Instance validates if `multipleOf` is an integer, but fails if it's a float

[Zac-HD]

from jsonschema import validate

instance = 9007199254740995
validate(instance, schema={"type": "integer", "multipleOf": 11})    # passes
validate(instance, schema={"type": "integer", "multipleOf": 11.0})  # fails

I'm pretty sure that this is a bug, on the basis that the spec says all integer-valued numbers should be treated as integers.

Found via python-jsonschema/hypothesis-jsonschema#103.

[Julian]

Nice :)

Needs thinking though / reading the spec language. It's true the spec says to treat floats with no fractional part as integers, but it also says (or really leaves the JSON spec to say) that floats are an acceptable implementation of number -- and I'm not sure where if anywhere covers what the requirements are on coercion to a common type.

Specifically here, 9007199254740995 is not representable as a float, the closest float is 9007199254740996.0 which is indeed not divisible by 11, so doing the division somehow will require coercing to float.

But it's possible you're both correct on whether this is covered by the spec and also possible this is fixable, not sure, needs thinking.

[Zac-HD]

Both passed on older versions of jsonschema, so for whatever it's worth this probably changed in version 4.18. I've just skipped the affected test though, and happy to go with your opinion 👍

[Julian]

Ah that indeed is interesting! Will have to remind myself.

…

On Sun, Sep 10, 2023, 19:27 Zac Hatfield-Dodds ***@***.***> wrote: Both passed on older versions of jsonschema, so for whatever it's worth this probably changed in version 4.18. I've just skipped the affected test though, and happy to go with your opinion 👍 — Reply to this email directly, view it on GitHub <#1159 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AACQQXWKRVOUYR6GRW3MERDXZZEHZANCNFSM6AAAAAA4SOQ2YY> . You are receiving this because you commented.Message ID: ***@***.***>

[IronAlpha3528]

Hi @Julian and @Zac-HD ,I think this bug is caused due to python handling floats by using IEEE 754 standard, and converts both the divisor and dividend to a float if one of them is a float. This can be fixed by using the Decimal library to like avoid that standard but I don't know if it will conflict with the spec or not.

[IronAlpha3528]

After some time sitting on this conversation, I think you were confused with the spec, and not with the actual solution😅. If you could link the specs you were referring, I could also provide an opinion on the matter.