Re: [Axiom-developer] Unit package question - Reply to 1st half

[William Sit]

C Y wrote:
> 
> --- William Sit <address@hidden> wrote:
> 
> > Actually, we can even do better, without the user doing a
> > setUnitLength(nm), if the only reason to use nm instead of
> > m is to get rid of annoying powers of 10. For numerical
> > output (the problem does not arise for symbolic output), the
> > output coerce routine can sense the magnitude of the value in the
> > unit and shift the decimal place appropriately and adjusting
> > using a suitable prefix. So one stays in SI but the output
> > gets rid of the powers of 10.
> 
> Are nanometer, etc. considered part of SI then?  

SI provides standardized prefixes and abbrevations in the range 10^(-24) to
10^(24), one for every three unit change in the exponent, which are applicable
to ANY physical quantity (with the exception of kilogram, see SI pages, to avoid
double prefixes). More interestingly, SI also has prefixes and abbreviations for
the computer industry: kilibit (1024 bits) vs kilobit (1000 bits). Axiom
implementation definitely should adhere to SI standards and implement all their
standardized and accepted units.

> Nifty idea, and one
> worth following up on, but I would still like to peg things at nm, for
> the following reason/scenario:  Let's say I know that my calculations
> and data SHOULD be coming out in the nm range.  I set my unit for
> length to be nm, proceed, and mostly see small numbers of nm.  All well
> and good, until suddenly I have a huge or very small number appearing.
> Red flag!  Not close to 1 nm.  With autoscaling, the number would have
> stayed near one and the unit would have changed.  Sometimes useful, I
> agree, but I can see arguments for both sides here.

Since SI provides prefixes only for every thousand-fold change, if all
computations you expect are in the scale of 1 nm, you would probably spot these
errors in hundreds of nm. But anyway, every scientist should be aware of units
attached to quantities, especially in graphs! I think autoscaling is good enough
(Note, all digital meters autoscales! Scientists are used to that.)

I think whoever implements this should first program for the correct framework,
and provide basic unit systems based on SI. Then as usage expands, add more unit
system domains. Then if that is not enough, add more friendly user settings. I
am fairly convinced that the framework proposed earlier is able to handle and
facilitate this plan of development. I would like to hear more critique from
others.

Thanks for this interesting discussion.

William