Crate validus

source ·
Expand description

validus — validated string slices

validus — validated string slices

// A three-step process.

// 1. Define your validation. No need to re-implement your validation logic!
// (though, we have a macro that makes it easy if you do have control.)
// Validus offers multiple tradeoff points between flexibility and convenience.

// 2. Replace `str` with `vstr<Rule>` where you need validation.
// `&vstr` coerces to `&str`, so you can use it deep inside your codebase,
// or even a foreign codebase. (Limitations exist.)

// 3. Use `vstr<Rule>` as though it were `str`.

use validus::prelude::*; // vstr, fast_rule, <&str>.validate(), etc.
use validus::fast_rule;

// Let's define a string validation rule that uses a Regex.
const USERNAME_RE_S: &str = r#"^[a-zA-Z0-9_]{1,16}$"#;
static USERNAME_RE: OnceLock<Regex> = OnceLock::new();

// 1. Defining the rule type and the error type.
// - I used fast_rule! because I have control over the error type.
// - If you are using existing error types and/or function,
// you can use easy_rule! instead. There's also a macro-free way;
// see the `vstr` module docs for more info.
struct BadUsernameError;
struct UsernameRule;
fast_rule!(
    UsernameRule,
    err = BadUsernameError,
    msg = "bad username",
    // The closure below could very well be a function.
    // So if the function were called `username_rule`,
    // I could do fast_rule!(..., ..., ..., username_rule)
    |s: &str| {
        let re = USERNAME_RE.get_or_init(|| Regex::new(USERNAME_RE_S).unwrap());
        re.is_match(s)
    }
);
// (There's also easy_rule that allows you to bring your own error type.)
// (That's helpful when you have existing error types.)
// (... or, walk the macro-free path and implement ValidateString manually.)
// (consult the `vstr` module docs for more info.)

// 2. Restrict your string slice with the rule.
type Username = vstr<UsernameRule>;

// 3. Use your `vstr<Rule>` as though it were `str`.
let input = "hello";
let username: Result<&Username, BadUsernameError> = input.validate();
assert!(username.is_ok());
assert_eq!(username.unwrap(), "hello");

let input = "haha 😊";
let username: Result<&Username, _> = input.validate();
assert!(username.is_err());

// It's that easy! Let's do a recap.

// - Error and the rule:
// struct MyError;
// struct MyRule;
// fn my_rule_priv(s: &str) -> Result<(), MyError> { ... }
// fast_rule!(MyRule, err = MyError, msg = "my error", my_rule_priv);

// - Use `vstr` as though it were `str`:
// type MyStr = vstr<MyRule>;
// let a: &vstr<MyRule> = "hello".validate().unwrap();

// Plus, this library has serde support with validation, and more.

Synopsis

  1. Bring your own validation, and use vstr<_> as though it were str (where immutable str references are expected).
  2. Mix it into your code. If you have existing validation modules, don’t touch it. Validus has multiple easy ways to wrap your validation into a vstr<_>-compatible rule type.
  3. Take a shortcut when starting out. If you don’t have existing validation modules, you can use the fast_rule! macro to quickly define a rule type with a predicate (closure or external function, which may or may not belong to your crate). Your error type will be a proper std::error::Error type; you can use an ad-hoc &'static str error message, even.
  4. Work with serde with no extra code. Simply add #[derive(Serialize, Deserialize)] to your struct as usual. Now, no need to worry about invalid strings getting deserialized.
  5. Opt into lazy validation if you want. With a special rule called Later<_>, you can also choose to either eagerly validate or defer validation until when you decide you need it. (Pretty useful for serde deserialization, for example.)
  6. Combine rules on the spot. Three logical connectives (and, or, not) are provided as an extension enabled by default; those are in your prelude, so you can use them right away.

The Validus library provides a newtype over regular string slices called vstr. This vstr is parameterized by your own validation rule type, like vstr<EmailRule>. Any zero-sized type can be used as a rule if it implements the trait ValidateString.

vstr is meant to be a replacement of str in certain contexts. So,

  • vstr implements Eq with vstr of other rules as well as str,
  • … and Ord and Hash the same as any other str reference.
  • vstr is aware of String, Cow and smart pointer types such as Box, Rc and Arc.

