Issue: Floating point rounding errors

[incansvl]

ucum.js: 0.0.2

The library seems to suffer from the effects of number representation errors, I assume these are caused by using binary floating point arithmetic for internal calculations.

The effects may be minor (hopefully) when dealing with real-world measurements, but they do cause ugly results when converting "pure" units by powers of ten, e.g:

Input            Canonicalises To
pg/L             {"value":9.999999999999999e-10,"units":{"g":1,"m":-3}}
(implies 1 unit)

I believe this would would be improved using a high (or arbitrary) precision decimal arithmetic library within the module. Several such libraries are available on Github.

I realise this might be a significant rewrite, and would carry a speed penalty, but it would improve the usability of the library for unit conversion projects.

[jmandel]

Thanks for the comment! This library isn't actively maintained, but I'd be happy to accept a contribution to integrate a bigdecimal library like https://github.com/peterolson/BigRational.js/ (for example).