A Rust validation library
- Basic usage example
- Validation rules
- Length modes
- Inner type validation
- Newtypes
- Handling Option
- Custom validation
- Context/Self access
- Implementing rules
- Implementing
Validate
- Rule adapters
- Integration with web frameworks
- Feature flags
- Why
garde
?
To get started, install garde
:
cargo add garde -F full
And attach the Validate
derive to your type. garde
will generate an implementation of the Validate
trait for you,
allowing you to call the validate
method.
Here's what that looks like in full:
use garde::{Validate, Valid};
#[derive(Validate)]
struct User<'a> {
#[garde(ascii, length(min=3, max=25))]
username: &'a str,
#[garde(length(min=15))]
password: &'a str,
}
let user = User {
username: "test",
password: "not_a_very_good_password",
};
if let Err(e) = user.validate() {
println!("invalid user: {e}");
}
Garde can also validate enums:
use garde::{Validate, Valid};
#[derive(Validate)]
enum Data {
Struct {
#[garde(range(min=-10, max=10))]
field: i32,
},
Tuple(
#[garde(ascii)]
String
),
}
let data = Data::Struct { field: 100 };
if let Err(e) = data.validate() {
println!("invalid data: {e}");
}
name | format | validation | feature flag |
---|---|---|---|
required | #[garde(required)] |
is value set | - |
ascii | #[garde(ascii)] |
only contains ASCII | - |
alphanumeric | #[garde(alphanumeric)] |
only letters and digits | - |
#[garde(email)] |
an email according to the HTML5 spec1 | email |
|
url | #[garde(url)] |
a URL | url |
ip | #[garde(ip)] |
an IP address (either IPv4 or IPv6) | - |
ipv4 | #[garde(ipv4)] |
an IPv4 address | - |
ipv6 | #[garde(ipv6)] |
an IPv6 address | - |
credit card | #[garde(credit_card)] |
a credit card number | credit-card |
phone number | #[garde(phone_number)] |
a phone number | phone-number |
length | #[garde(length(<mode>, min=<usize>, max=<usize>)] |
a container with length in min..=max |
- |
range | #[garde(range(min=<expr>, max=<expr>))] |
a number in the range min..=max |
- |
contains | #[garde(contains(<string>))] |
a string-like value containing a substring | - |
prefix | #[garde(prefix(<string>))] |
a string-like value prefixed by some string | - |
suffix | #[garde(suffix(<string>))] |
a string-like value suffixed by some string | - |
pattern | #[garde(pattern("<regex>"))] |
a string-like value matching some regular expression | regex |
pattern | #[garde(pattern(<matcher>))] |
a string-like value matched by some Matcher | - |
dive | #[garde(dive)] |
nested validation, calls validate on the value |
- |
skip | #[garde(skip)] |
skip validation | - |
custom | #[garde(custom(<function or closure>))] |
a custom validator | - |
Additional notes:
required
is only available forOption
fields.- For
length
andrange
, eithermin
ormax
may be omitted, but not both. length
andrange
use an inclusive upper bound (min..=max
).- The
<mode>
argument forlength
is explained here - For
contains
,prefix
, andsuffix
, the pattern must be a string literal, because thePattern
API is currently unstable. - Garde does not enable the default features of the
regex
crate - if you need extra regex features (e.g. Unicode) or better performance, add a dependency onregex = "1"
to yourCargo.toml
.
If most of the fields on your struct are annotated with #[garde(skip)]
, you may use #[garde(allow_unvalidated)]
instead:
#[derive(garde::Validate)]
struct Foo<'a> {
#[garde(length(min = 1))]
a: &'a str,
#[garde(skip)]
b: &'a str, // this field will not be validated
}
#[derive(garde::Validate)]
#[garde(allow_unvalidated)]
struct Bar<'a> {
#[garde(length(min = 1))]
a: &'a str,
b: &'a str, // this field will not be validated
// note the lack of `#[garde(skip)]`
}
The length
rule accepts an optional mode
argument, which determines what kind of length it will validate.
The options are:
simple
bytes
graphemes
utf16
chars
The simple
is the default used when the mode
argument is omitted. The meaning of "simple length"
depends on the type. It is currently implemented for strings, where it validates the number of bytes,
and std::collections
, where it validates the number of items.