To show you how vstr<_> compares and hashes the same as str references, I will give you an example of directly using vstr as keys in HashMaps and HashSets.

// Illustration: using vstr<_> as a key in a HashMap.

struct BadUsernameError;
struct UsernameRule;
fast_rule!(
    UsernameRule,
    err = BadUsernameError,
    msg = "bad username",
    |s: &str| s.len() <= 16
);

type Username = vstr<UsernameRule>;

let mut map = HashMap::<&Username, i32>::new();
map.insert("hello".validate().unwrap(), 1);
map.insert("world".validate().unwrap(), 2);

// assume_valid bypasses validation, incurring no computational cost,
// so it's useful in this case.
assert_eq!(map.get("hello".assume_valid()), Some(&1));
assert_eq!(map.get("world".assume_valid()), Some(&2));

// So every time you need a `&str` but with validation,
// you know that Validus and `vstr<_>` have got you covered,
// anywhere in your codebase, or a foreign codebase.
// (Limitations exist.)

serde with validation

With the optional serde feature, this crate also supports serialization and deserialization with validation. This means that you can use vstr<_> as a field in a serde-powered struct, and if the input fails the validation, it will be rejected and an error according to the validation rule’s associated Error type will be returned.

  • The serde feature is enabled by default. Disable it using default-features = false in your Cargo.toml.
// Illustration: a struct with a validated email field.

#[cfg(feature = "serde")] {

use validus::prelude::*;
use validus::fast_rule;
use serde::Deserialize;

// This rule is very generous. It accepts any string that
// contains an at-symbol.
// (When the error type is not specified, it is inferred to
// be &'static str.)
struct EmailRule;
fast_rule!(EmailRule, msg = "no at-symbol", |s: &str| s.contains('@'));

#[derive(Deserialize)]
pub struct User {
    pub email: Box<vstr<EmailRule>>,
}

let input = r#"{"email": "notgood"}"#;
let result = serde_json::from_str::<User>(input);
assert!(result.is_err());

let input = r#"{"email": "hi@example.com"}"#;
let result = serde_json::from_str::<User>(input);
assert!(result.is_ok());
assert!(result.unwrap().email.as_str() == "hi@example.com");

}

Deferred validation with Later

Sometimes, you want to validate a string slice only when it is actually used.

For this need, there is a rule called Later that bypasses all validation, but specifies what rule it is supposed to be validated with. When the validation is actually needed, you can call make_strict to validate the string slice and convert it to a vstr with the specified rule.

Here, I copy the example code from the Later type documentation.

use validus::prelude::*;
use validus::fast_rule;

struct EmailError;
struct Email;
fast_rule!(Email,
    err = EmailError,
    msg = "no @ symbol",
    |s: &str| s.contains('@')
);

// Here, we start with an email with deferred (postponed) validation.
// Validation of `Later<_>` is infallible.
let v1: &vstr<Later<Email>> = "hi@example.com".validate().unwrap();
// Now, we truly validate it.
let v1: Result<&vstr<Email>, _> = v1.make_strict();
assert!(v1.is_ok());

// So, again, this is going to succeed.
let v2 = "notgood".validate::<Later<Email>>().unwrap();
// But, when we check it, it will fail, since it is not a good email address
// (according to the rule we defined).
let v2 = v2.make_strict();
assert!(v2.is_err());

// With the extension `StrExt`, we can also call `.assume_valid()`
// to skip validation, since we know that `Later<_>` doesn't validate.

let relaxed = "hi@example.com".assume_valid::<Later<Email>>();
assert!(relaxed.check().is_ok()); // This is infallible because `Later<_>` is infallible.
assert!(relaxed.make_strict().is_ok()); // Later<Email> -> Email.

let relaxed = "nonono".assume_valid::<Later<Email>>();
assert!(relaxed.check().is_ok()); // Yup, it is still infallible.
let strict = relaxed.make_strict(); // Now, we made it strict.
assert!(strict.is_err()); // It didn't make it (it was a bad email address.)

Overriding a rule with assume_valid and checking with check

You are also given the power to override the underlying mechanism using assume_valid. This is useful when you have a vstr<_> that you know is valid, but that is difficult to decide at a given moment; or, when, for some reason, you don’t need to validate a vstr<_> (for example, when you are using it as a look-up key in a HashMap). The crate provides the check() method that can be used to establish the validity of a vstr<_>.

