mirror of
https://github.com/uuid-rs/uuid.git
synced 2025-09-30 06:21:02 +00:00
1846 lines
52 KiB
Rust
1846 lines
52 KiB
Rust
// Copyright 2013-2014 The Rust Project Developers.
|
|
// Copyright 2018 The Uuid Project Developers.
|
|
//
|
|
// See the COPYRIGHT file at the top-level directory of this distribution.
|
|
//
|
|
// Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or
|
|
// http://www.apache.org/licenses/LICENSE-2.0> or the MIT license
|
|
// <LICENSE-MIT or http://opensource.org/licenses/MIT>, at your
|
|
// option. This file may not be copied, modified, or distributed
|
|
// except according to those terms.
|
|
|
|
//! Generate and parse universally unique identifiers (UUIDs).
|
|
//!
|
|
//! Here's an example of a UUID:
|
|
//!
|
|
//! ```text
|
|
//! 67e55044-10b1-426f-9247-bb680e5fe0c8
|
|
//! ```
|
|
//!
|
|
//! A UUID is a unique 128-bit value, stored as 16 octets, and regularly
|
|
//! formatted as a hex string in five groups. UUIDs are used to assign unique
|
|
//! identifiers to entities without requiring a central allocating authority.
|
|
//!
|
|
//! They are particularly useful in distributed systems, though can be used in
|
|
//! disparate areas, such as databases and network protocols. Typically a UUID
|
|
//! is displayed in a readable string form as a sequence of hexadecimal digits,
|
|
//! separated into groups by hyphens.
|
|
//!
|
|
//! The uniqueness property is not strictly guaranteed, however for all
|
|
//! practical purposes, it can be assumed that an unintentional collision would
|
|
//! be extremely unlikely.
|
|
//!
|
|
//! UUIDs have a number of standardized encodings that are specified in [RFC4122](http://tools.ietf.org/html/rfc4122),
|
|
//! with recent additions [in draft](https://datatracker.ietf.org/doc/html/draft-peabody-dispatch-new-uuid-format-04).
|
|
//!
|
|
//! # Getting started
|
|
//!
|
|
//! Add the following to your `Cargo.toml`:
|
|
//!
|
|
//! ```toml
|
|
//! [dependencies.uuid]
|
|
//! version = "1.6.0"
|
|
//! features = [
|
|
//! "v4", # Lets you generate random UUIDs
|
|
//! "fast-rng", # Use a faster (but still sufficiently random) RNG
|
|
//! "macro-diagnostics", # Enable better diagnostics for compile-time UUIDs
|
|
//! ]
|
|
//! ```
|
|
//!
|
|
//! When you want a UUID, you can generate one:
|
|
//!
|
|
//! ```
|
|
//! # fn main() {
|
|
//! # #[cfg(feature = "v4")]
|
|
//! # {
|
|
//! use uuid::Uuid;
|
|
//!
|
|
//! let id = Uuid::new_v4();
|
|
//! # }
|
|
//! # }
|
|
//! ```
|
|
//!
|
|
//! If you have a UUID value, you can use its string literal form inline:
|
|
//!
|
|
//! ```
|
|
//! use uuid::{uuid, Uuid};
|
|
//!
|
|
//! const ID: Uuid = uuid!("67e55044-10b1-426f-9247-bb680e5fe0c8");
|
|
//! ```
|
|
//!
|
|
//! # Working with different UUID versions
|
|
//!
|
|
//! This library supports all standardized methods for generating UUIDs through individual Cargo features.
|
|
//!
|
|
//! By default, this crate depends on nothing but the Rust standard library and can parse and format
|
|
//! UUIDs, but cannot generate them. Depending on the kind of UUID you'd like to work with, there
|
|
//! are Cargo features that enable generating them:
|
|
//!
|
|
//! * `v1` - Version 1 UUIDs using a timestamp and monotonic counter.
|
|
//! * `v3` - Version 3 UUIDs based on the MD5 hash of some data.
|
|
//! * `v4` - Version 4 UUIDs with random data.
|
|
//! * `v5` - Version 5 UUIDs based on the SHA1 hash of some data.
|
|
//! * `v6` - Version 6 UUIDs using a timestamp and monotonic counter.
|
|
//! * `v7` - Version 7 UUIDs using a Unix timestamp.
|
|
//! * `v8` - Version 8 UUIDs using user-defined data.
|
|
//!
|
|
//! Versions that are in draft are also supported. See the _unstable features_ section for details.
|
|
//!
|
|
//! This library also includes a [`Builder`] type that can be used to help construct UUIDs of any
|
|
//! version without any additional dependencies or features. It's a lower-level API than [`Uuid`]
|
|
//! that can be used when you need control over implicit requirements on things like a source
|
|
//! of randomness.
|
|
//!
|
|
//! ## Which UUID version should I use?
|
|
//!
|
|
//! If you just want to generate unique identifiers then consider version 4 (`v4`) UUIDs. If you want
|
|
//! to use UUIDs as database keys or need to sort them then consider version 7 (`v7`) UUIDs.
|
|
//! Other versions should generally be avoided unless there's an existing need for them.
|
|
//!
|
|
//! Some UUID versions supersede others. Prefer version 6 over version 1 and version 5 over version 3.
|
|
//!
|
|
//! # Other features
|
|
//!
|
|
//! Other crate features can also be useful beyond the version support:
|
|
//!
