[As an individual] I agree with Juergen that in many configuration cases, using decimal64 is better/safer than binary float/double. However, there are other scenarios, such as operational data coming from sensors, where float/double is probably more appropriate/useful, hence I would still like to see the next version of YANG supporting float/double, possibly restricted to operational data only.
For configuration, with regards to the rounding errors alluded to below, I do have some sympathy with Chris's suggestion of support for arbitrary precision decimal numbers. It seems that more and more languages have native support for arbitrary precision decimal maths. I note that CBOR also has an encoding for them, and a JSON/XML encoding of them is seemingly trivial. Regards, Rob > -----Original Message----- > From: netmod <[email protected]> On Behalf Of Juergen Schoenwaelder > Sent: 07 July 2020 12:25 > To: Christian Hopps <[email protected]> > Cc: NetMod WG <[email protected]> > Subject: Re: [netmod] Justification for decimal64 over string for floating > point values in geo location data? > > Precision often means different things to different people. Here is my > take: > > - Floating point numbers have almost always rounding errors. And > floating point numbers use binary fractions, a decimal fraction like > 1.0 has no precise representation as a binary fraction. Type 0.1 + > 0.2 into python or haskell or any other language that gives you bare > floating point numbers and enjoy the result. > > - Fixed precision decimal numbers do not have rounding errors since > they are essentially scaled integers and hence they are precise as > long as calculations stay within the range. > > - Floating point numbers can cover a large number space (from very > tiny to really big), fixed precision decimal numbers are much more > restrictive. > > - In XML and JSON, numbers are rendered in strings that likely do not > look much different if its a decimal64 or a float or ... If you really > care about size, use a binary encoding like CBOR. > > /js > > On Tue, Jul 07, 2020 at 07:06:20AM -0400, Christian Hopps wrote: > > I received feedback in my YANG doctor review (thanks Mahesh) regarding > the use of decimal64 for most of the values in the geo location grouping > (https://tools.ietf.org/html/draft-ietf-netmod-geo-location-04). In my > comparison sections I note that some precision (at the very extremes) may > be lost when converting from other geo location formats that use string > (or double for w3c) to decimal64. Given that mention of loss of extreme > precision, the reviewer was asking if some justification for the decimal64 > should be given in the document. > > > > What are the advantages to using decimal64 for floating point numbers vs > using a string with a pattern "[0-9]+(\.[0-9]+)?" (convert that to yang > pattern language). The advantage of using a string is that the precision > of the value is not restricted by the model. Does the YANG decimal64 > values have a concise binary format that can be more efficiently > transported or stored in binary form? If so is this the only advantage, > and is it enough of one to limit the precision in the model? > > > > It's definitely worth noting that the precision of the decimal64 values > seem vastly adequate for geo location data (e.g., for Cartesian > coordinates and height values which are measured in meters the fractional > digits is 6 which means the surface could be up to 9 billion kilometers > large (or away from for height) and the precision is to the micrometer. > For ellipsoidal coordinates there are 12 fractional digits for the > degrees. > > > > Thanks, > > Chris. > > > > > _______________________________________________ > > netmod mailing list > > [email protected] > > https://www.ietf.org/mailman/listinfo/netmod > > > -- > Juergen Schoenwaelder Jacobs University Bremen gGmbH > Phone: +49 421 200 3587 Campus Ring 1 | 28759 Bremen | Germany > Fax: +49 421 200 3103 <https://www.jacobs-university.de/> > > _______________________________________________ > netmod mailing list > [email protected] > https://www.ietf.org/mailman/listinfo/netmod _______________________________________________ netmod mailing list [email protected] https://www.ietf.org/mailman/listinfo/netmod
