validatrix

A lightweight validator library for rust.

Validatrix contains no built-in validators, just traits and error types for your own custom validation.

Designed for cases where:

• possibly-valid types are built up of other possibly-valid types
• there is additional schema-level validation
• data is modelled as JSON-like, where sequences are ordered and maps' keys are stringy

The Display implementation of validatrix::Error can list multiple validation errors, pointing to the location of the errors with JSONPath-like syntax, although implementors can choose to fail fast instead.

Usage

use validatrix::{Validate, Accumulator, Valid};

#[derive(Debug, Clone, serde::Serialize, serde::Deserialize)]
struct A {
    /// Must not be divisible by 3.
    avalue: u8,
    /// Must be valid.
    b: B,
}

#[derive(Debug, Clone, serde::Serialize, serde::Deserialize)]
struct B {
    /// Must not be divisible by 5.
    bvalue: u8,
    /// All must be valid.
    cs: Vec<C>,
}

#[derive(Debug, Clone, serde::Serialize, serde::Deserialize)]
struct C {
    /// Must not be divisible by 3 and 5.
    cvalue: u8,
}

// Implement `validatrix::Validate` on your structs.
impl Validate for A {
    // `Accumulator` allows you to continue looking for validation errors after the first.
    // But you can return early if you prefer.
    fn validate_inner(&self, accum: &mut Accumulator) {
        if self.avalue % 3 == 0 {
            // Each failure is added with a context: the name of the field
            // (or index of a sequence) which failed.
            accum.add_failure_at("avalue", "fizz");
        }

        // Fields implementing validatrix::Validate can have validation errors accumulated too.
        accum.validate_member_at("b", &self.b);
    }
}

impl Validate for B {
    fn validate_inner(&self, accum: &mut Accumulator) {
        if self.bvalue % 5 == 0 {
            // You can also manually do validation within a prefix context
            accum.with_key("bvalue", |a| a.add_failure("buzz"));
        }

        // Helper method for validating a sequence of validatrix::Validate structs
        accum.validate_iter_at("cs", &self.cs);
    }
}


impl Validate for C {
    fn validate_inner(&self, accum: &mut Accumulator) {
        if (self.cvalue % 3 * self.cvalue % 5) == 0 {
            accum.add_failure_at("cvalue", "fizzbuzz")
        }
    }
}

// valid
let valid = A {
    avalue: 1,
    b: B {
        bvalue: 1,
        cs: vec![C { cvalue: 1 }, C { cvalue: 1 }],
    },
};
valid.validate().unwrap();

// all of the value fields are fizz/buzz, and therefore invalid
let invalid = A {
    avalue: 3,
    b: B {
        bvalue: 5,
        cs: vec![C { cvalue: 15 }, C { cvalue: 30 }],
    },
};
let err = invalid.validate().unwrap_err();
let validation_report = format!("{err}");
assert_eq!(validation_report, "
Validation failure(s):
   $.avalue: fizz
   $.b.bvalue: buzz
   $.b.cs[0].cvalue: fizzbuzz
   $.b.cs[1].cvalue: fizzbuzz
".trim());

// the `Valid` wrapper type enforces validity
let valid_wrapped = Valid::try_new(valid.clone()).expect("is valid");
assert!(Valid::try_new(invalid.clone()).is_err());
// `Valid` implements AsRef, Borrow, and Deref for the contained type

#[cfg(feature = "serde")]
{
    // You can also deserialize directly into a Valid;
    // validation errors are raised by serde.
    let valid_wrapped_deser: Valid<A> = serde_json::from_str(
        &serde_json::to_string(&valid).unwrap()
    ).unwrap();

    // serialization is handled transparently
    let invalid_str = serde_json::to_string(&invalid).unwrap();
    assert!(serde_json::from_str::<Valid<A>>(&invalid_str).is_err());
}

There is also an asynchronous variant in the validatrix::asynch module. See also validatrix(::asynch)::ValidateContext, which allows passing a reference to some external data as context for the validation.

Why not

• validator
• validators
• serde_valid

Other validation crates have focused on providing validator functions and proc macros to decorate types. I found those validators are often trivial implement yourself, the DSLs for decorating fields just look worse than regular rust code, and composing custom and built-in validators behaved in unclear ways.

JSONSchema-like validators tend not to be good at schema-level validation.

Development

Make releases using cargo-release.

Use prek to manage pre-commit hooks.

To do

• use smallvec to decrease allocations Doesn't really help
• use Cow<str> (or alternative like hipstr, ecow etc.) for messages
• investigate whether RAII can be used instead of pushing and popping accumulator prefixes
• Accumulator could have a fail-fast mode
  • methods would return Results (Err if fail-fast is true, otherwise Ok) so they can be ?'d and propagate
  • this would cause weirdness in the &mut self methods which would then need to cede their failures to the returned errors
• replace String in Failure with generic error