JMESPath functions floor and ceil return integer

[BartelNieuwenhuyse]

JMESPath function floor and ceil return a json float number, while the specs and the examples in it suggest that an integer should be returned.

Expected:
floor(`3.7`) -> 3
Actual:
floor(`3.7`) -> 3.0

[BartelNieuwenhuyse]

Is there a way to cast to the underlying type of json_type::int64 instead of to int64_t?

[danielaparker]

For the int64/uint64 case, instead of

return *context.create_json(arg0);

you can safely return arg0.

For the float64 case, you should

• check if the double value is nan or inf using std::finite, and if so, set ec to jmespath_errc::invalid_type and return null.
• check if the return value of std::ceil/std::floor is inside the range of an int64_t before doing the cast, and if it isn't, I would suggest leaving it as a double (the alternative would be to convert it to a typename Json::string_type and return it as a Json string value with a sementic_tag::bigint tag)

Regarding,

Is there a way to cast to the underlying type of json_type::int64 instead of to int64_t?

there isn't, but the underlying type is always int64_t.

[BartelNieuwenhuyse]

I'm not sure if defining the numbers is an option, but I seem to hit warnings in my other attempts to define the largest\smallest double that can be safely converted to int64_t.

[danielaparker]

The cases that fail are failing because a double doesn't have enough precision to represent std::numeric_limits<int64_t>::max() exactly. The upper bound for preserving precision exactly is std::pow(2, std::numeric_limits<double>::digits). Numbers above that limit should remain in double representation. double values above that bound can't be sensibly cast to int types.