Consider using an existing rust validation library?

[jmetz]

When looking into the feasibility of this recently, I came across eg

• validator - https://crates.io/crates/validator
• garde - https://crates.io/crates/garde
• serde_valid - https://crates.io/crates/serde_valid

Perhaps it would make more sense to use one of these for the "easy" fields and only worry about the more complicated validation logic ourselves - or is there a reason not to?

[Tomaz-Vieira]

So I looked into the libraries and if I understand correctly, the flow is something like this:

pub struct User{
  #[validate(minimum = 0)]
  #[validate(maximum = 120)] // users can be at most 120 years old
  age: u8,
}

let user = User{age: 255}; // instantiate an invalid user
if let Err(err) = user.validate(){
  # handle invalid value
}

It's nice that we don't have to write the validation code ourselves, but it does come at a cost: Because this validation is not done at construction-time, it is possible to have invalid Users floating around the code, meaning that any code that touches a User can't be sure that it's dealing with a valid User, and that limits the compiler's ability to catch errors; the compiler can never enforce that we are actually calling .validate() and then handling the errors accordingly.

Still, what we have today might not be too different from that... Under bioimg_spec::rdf we have the stuff that is serialized directly to YAML, and under bioimg_spec::runtime we have structs that are parsed, "valid" objects, constructed from those in bioimg_spec::rdf. So, the rdf::* stuff is like "syntactic" checking and the runtime::* stuff is like "semantic" validation. So it's conceivable we could use those validation libraries inside runtime::*, though I don't know if it will actually make our lives easier than to simply write the validation code 🤔

I'll come back to this one runtime::* is more populated so I can have a feel of how annoying the validation code is actually getting.

[jmetz]

Hmm I guess this is almost always going to be used for runtime validation - that's kinda the point of the library.

The idea of enforcing compile-time validation, for the rare scenario where someone is hard-coding an RDF in rust... I'm not sure it made sense to spend time on that tbh 🤷

That said, you've done that now I guess. I'm still not 100% clear as to the utility... am testing now.

[jmetz]

I just tested your argument, and it does not really work as you had intentioned - see #3

[jmetz]

See #3