// Illustration: overriding the validation mechanism.

use validus::prelude::*;
use validus::easy_rule;

// easy_rule is a different macro that helps you define rules.
// The difference with fast_rule! is that leaves the error type
// untouched, and you need to return a Result<(), YourError>
// instead of a bool in the closure.
struct No;
easy_rule!(No, err = &'static str, |s: &str| Err("i won't accept anything"));

let s = "hello";
let v: &vstr<No> = vstr::assume_valid(s);

// Yup, it works. We overrode the validation mechanism.
assert_eq!(v, "hello");

// But it's not valid. Let's test that.
assert!(v.check().is_err());

(assume_valid is NOT unsafe: vstr makes no further guarantees about the validity of the string slice beyond what str provides. [it also doesn’t make any fewer]. Thus, assume_valid may not be blamed for causing undefined behavior.)

Defining implication among rules

Furthermore, since some pairs of rules can be converted automatically (there is an IMPLIES relation between them), you can use the change_rules associated method to convert a reference to vstr<Rule1> to a reference to vstr<Rule2>. This requires Rule to implement Into<Rule2>. (Otherwise, the regular try_change_rules can be used between any two rules.)

// Illustration: declaring implication.
// Implication means: "Whenever [rule] A says good, so does B."

use validus::prelude::*;
use validus::fast_rule;

// Less generous
struct A;
fast_rule!(A, msg = "no wow", |s: &str| s.contains("wow"));

// More generous: includes all strings that A accepts and
// perhaps more.
struct B;
fast_rule!(B, msg = "neither wow nor bad found", |s: &str| {
    s.contains("wow") || s.contains("bad")
});

// Assert that A implies B.
// In English: "whatever string A accepts, B accepts, too."
impl From<A> for B {
    // This particular formulation is idiomatic
    // to the `validus` crate because all rules are supposed
    // to be freely constructed Zero-Sized Types (ZSTs).
    fn from(_: A) -> Self {
        // And, this value never gets used, anyway.
        // All methods of `ValidateString` (trait that
        // defines rules) have static methods, not instance
        // methods.
        B
    }
}

// The declaration of implication unlocks the `change_rules`
// method that converts a reference to `vstr<A>` to a reference
// to `vstr<B>` infallibly.

let good = "wow bad";
let a: &vstr<A> = vstr::assume_valid(good); // we know it works, so.
let _: &vstr<B> = a.change_rules(); // infallible. see, no Result or unwrap().

The special rule ValidateAll

Oh, one more. There are two special rules which validate all strings and no strings, respectively. They are called ValidateAll and (). Though you can’t use change_rules to convert your rule to ValidateAll, you can still use a dedicated method called erase_rules just for that. From ValidateAll, you can use try_change_rules to convert to any other rule.

Batteries included … (but I need your help!)

Check out some of the prepared validation rules in the module vstrext. The module should already have been imported in the prelude module (it’s feature-gated by ext, which is enabled by default.)

Currently, three logical connectives and a few rules are implemented in the extension module:

  • Conj, which requires two rules to be satisfied.
  • Disj, which requires at least one of two rules to be satisfied.
  • Neg, which requires a rule to be un-satisfied.
  • StringSizeRule (and its variants), which check the size of a string.
  • StringAsciiRule, our only rule that checks the content of a string so far.

I would really appreciate your help in adding more rules to the extension module.

Experimental features

Contingent validation

The experimental cow feature introduces a new type, VCow that represents a Cow<'_, vstr<_>> that is either valid or invalid. The validity is tracked at runtime.

NOTE: vstr already supports being in a Cow like this: Cow<'_, vstr<_>> even without the cow feature. The cow feature adds an experimental wrapper type that tracks the validity of the vstr that may change at runtime.

Features

  • (default) serde: enables serde support. Pulls in serde as a dependency.
  • (default) ext: enables built-in extensions. Pulls in thiserror as a dependency.
  • (default) std: enables std support. Necessary to implement the Error trait. Pulls in std as a dependency.
  • (default) alloc: integrates with alloc crate., and enables smart pointers.
  • cow: enables the experimental VCow type. Pulls in alloc as a dependency.

Modules

Macros

  • Define a rule using a closure.
  • Define a rule using a predicate with ad-hoc errors.