Re: struct interface

[Michael Livshin <cmm at verisity dot com>]

Mikael Djurfeldt <mdj@mdj.nada.kth.se> writes:

> Clark McGrew <mcgrew@ale.physics.sunysb.edu> writes:
> 
> >     Mikael> Regarding structs: As I've said earlier, structs will be
> >     Mikael> replaced, so it's probably waste of time change them.
> > 
> > How soon?
> 
> When someone does it.

you mean the GOOPS FDI or something else?

> >     Mikael> Regarding your suggestion above: When we design the new
> >     Mikael> record representation, the "layouts" of objects should
> >     Mikael> probably allow for SCM values to be located non-word
> >     Mikael> aligned within the object.
> > 
> > Why?  I don't see the utility and it will increase the complexity of
> > the access macros since misaligned SCM values can't be accessed
> > directly on some architectures.
> 
> Because SCM values situated in a struct can be positioned non-SCM
> aliged on some architectures.  Consider an application object which
> has an SCM value at some location within a struct.

it's good to support such things, but they are definetely pathologic.
I don't think that removing the word-alignment assumption from the
core Guile memory access macros is a good idea.

[ but maybe I'm not seeing some obvious use for non-word-aligned SCM
values.  the only two things I can think of are "packed" data and data 
coming from another architecture on the netword or something. ]

> We would also like to have a tool which can automatically build GOOPS
> glue-code for library header files.  (It might turn out that it is
> necessary to annotate such header files in some way.)

is Swig easy enough to mangle in this direction?  does anybody know?

-- 
The journey of a thousand miles begins with an open parenthesis.
                                                           -- Rainer Joswig