struct displacements

struct displacements

[Andy Wingo]

Hello Ludovic,

    -  GC_REGISTER_DISPLACEMENT (2*sizeof(scm_t_bits)); /* for the self data pointer */
    -  GC_REGISTER_DISPLACEMENT (2*sizeof(scm_t_bits)
    -                            + scm_tc3_struct); /* for the vtable data pointer */
    +  /* The first word of a struct is equal to `SCM_STRUCT_DATA (vtable) +
    +     scm_tc3_struct', and `SCM_STRUCT_DATA (vtable)' is 2 words after VTABLE by
    +     default.  */
    +  GC_REGISTER_DISPLACEMENT (2 * sizeof (scm_t_bits) + scm_tc3_struct);

I'm pretty sure the first case is still needed, for instance
reallocation. It is, as the comment indicates, "for the self data
pointer" -- the second word of the object, which normally points to the
third word of the object, but for reallocated instances points to the
third word of *another* object. See my article on redefining classes for
more info.

Can you change this?

Andy
-- 
http://wingolog.org/

Re: struct displacements

[Ludovic Courtès]

Hello,

Andy Wingo <wingo@pobox.com> writes:

>     -  GC_REGISTER_DISPLACEMENT (2*sizeof(scm_t_bits)); /* for the self data pointer */
>     -  GC_REGISTER_DISPLACEMENT (2*sizeof(scm_t_bits)
>     -                            + scm_tc3_struct); /* for the vtable data pointer */
>     +  /* The first word of a struct is equal to `SCM_STRUCT_DATA (vtable) +
>     +     scm_tc3_struct', and `SCM_STRUCT_DATA (vtable)' is 2 words after VTABLE by
>     +     default.  */
>     +  GC_REGISTER_DISPLACEMENT (2 * sizeof (scm_t_bits) + scm_tc3_struct);
>
> I'm pretty sure the first case is still needed, for instance
> reallocation. It is, as the comment indicates, "for the self data
> pointer" -- the second word of the object, which normally points to the
> third word of the object, but for reallocated instances points to the
> third word of *another* object.

As an excuse, the comment was a bit too terse for me to understand, but
I get your point now.  However, it’s still unclear to me how an ‘s’
field can point to something else than “self”.

> See my article on redefining classes for more info.

By ‘self’ here, did you mean what’s known as ‘data’ on the diagrams at
<http://wingolog.org/archives/2009/11/09/class-redefinition-in-guile>?

Thanks!

Ludo’.

PS: It would be really cool if this blog post were part of the “Data
    Representation” node of the manual.  :-)

Re: struct displacements

[Andy Wingo]

Hi Ludo' :)

On Mon 25 Jan 2010 14:44, ludo@gnu.org (Ludovic Courtès) writes:

> Andy Wingo <wingo@pobox.com> writes:
>
>>     -  GC_REGISTER_DISPLACEMENT (2*sizeof(scm_t_bits)); /* for the self data pointer */
>>     -  GC_REGISTER_DISPLACEMENT (2*sizeof(scm_t_bits)
>>     -                            + scm_tc3_struct); /* for the vtable data pointer */
>>     +  /* The first word of a struct is equal to `SCM_STRUCT_DATA (vtable) +
>>     +     scm_tc3_struct', and `SCM_STRUCT_DATA (vtable)' is 2 words after VTABLE by
>>     +     default.  */
>>     +  GC_REGISTER_DISPLACEMENT (2 * sizeof (scm_t_bits) + scm_tc3_struct);
>>
>> I'm pretty sure the first case is still needed, for instance
>> reallocation. It is, as the comment indicates, "for the self data
>> pointer" -- the second word of the object, which normally points to the
>> third word of the object, but for reallocated instances points to the
>> third word of *another* object.
>
> As an excuse, the comment was a bit too terse for me to understand, but
> I get your point now.  However, it’s still unclear to me how an ‘s’
> field can point to something else than “self”.

Hm, I think you're right; the wording was not only terse but misleading.
It is "for the data pointer, which might point to another object's
data", or something like that. Nothing to do with `s' fields.

>> See my article on redefining classes for more info.
>
> By ‘self’ here, did you mean what’s known as ‘data’ on the diagrams at
> <http://wingolog.org/archives/2009/11/09/class-redefinition-in-guile>?

Yes indeed.

> PS: It would be really cool if this blog post were part of the “Data
>     Representation” node of the manual.  :-)

Heh, yeah. That whole chapter needs revision...

Andy
-- 
http://wingolog.org/

Re: struct displacements

[Ludovic Courtès]

Hey,

Andy Wingo <wingo@pobox.com> writes:

> On Mon 25 Jan 2010 14:44, ludo@gnu.org (Ludovic Courtès) writes:
>
>> Andy Wingo <wingo@pobox.com> writes:
>>
>>>     -  GC_REGISTER_DISPLACEMENT (2*sizeof(scm_t_bits)); /* for the self data pointer */
>>>     -  GC_REGISTER_DISPLACEMENT (2*sizeof(scm_t_bits)
>>>     -                            + scm_tc3_struct); /* for the vtable data pointer */
>>>     +  /* The first word of a struct is equal to `SCM_STRUCT_DATA (vtable) +
>>>     +     scm_tc3_struct', and `SCM_STRUCT_DATA (vtable)' is 2 words after VTABLE by
>>>     +     default.  */
>>>     +  GC_REGISTER_DISPLACEMENT (2 * sizeof (scm_t_bits) + scm_tc3_struct);
>>>
>>> I'm pretty sure the first case is still needed, for instance
>>> reallocation. It is, as the comment indicates, "for the self data
>>> pointer" -- the second word of the object, which normally points to the
>>> third word of the object, but for reallocated instances points to the
>>> third word of *another* object.
>>
>> As an excuse, the comment was a bit too terse for me to understand, but
>> I get your point now.  However, it’s still unclear to me how an ‘s’
>> field can point to something else than “self”.
>
> Hm, I think you're right; the wording was not only terse but misleading.
> It is "for the data pointer, which might point to another object's
> data", or something like that. Nothing to do with `s' fields.

OK, done:

  http://git.sv.gnu.org/cgit/guile.git/commit/?id=227eff6a77dd58d64775fe91ae0f1596bfe3371b

Thanks,
Ludo'.