That's a good point.

Another relatively easy optimization would be to use a bit from the header
on 64-bit platforms (32-bit platforms have no available bits) to indicate
another form of extension, whereby an extra word is used as a bitmap to
indicate which words are floats. Haskell uses a similar trick to indicate
which words are pointers on the stack. This would remove the indirection of
floats in the majority of cases, except of course in the stack itself.
This shouldn't have an impact on marshaling.

BTW bits in the 64-bit header should probably have been marked as reserved
rather than making the wo_size field impossibly large.

Yotam


On Mon, Sep 16, 2013 at 1:14 PM, Markus Mottl <markus.mottl@gmail.com>wrote:

> On Mon, Sep 16, 2013 at 12:49 PM, Yotam Barnoy <yotambarnoy@gmail.com>
> wrote:
> > Why do doubles need special handling though, even on a 32-bit system? My
> > suggestion is that the Double_tag be changed to Flat_tag, meaning that
> all
> > non-pointer objects can reside in this tag. The only issue I've found so
> far
> > is that polymorphic <, <=, > and >= would not work. However, these
> operators
> > should not be allowed on a vector anyway since there is no natural
> ordering
> > scheme for vectors. If there are other issues, please let me know.
>
> Here is a problem: If you marshal a float array on a 64-bit platform,
> how is the 32-bit platform supposed to know about the "logical" and
> "physical" size of the array?  On 64-bit platforms where everything is
> the same size the distinction wouldn't matter, of course, but on
> 32-bit it does.  The header word can only provide one size (but tag
> distinctions).
>