Re: [Pnet-developers] CVMWord union and ReadLong Int32/Int64 mismatch?

[Peter Colson]

On 24/09/2004, at 11:30 AM, Rhys Weatherley wrote:

> On Friday 24 September 2004 11:18 am, Peter Colson wrote:
> > In cvm.c, the function 'ReadLong' (where long presumably means an
> > integral type) is assumed to be an Int64
> > taken from the same union above that only contains Int32 integrals:
>
> Int64 quantities may occupy 1 or 2 stack words, depending upon the 
> platform
> size/alignment rules.  Adding a "ILInt64" field to the CVMWord union 
> would
> cause stack slots to expand to 64 bits on all platforms, wasting space 
> on
> those (like x86) that don't need 64-bit alignment of 64-bit values.
>
> The "IL_BEST_ALIGNMENT" macro will normally pad the union to the 
> correct
> boundary to prevent mis-aligned accesses on picky platforms.
>
> > I don't understand the Int32/Int64 mismatch here?
>
> The "ReadLong" function is not taking a pointer to 1 stack word.  It 
> is taking
> a pointer to the stack word containing the beginning of the value.  
> The value
> may extend into the following stack words on some platforms.

OK that makes sense.

> P.S. Wouldn't it be easier to simply migrate your customers off the
> bletcherous AS/400 onto a real hardware platform and operating system? 
> :-)

Not my customers, so not my call. I'm just trying to see if there's a 
future
porting pnet to an AS/400 platform that now allows for 8 byte pointers 
that
can be bit manipulated (but provides 4 byte longs).

Regards,
Peter Colson.