Re: Goops & Valgrind

[Neil Jerram]

2008/9/12 Andy Wingo <address@hidden>:
> On Thu 11 Sep 2008 23:06, "Neil Jerram" <address@hidden> writes:
>
>> Are you sure?  Surely that would require a call somewhere to
>> scm_alloc_struct() with n_extra = 0, and I can't see any of those.
>
> I'm sure -- goops.c:1541 in master. Doesn't go through scm_alloc_struct
> at all.

Thanks, I see now.

>> Also, is Mikael right with his error #1?  I'm thinking not, because I
>> believe that instances are structs too, so surely it's OK to call
>> SCM_STRUCT_DATA (x)[...] on them?
>
> I can't recall the mail at the moment. Please reply if you want me to
> dig through this -- I'm happy to do so. But instances are structs, yes.
> Calling SCM_STRUCT_DATA (x)[] does work. You have to know how many
> fields there are, though -- you get that from the vtable.

Agreed.  So I think the right fix here is along the lines of your
second suggestion:

> #define SCM_NUMBER_OF_FIELDS(x) (SCM_STRUCT_VTABLE (x)[scm_si_nfields])

I propose specifically that we:

- remove the SCM_NUMBER_OF_SLOTS macro - because it's never been
right, so there can't be external code relying on it

- change scm_sys_fast_slot_ref and scm_sys_fast_slot_set_x to say

  i = scm_to_unsigned_integer (index, 0, SCM_SLOT (SCM_CLASS_OF (obj),
scm_si_nfields) - 1);

OK?  (There are way too many goops/struct macros already, so let's not
introduce another one!)

One last concern, though: I didn't understand what you meant by "would
probably have a different purpose".  (In:

> assumption. The other would be to use a different definition of
> SCM_NUMBER_OF SLOTS, which would probably have a different purpose:

)

Regards,
     Neil