RE: [office-formula] 48 bits?

["Dennis E. Hamilton" <dennis.hamilton@acm.org>]

1. I think that the note under BITAND should be deleted.  It provides
unnecessary rationale.  We have no idea how a spreadsheet implementation
might handle exact integer values, which are all that are required for the
bit-wise logical functions.  Implications of using an IEEE standard do not
apply here.

2. Some things about how Numbers (better: *exact* Integers) are mapped to
and from bit-wise values for these functions is called for, perhaps in
subsection 5.6.1 on General information for Bit operation functions.

 - Dennis


FURTHER THOUGHTS

1. To satisfy the BITAND requirement we should probably specify *exact*
Integer parameters (since the INT(n)=n constraint isn't assured to filter
out values that are too large to be exact Integer values [and I am amazed to
see that INT is defined as floor, i.e., the largest integer not greater than
n).  

2. We only need to require that the implementation define a value kbits so
that exact integer values 0 to 2^kbits-1 are supported and kbits >=48.  

3. We then describe the logical operations as operating on the kbits-bit
binary numeral that represents the value supplied as the parameter, with the
high-order bit, b[kbits-1] defined as at the left and low-order bit, b[0]
defined to be at the right (for understanding shift operations).  

4. You probably need to know what kbits are in order to completely specify
what left shifting does.  (That is, whether a shift left of one produces
(x*2) mod 2^kbits or whether shifts off the high end have
implementation-dependent results).  

4. There are other things needed to tighten or to explicitly make
implementation dependent.  But no appeal to IEEE specifications of any kind
is necessary here.



-----Original Message-----
From: David A. Wheeler [mailto:dwheeler@dwheeler.com] 
http://lists.oasis-open.org/archives/office-formula/200912/msg00114.html
Sent: Tuesday, December 22, 2009 07:02
To: patrick@durusau.net
Cc: office-formula@lists.oasis-open.org
Subject: Re: [office-formula] 48 bits?

Patrick Durusau:
http://lists.oasis-open.org/archives/office-formula/200912/msg00113.html
> I can't speak to the "need" to handle exactly 53 bits but why create a 
> variation from IEEE 754's 64-bit representation for numbers?

There's *no* variation.  The text explains the implications of *using* the
IEEE standard... we are NOT changing anything.

Many implementations do not support an infinite number of bits :-).
Therefore, it's useful to specify a range that *is* guaranteed to be
supported.  Most people use IEEE 64-bit or better (e.g., 80-bit), and thus
have at least 53 bits for the mantissa.  Thus, if you're using integers, 48
bits of significance is within a portable range.

--- David A. Wheeler

---------------------------------------------------------------------
To unsubscribe from this mail list, you must leave the OASIS TC that
generates this mail.  Follow this link to all your TCs in OASIS at:
https://www.oasis-open.org/apps/org/workgroup/portal/my_workgroups.php