Bug: Binary prefix and SI prefix incorrectly used for byte ('B').

[andrewmtoy]

I was trying to convert a quantity from B to MB and I got an unexpected result.

console.log(math.format(math.eval('5000000 B to MB'), 14))
Result:  4.7683715820313 MB

I was expecting to see 5MB.

I saw in the documentation, units.md, that SI prefixes such as kilo, mega, giga, etc. have the same value as their IEC counterparts kibi, mebi, gibi, etc.

According to IEC, this is incorrect. Their definitions for kibibyte and kilobyte are as follows.

The definition of the kibibyte is:
1 kibibyte = 210 bytes = 1024 bytes.

The definition of the kilobyte is:
1 kB = 1000bytes = 10^3

[josdejong]

Yes, you're correct about the definitions. Good point. It will be better in the long run to stick to the definitions (even if most people mean 1024 B when they talk about 1 kB, and hardly anybody knows about these binary prefixes like kibi).

[jmichae3]

I know about them and so do other people.
deci d 1e-1=10^-1
centi c 1e-2=10^-2
milli m 1e-3=10^(3*-1)
micro u 1e-6=10^(3*-2)
nano n 1e-9=10^(3*-3)
pico p 1e-12=10^(3*-4)
femto f 1e-15=10^(3*-5)
atto a 1e-18=10^(3*-6)
zepto z 1e-21=10^(3*-7)
yocto y 1e-24=10^(3*-8)

the powers of -3 have equivalent IEC units. and, they ARE case sensitive, because some names clash.
mibi mi 1024^-1=2^-10=2^(10*-1) (note that this can be confused with "mile" if not done correctly)
mubi ui 1024^-2=2^-20=2^(10*-2)
nabi ni 1024^-3=2^-30=2^(10*-3)
pibi pi 1024^-4=2^-40=2^(10*-4)
fembi fi 1024^-5=2^-50=2^(10*-5)
abi ai 1024^-6=2^-60=2^(10*-6)
zebi zi 1024^-7=2^-70=2^(10*-7)
yobi yi 1024^-8=2^-80=2^(10*-8)

http://jesusnjim.com/conversions/SI-units-IEC-units.html

[jmichae3]

here's what you have so far. https://github.com/josdejong/mathjs/blob/master/docs/reference/units.md

[jmichae3]

https://www.dropbox.com/s/awfcks811etnkms/math.js.modified.diff?dl=0
here is a diff you can apply with patch.exe that (?) should fix the problems with "binary" IEC prefixes. I noticed that there were SI units in the "binary" prefixes.

[jmichae3]

I also added the negative entries

[jmichae3]

https://www.dropbox.com/s/uwa4rho8q30j8nt/math.js.modified?dl=0
this is the whole modified math.js 1.20 file, in case you want to diff-merge it.

[josdejong]

Thanks, it will be just a few minutes of work to fix this in the code, unit tests, and docs.

The main concern is that this is a breaking change and so we have to be careful when/how to change this. Will require the release of a breaking v2.0. I would like to bundle all pending breaking changes in a single release (like string conversion, see #249). I have to get clear which changes we want to apply.

[josdejong]

I have created a v2 branch, which already contains the fixed behavior for bits and bytes.

It will take some time though before v2 will be ready, I'm working on a large refactoring which will enable modularization of math.js.

[josdejong]

Rethinking this, I think it may be better to fix this bug now. It's sort of a backward incompatible "change" but on the other hand it's just a fix of wrong behavior which should not be like that anyway. This bug has been open way too long, and v2 isn't yet there...