Re: [PATCH 5/7] widl: Fix type_memsize to follow aliases so that the correct size is calculated.

[Dan Hipschman]

On Mon, Jun 23, 2008 at 10:29:11PM +0100, Rob Shearman wrote:
> diff --git a/tools/widl/typegen.c b/tools/widl/typegen.c
> index 1f8695a..84115fd 100644
> --- a/tools/widl/typegen.c
> +++ b/tools/widl/typegen.c
> @@ -809,7 +809,9 @@ size_t type_memsize(const type_t *t, unsigned int *align)
>  {
>      size_t size = 0;
> - if (t->declarray && is_conformant_array(t))
> - if (t->kind == TKIND_ALIAS)
> -
>    size = type_memsize(t->orig, align);
> - else if (t->declarray && is_conformant_array(t))
> {
>     type_memsize(t->ref, align);
>     size = 0;

What type of thing is this needed for?  A typedef'd type should have all
the same fields as the original except for the attributes, so for most
cases in type_memsize it shouldn't be necessary to check whether the
type is an alias.  For anything that needs to check attributes, I would
think it would be better to fix the predicates (such as
is_conformant_array, etc.) if they do not detect that correctly 100% of
the time.

[PATCH 5/7] widl: Fix type_memsize to follow aliases so that the correct size is calculated.

[Rob Shearman]

2008/6/24 Dan Hipschman :
> On Mon, Jun 23, 2008 at 10:29:11PM +0100, Rob Shearman wrote:
> > diff --git a/tools/widl/typegen.c b/tools/widl/typegen.c
> > index 1f8695a..84115fd 100644
> > --- a/tools/widl/typegen.c
> > +++ b/tools/widl/typegen.c
> > @@ -809,7 +809,9 @@ size_t type_memsize(const type_t *t, unsigned int *align)
> >  {
> >      size_t size = 0;
> > - if (t->declarray && is_conformant_array(t))
> > - if (t->kind == TKIND_ALIAS)
> > -
> >    size = type_memsize(t->orig, align);
> > - else if (t->declarray && is_conformant_array(t))
> > {
> >     type_memsize(t->ref, align);
> >     size = 0;
>
> What type of thing is this needed for?  A typedef'd type should have all
> the same fields as the original except for the attributes, so for most
> cases in type_memsize it shouldn't be necessary to check whether the
> type is an alias.  For anything that needs to check attributes, I would
> think it would be better to fix the predicates (such as
> is_conformant_array, etc.) if they do not detect that correctly 100% of
> the time.

Sorry, I hadn't considered the order in which I sent these patches and
didn't send the subsequent patches that show why it is needed. The
direction I'm going with this is this patch:
http://repo.or.cz/w/wine/dcerpc.git?a=commit;h=df37407c4a67a536950d1b8fbcbd8...
In this series:
http://repo.or.cz/w/wine/dcerpc.git?a=log;h=refs/heads/idl-parser

The idea is that in order to switch to generating a parse tree and
generating a header file from that we need to differentiate between a
type being defined and a type being used and that cannot be done by
using the same type pointer for both instances.

The idea, therefore, is to move the definition-specific data to a
separate details object which is pointed to by both the type being
used and the defined type. The code then needs to fix the former when
the latter is created.

Thoughts?