#[derive(garde::Validate)]
struct Foo {
#[garde(length(min = 1, max = 100))]
string: String,
#[garde(length(min = 1, max = 100))]
collection: Vec<u32>
}
The bytes
, graphemes
, utf16
, and chars
exist mostly for string validation:
bytes
validates the number of bytesgraphemens
uses theunicode-segmentation
crate, and validates the number of graphemesutf16
usesencode_utf16
, and validates the number of UTF-16 code pointschars
useschars
, and validates the number of unicode scalar values
#[derive(garde::Validate)]
struct Foo {
#[garde(length(bytes, min = 1, max = 100))]
a: String, // `a.len()`
#[garde(length(graphemes, min = 1, max = 100))]
b: String, // `b.graphemes().count()`
#[garde(length(utf16, min = 1, max = 100))]
c: String, // `c.encode_utf16().count()`
#[garde(length(chars, min = 1, max = 100))]
d: String, // `d.chars().count()`
}
If you need to validate the "inner" type of a container, such as the String
in Vec<String>
, then use the inner
modifier:
#[derive(garde::Validate)]
struct Test {
#[garde(
length(min = 1),
inner(ascii, length(min = 1)), // wrap the rule in `inner`
)]
items: Vec<String>,
}
The above type would fail validation if:
- the
Vec
is empty - any of the inner
String
elements is empty - any of the inner
String
elements contains non-ASCII characters
The best way to re-use validation rules on a field is to use the newtype idiom
with #[garde(transparent)]
:
#[derive(garde::Validate)]
#[garde(transparent)]
struct Username(#[garde(length(min = 3, max = 20))] String);
#[derive(garde::Validate)]
struct User {
// later used with `dive`:
#[garde(dive)]
username: Username,
}
The username
field in the above example will inherit all the validation rules from the String
field on Username
. The result is that the error path will be flattened by one level, resulting in cleaner error messages:
User {
username: Username("")
}.validate()
"username: length is lower than 3"
Without the #[garde(transparent)]
attribute, it would instead be:
User {
username: Username("")
}.validate()
"username[0]: length is lower than 3"
Structs with the #[garde(transparent)]
attribute may have more than one field, but there must be only one unskipped field. That means every field other than the one you wish to validate must be #[garde(skip)]
.
Every rule works on Option<T>
fields. The field will only be validated if it is Some
. If you additionally want to validate that the Option<T>
field is Some
, use the required
rule:
#[derive(garde::Validate)]
struct Test {
#[garde(required, ascii, length(min = 1))]
value: Option<String>,
}
The above type would fail validation if:
value
isNone
- the inner
value
is empty - the inner
value
contains non-ASCII characters
Validation may be customized via the custom
rule, and the context
attribute.
The context may be any type without generic parameters. By default, the context is ()
.
#[derive(garde::Validate)]
#[garde(context(PasswordContext))]
struct User {
#[garde(custom(is_strong_password))]
password: String,
}
struct PasswordContext {
min_entropy: f32,
entropy: cracken::password_entropy::EntropyEstimator,
}
fn is_strong_password(value: &str, context: &PasswordContext) -> garde::Result {
let bits = context.entropy.estimate_password_entropy(value.as_bytes())
.map(|e| e.mask_entropy)
.unwrap_or(0.0);
if bits < context.min_entropy {
return Err(garde::Error::new("password is not strong enough"));
}
Ok(())
}
let ctx = PasswordContext { /* ... */ };
let user = User { /* ... */ };
user.validate(&ctx)?;
The validator function may accept the value as a reference to any type which it derefs to.
In the above example, it is possible to use &str
, because password
is a String
, and String
derefs to &str
.
The #[garde(custom(...))]
attribute accepts any expression which evalutes to a something which implements the following trait:
FnOnce(&T, &<T as Validate>::Context) -> garde::Result
That means it's possible to use higher order functions:
// Returns a function which does the actual validation.
fn my_equals(other: &str) -> impl FnOnce(&str, &()) -> garde::Result + '_ {
move |value, _| {
if value != other {
return Err(garde::Error::new(format!("not equal to {other}")));
}
Ok(())
}
}
#[derive(garde::Validate)]
struct User {
#[garde(length(min = 1, max = 255))]
password: String,
// Combined with `self` access in rules:
#[garde(custom(my_equals(&self.password2)))]
password2: String,
}
It's generally possible to also access the context and self
, because they are in scope in the output of the proc macro:
struct Limits {
min: usize,
max: usize,
}
struct Config {
username: Limits,
}
#[derive(garde::Validate)]
#[garde(context(Config as ctx))]
struct User {
#[garde(length(min = ctx.username.min, max = ctx.username.max))]
username: String,
}
Say you want to implement length checking for a custom string-like type.
