multipleOf for floats result in false validation errors

[gmkeith]

Given a schema like the following:
"salary": {
"type": "number",
"multipleOf": 0.01
}
and the value 10000.51.
The validation returns a false validation error. This appears to be due to floating point precision errors of the modulus operator used. This error may appear for any number < 1 and > 0.

I'm happy to create a fix and pull request for this. Due to the nature of the error it may require adding a dependency like http://jsfromhell.com/classes/bignumber, https://github.com/josdejong/mathjs or something similar as floating point calculations tend to quite complex problems to solve.

Let me know
Thanks
Glenn

Hey there,
I have incidentally created a dup. of this.
Do you agree to Geraints proposal here: #167 ?
I made a pull request from that here: #168
If so, you could close this issue.

[kmbilly]

The proposed fix seems cannot solve the problem, tried on chrome 39.0.2171.65 m (64-bit), an example is using data=2.3 and multipleOf=0.1, may be need to further adjust the CLOSE_ENOUGH constant? Or any better approach, like changing the calculation into integer division according to the decimal points of multipleOf?

@kmbilly Some other approaches are in the closed duplicate #167
Also found this stackoverflow discussion ...
@geraintluff

[gmkeith]

Unfortunately this is one of those situations where, imho, there are two choices are neither are entirely satisfying. Floating point calculation such as these where the required precision and range are unknown, by the nature of the problem we are trying to solve, results in complex solutions.
Such complex solutions can be either solved directly in the tv4 library, adding a lot of code about foating point calculations which adds a lot of complexity to the clean tv4 code. Or a third party library needs to be added, which is contrary, I assume, to the ethos of tv4.
I don't like either of the solutions I suggest, but to ensure accuracy you cannot fix the precision as this will likely lead to range overflows or to false errors. Anything other than fixing the precision that maintains accuracy, tends towards the complex.

@gmkeith @kmbilly However - have you actually seen "Alternative 2" in the mentioned fiddle ?
A polyfill for ECMA6 isInteger ...
http://jsfiddle.net/we1nt9ye/1/
@kmbilly that was my proposed fix and the handling for the values you described is fine, see:
http://jsfiddle.net/we1nt9ye/3/

[doingweb]

While I'd personally prefer to use a well-tested third-party library to help me take care of complicated stuff like this, I could appreciate wanting to be dependency-free.

@gmkeith @kmbilly - What do you think of @redaktor's isInteger() solution? I'd certainly find a 15-place limit preferable to false errors in these really basic use cases.

[bruderol]

any update on that??? having this problem as well.

[bruderol]

I thought a little bit about this problem. and realised that the problem is, that 0.1 as a floating point number, is allready not precisely. So how should you now correctly decide ever (in floating point arithmetics) that something is a multipleOf of a number that you can allready not store precisely in float. I guess therefore that this problem is just not solveable (no matter what you take as a allowed delta value to 0 or 1) and would recommend to just close this issue. multipleOf in json schema should probably just never be used to validate for something like multipleOf 0.1, because not really decideable correctly.

On the other hand, what should be possible: at least handle the multipleOf for integer values correctly, as proposed above.

[pateketrueke]

We're also suffering from this: json-schema-faker/json-schema-faker#379