Range for standard number types

[jakob0]

Is there a reason, why no default minimum and maximum values are generated with schema_for! for rust built-in number types, like i8-i128, u8-u128, isize, usize, f32, f64?

[GREsau]

I try to strike a balance simplicity/clarity with specificity - including minimum/maximum would definitely make the schemas more specific, but could also increase noise in the output.

I suppose that the smaller the range of possible values, the more valuable it would be to specify the actual min/max. Considering each type:

• u8/i8 - including minimum/maximum for these would probably be useful, so I'm very open to it
• u16/i16 - I'm open to including minimum/maximum for these types too
• u32/i32 - here's where I'm less sure. I'm a little concerned that some people choose these types for their integers somewhat arbitrarily, rather than making a conscious choice about the actual range. Although maybe that's not the case, and even if it is maybe it shouldn't really be relevant
• u64/i64 - similar to above, and also -9223372036854775808/9223372036854775807/18446744073709551615 are large enough that they're starting to add noticeable noise to the schema which reduces human-readability. They also can't be accurately represented as a double-precision floating point number, which is allowed by JSON (and serde_json and JSON schema) but can cause interoperability problems with other languages/frameworks
• u128/i128 - as above, plus these can't be serialized by serde_json without the arbitrary_precision feature
• f32/f64 - I'm concerned that including minimum/maximum for these could be misleading due to rounding and imprecision

I'm curious what your use-case is for wanting these values, as that may help make a decision as to how schemars should behave

[jakob0]

You are right, it would definitely be much more relevant for the small types. I just thought, it might help guard against overflow errors - especially for people who choose their integers 'somewhat arbitrarily'. I must admit, I did not really care about the simplicity in human-readable schemas, but I get your point.
About my use-case: I'm using the schemas to generate ui-forms (like jsonforms.io) to initialize my rust structs. For the input, I need to set the limits, so I don't get dangerous errors later on.
For me it's not that relevant anymore, since I have my custom implementation now (taking into account the limits of JSON and f64, like you mentioned). It has never been a real issue anyways, I was just curious.
Thanks for the great work!