To do this, you would implement one of the length
traits for it, depending
on what kind of validation you are looking for.
#[repr(transparent)]
pub struct MyString(String);
impl garde::rules::length::HasSimpleLength for MyString {
fn length(&self) -> usize {
self.0.len()
}
}
#[derive(garde::Validate)]
struct Foo {
// Now the `length` check may be used with `MyString`
#[garde(length(min = 1, max = 1000))]
field: MyString,
}
Each rule comes with its own trait that may be implemented by custom types in your code.
They are all available under garde::rules
.
In case you have a container type for which you'd like to support nested validation (using the #[garde(dive)]
rule),
you may implement Validate
for it:
#[repr(transparent)]
struct MyVec<T>(Vec<T>);
impl<T: garde::Validate> garde::Validate for MyVec<T> {
type Context = T::Context;
fn validate_into(
&self,
ctx: &Self::Context,
mut parent: &mut dyn FnMut() -> garde::Path,
report: &mut garde::Report
) {
for (index, item) in self.0.iter().enumerate() {
let mut path = garde::util::nested_path!(parent, index);
item.validate_into(ctx, &mut path, report);
}
}
}
#[derive(garde::Validate)]
struct Foo {
#[garde(dive)]
field: MyVec<Bar>,
}
#[derive(garde::Validate)]
struct Bar {
#[garde(range(min = 1, max = 10))]
value: u32,
}
Adapters allow you to implement validation for third-party types without using a newtype.
An adapter may look like this:
mod my_str_adapter {
#![allow(unused_imports)]
pub use garde::rules::*; // re-export garde's rules
pub mod length {
pub use garde::rules::length::*; // re-export `length` rules
pub mod simple {
// re-implement `simple`, but _only_ for the concrete type &str!
pub fn apply(v: &str, (min, max): (usize, usize)) -> garde::Result {
if !(min..=max).contains(&v.len()) {
Err(garde::Error::new("my custom error message"))
} else {
Ok(())
}
}
}
}
}
You create a module, add a public glob re-export of garde::rules
inside of it,
and then re-implement the specific rule you're interested in. This is a form of
duck typing. Any rule which you have
not re-implemented is simply delegated to garde
's impl.
It's quite verbose, but in exchange it is maximally flexible. To use the adapter,
add an adapt
attribute to a field:
#[derive(garde::Validate)]
struct Stuff<'a> {
#[garde(
adapt(my_str_adapter),
length(min = 1),
ascii,
)]
v: &'a str,
}
The length
rule will now use your custom implementation, but the ascii
rule
will continue to use garde
's implementation.
name | description | extra dependencies |
---|---|---|
derive |
Enables the usage of the derive(Validate) macro |
garde_derive |
url |
Validation of URLs via the url crate. |
url |
email |
Validation of emails according to HTML5 | regex , once_cell |
email-idna |
Support for Internationalizing Domain Names for Applications in email addresses | idna |
regex |
Support for regular expressions in pattern via the regex crate |
regex , once_cell |
credit-card |
Validation of credit card numbers via the card-validate crate |
card-validate |
phone-number |
Validation of phone numbers via the phonenumber crate |
phonenumber |
unicode |
Validation of grapheme count via the unicode-segmentation crate |
unicode-segmentation |
Garde means guard in French. I am not French, nor do I speak the language, but guard
was taken, and this is close enough :).
Contributing to garde
only requires a somewhat recent version of Rust
.
This repository also makes use of the following tools, but they are optional:
insta
for snapshot testing (tests/rules).just
for running recipes defined in thejustfile
. Runjust -l
to see what recipes are available.
Licensed under either of
- Apache License, Version 2.0 (LICENSE-APACHE or http://www.apache.org/licenses/LICENSE-2.0)
- MIT license (LICENSE-MIT or http://opensource.org/licenses/MIT)
at your option.
Unless you explicitly state otherwise, any contribution intentionally submitted for inclusion in the work by you, as defined in the Apache-2.0 license, shall be dual licensed as above, without any additional terms or conditions.
This crate is heavily inspired by the validator crate. It is essentially a full rewrite of validator
.
The creation of this crate was prompted by this comment
and a few others talking about a potential rewrite.