Re: [h5md-user] units module

[Felix Höfling]

Am 11.10.2013, 17:06 Uhr, schrieb Peter Colberg
<address@hidden>:

> Hi,
>
> Before adding a units module to the repository that defines an
> optional unit system for the "unit" attributes, I would like to
> solicit feedback from the H5MD community.
>
> The idea is to have a module named "units" that carries a module
> attribute "system" that defines the unit system. Rather than having
> a separate module for each unit system, this way a program can query
> the unit system by reading the attribute "/h5md/modules/address@hidden".
>
> The module would initially support the unit system "si". Here the
> question is whether and how to define the format of the unit strings.
> Konrad suggested the udunits2 package earlier. It seems a good basis:
>
> http://www.unidata.ucar.edu/software/udunits/udunits-2/udunits2lib.html#Syntax
>
> The udunits2 grammar is very flexible. It would probably be sensible
> to define a smaller grammar, so programs wishing to interpret units do
> not have to use the udunits2 library due to the complexity of parsing.
>
> Debian ships with a udunits-bin package if you wish to try the syntax.
>
> Peter

The package brings an impressive collection of units, SI-base, SI-derived,  
SI-accepted, non-SI, ... it knows even about Poise and Dyne ;-) I can  
hardly see anything that is missing -- except for "Lennard-Jones" units  
which usually are kind of abstract and without a conversion to physical  
units (I'm not talking about Argon ...)

http://www.unidata.ucar.edu/software/udunits/udunits-2/udunits2.html#Database

The syntax looks quite natural to me, a human has a good chance to readily  
grip the meaning.

Using the package as a reference for how the unit strings are to specified  
seems reasonable. I'm a bit unsure whether the spec of the units module  
should depend on an external package, or whether we should copy all the  
strings verbosely (i.e., converting the XML files to a pretty table).

One task that the units module (or the udunits library) should be able to  
do: convert between units, e.g., recognise that "1 N m = 1 Pa = 10⁻⁵ bar"?  
At least the conversion factor is listed in one of the XMLs. Without such  
a functionality, the automatic parsing would be of limited value.

Restricting the set of units to "SI-base" or "SI-base plus SI-derived"  
makes sense. But we have to think about arbitrary combinations of sets of  
units, one part of the H5MD file may want to use SI-base with SI-derived,  
but mix it with SI-accepted (e.g., because these data come from another  
source). I tend to make the "system" attribute a list rather than a single  
entry.

Cheers,

Felix