[Stds-754] Prolegomena to any future decimal discussion

[Michel Hack (1-914-784-7648) <hack@xxxxxxxxxxxxxx>]

John Harrison wrote:
> What about issues besides performance?  DPD is significantly more
> complex intellectually.  (Can you remember all the details of the
> DPD encoding, or sight-read DPD numbers?)  And DPD retains no
> vestige of lexicographic ordering, while ZBD with the appropriate
> field layout has some potentially useful ordering properties.

I'll buy the lexicographic observation, though that is seriously messed
up by the combination field in all three formats presented here.  (Some
earlier Intel proposals had tried to deal with this.)

I'm amazed however that the sight-reading argument is cited *against* DPD.

It's true that BID has an advantage when the coefficient has some binary
structure, e.g. near powers of two, or bit fields, but in a DFP environment
that is likely to happen only for NaN payloads.  ZBD and DPD both suffer
from the fact that every other declet is misaligned relative to hex digits
(I'm assuming here that sight-reading in hex is implied), but this may be
enough to recognise parts of a decimal number.  Many people can read small
integers in hex (e.g. up to 256 or so), but I should point out that, for
DPD, numbers from 000 to 079 stand for themselves.  I have stared at many
DPD encodings in hex, and the ability to choose test cases that are in fact
recognisable has been quite helpful.

Michel.
Sent: 2006-06-28 14:46:17 UTC