Re: guile bugs

[Dirk Herrmann <dirk AT ida.ing.tu-bs.de>]

On 16 Sep 1999, Jim Blandy wrote:

> > 1) SCM_BOOL_F should be 0 so it is false in C
> 
> If someone were to make this change and show me before and after
> benchmark results (see guile-modules/benchmarks), then that would help
> persuade me.

I think this should not be done, independent of any benchmark results.
The main reason for this change would be to make sure that SCM_BOOL_F is
not accidentally used as a boolean value.  If, instead, it would be
possible to guarantee that using a SCM value as a boolean value will lead
to a compile time error, everything is fine.

If I understand correctly from some previous mail, defining SCM as a
union would serve that purpose.

> > 2) SCM should not be a long, but instead be a pointer to undefined
> > struct as we discussed before.  This would prevent misuse of an integer
> > as a Scheme value.
> 
> I'd love to see this change happen.

If such a change is done, it might be done in a way that would let guile
compile independent of whether SCM is defined either as a union or as a
pointer to an undefined struct.

Then, we could have something like a "find compile time error"-mode by
defining SCM as a union.  This would only be used for developers, since it
would lead to performance penalties.  In all other cases, SCM would just
be defined as a pointer to an undefined struct.

example:

#ifdef FIND_COMPILE_TIME_ERROR_MODE
[define SCM as a union]
#else
[define SCM as a pointer to an undefined struct]
#endif

This way, 1) would become unnecessary.  However, it possibly requires more
effort when doing 2).  Still, even 2) will introduce an awful lot of
casting.

Best regards,
Dirk Herrmann