Re: eval.c

[Dirk Herrmann <dirk at ida dot ing dot tu-bs dot de>]

Thanks for the answer.

On 18 Apr 2000, Jost Boekemeier wrote:

> Dirk Herrmann <dirk@ida.ing.tu-bs.de> writes:
> 
> > Hi!
> > 
> > in eval.c there is a cell 'undef_cell' defined:
> >   static scm_cell undef_cell = { SCM_UNDEFINED, SCM_UNDEFINED };
> > 
> > The problem is:  what happens, if this cell is not aligned on a 64-Bit
> > boundary?  
> 
> Hä?  What do you mean?

Every cell that you get via SCM_NEWCELL is guaranteed to be on a 8 byte
boundary.  This guarantee is the reason, why SCM_NIMP works the way it is
implemented:  x & 7 == 0 --> non immediate, x is a pointer to a cell.

In this case, the static variable scm_cell is placed in memory at an
arbitrary address at the will of the compiler, which will on most
architectures be 4-byte aligned, but not necessarily 8-byte
aligned.  Thus, it can happen that the address of this pseudo cell is
something, for which SCM_NIMP might _not_ be fulfilled.  Thus, if somebody
was to play around with that cell, like making a SCM value of the cell
address, an invalid SCM variable may be the result.

> > The only access is:
> >   return SCM_CDRLOC (&undef_cell);
> 
> Yep.  It is used in the macro expander.
> 
> 
> > * Question:  Is it really necessary to have a _cell_ here?  It seems that
> > only the cdr is needed, and that it even doesn't matter if it really is a
> > valid cdr, i.e. at at valid cdr heap address.  If so, I suggest to simply
> > use a single SCM variable, and call that undef_object.
> 
> 
> Take a look at `macroexp'. 

Thanks for the hint.  I will try to figure it out.

Best regards
Dirk Herrmann