[m-dev.] for review: nicer stack dumps

[Zoltan Somogyi]

> +		        fprintf(stderr, "\t%s:%s/%ld (mode %ld, %s)\n",
> +				entry_layout->MR_sle_def_module,
> +				entry_layout->MR_sle_name,
> +				(long) entry_layout->MR_sle_arity,
> +		                (long) entry_layout->MR_sle_mode,
> +				detism_names[entry_layout->MR_sle_detism]);

This is fine as long as we are not too worried about what we print for
automatically generated procedures. (The same goes for similar code
in the tracer.)

The rest of the diff looks OK.

There are two further changes related to this, but they can be in
separate checkins. One is to fill in the label name field of the
label table with a NULL unless profiling is defined. This will cut
down executable sizes significantly. The other is to give to this
stack dump utility the capability of the one in mercury_memory.c,
i.e. compression of the stack dump via run-length encoding.
I want to add a command to the debugger that will print the stack dump,
but don't want to annoy people by printing

	list:member/2 (mode 1, semidet)

2000 times instead of printing

	2000 * list:member/2 (mode 1, semidet)

just once. (This is a real situation.) Actually, it would be even better
if we recognized cycles of length > 1, up to a configurable limit, since
sometimes a deep recursion involves two or more procedures (I recall that
the parser has such code).

Zoltan.