the pursuit
WWW: <http://www.cs.mu.oz.au/~fjh>   |  of excellence is a lethal habit"
PGP: finger fjh at 128.250.37.3         |     -- the last words of T. S. Garp.