|
|
//! * `macro-diagnostics` - enhances the diagnostics of `uuid!` macro.
|
|
//! * `serde` - adds the ability to serialize and deserialize a UUID using
|
|
//! `serde`.
|
|
//! * `arbitrary` - adds an `Arbitrary` trait implementation to `Uuid` for
|
|
//! fuzzing.
|
|
//! * `fast-rng` - uses a faster algorithm for generating random UUIDs.
|
|
//! This feature requires more dependencies to compile, but is just as suitable for
|
|
//! UUIDs as the default algorithm.
|
|
//! * `bytemuck` - adds a `Pod` trait implementation to `Uuid` for byte manipulation
|
|
//!
|
|
//! # Unstable features
|
|
//!
|
|
//! Some features are unstable. They may be incomplete or depend on other
|
|
//! unstable libraries. These include:
|
|
//!
|
|
//! * `zerocopy` - adds support for zero-copy deserialization using the
|
|
//! `zerocopy` library.
|
|
//! * `borsh` - adds the ability to serialize and deserialize a UUID using
|
|
//! `borsh`.
|
|
//!
|
|
//! Unstable features may break between minor releases.
|
|
//!
|
|
//! To allow unstable features, you'll need to enable the Cargo feature as
|
|
//! normal, but also pass an additional flag through your environment to opt-in
|
|
//! to unstable `uuid` features:
|
|
//!
|
|
//! ```text
|
|
//! RUSTFLAGS="--cfg uuid_unstable"
|
|
//! ```
|
|
//!
|
|
//! # Building for other targets
|
|
//!
|
|
//! ## WebAssembly
|
|
//!
|
|
//! For WebAssembly, enable the `js` feature:
|
|
//!
|
|
//! ```toml
|
|
//! [dependencies.uuid]
|
|
//! version = "1.6.0"
|
|
//! features = [
|
|
//! "v4",
|
|
//! "v7",
|
|
//! "js",
|
|
//! ]
|
|
//! ```
|
|
//!
|
|
//! ## Embedded
|
|
//!
|
|
//! For embedded targets without the standard library, you'll need to
|
|
//! disable default features when building `uuid`:
|
|
//!
|
|
//! ```toml
|
|
//! [dependencies.uuid]
|
|
//! version = "1.6.0"
|
|
//! default-features = false
|
|
//! ```
|
|
//!
|
|
//! Some additional features are supported in no-std environments:
|
|
//!
|
|
//! * `v1`, `v3`, `v5`, `v6`, and `v8`.
|
|
//! * `serde`.
|
|
//!
|
|
//! If you need to use `v4` or `v7` in a no-std environment, you'll need to
|
|
//! follow [`getrandom`'s docs] on configuring a source of randomness
|
|
//! on currently unsupported targets. Alternatively, you can produce
|
|
//! random bytes yourself and then pass them to [`Builder::from_random_bytes`]
|
|
//! without enabling the `v4` feature.
|
|
//!
|
|
//! # Examples
|
|
//!
|
|
//! Parse a UUID given in the simple format and print it as a URN:
|
|
//!
|
|
//! ```
|
|
//! # use uuid::Uuid;
|
|
//! # fn main() -> Result<(), uuid::Error> {
|
|
//! let my_uuid = Uuid::parse_str("a1a2a3a4b1b2c1c2d1d2d3d4d5d6d7d8")?;
|
|
//!
|
|
//! println!("{}", my_uuid.urn());
|
|
//! # Ok(())
|
|
//! # }
|
|
//! ```
|
|
//!
|
|
//! Generate a random UUID and print it out in hexadecimal form:
|
|
//!
|
|
//! ```
|
|
//! // Note that this requires the `v4` feature to be enabled.
|
|
//! # use uuid::Uuid;
|
|
//! # fn main() {
|
|
//! # #[cfg(feature = "v4")] {
|
|
//! let my_uuid = Uuid::new_v4();
|
|
//!
|
|
//! println!("{}", my_uuid);
|
|
//! # }
|
|
//! # }
|
|
//! ```
|
|
//!
|
|
//! # References
|
|
//!
|
|
//! * [Wikipedia: Universally Unique Identifier](http://en.wikipedia.org/wiki/Universally_unique_identifier)
|
|
//! * [RFC4122: A Universally Unique Identifier (UUID) URN Namespace](http://tools.ietf.org/html/rfc4122)
|
|
//! * [Draft RFC: New UUID Formats, Version 4](https://datatracker.ietf.org/doc/html/draft-peabody-dispatch-new-uuid-format-04)
|
|
//!
|
|
//! [`wasm-bindgen`]: https://crates.io/crates/wasm-bindgen
|
|
//! [`cargo-web`]: https://crates.io/crates/cargo-web
|
|
//! [`getrandom`'s docs]: https://docs.rs/getrandom
|
|
|
|
#![no_std]
|
|
#![deny(missing_debug_implementations, missing_docs)]
|
|
#![doc(
|
|
html_logo_url = "https://www.rust-lang.org/logos/rust-logo-128x128-blk-v2.png",
|
|
html_favicon_url = "https://www.rust-lang.org/favicon.ico",
|
|
html_root_url = "https://docs.rs/uuid/1.6.0"
|
|
)]
|
|
|
|
#[cfg(any(feature = "std", test))]
|
|
#[macro_use]
|
|
extern crate std;
|
|
|
|
#[cfg(all(not(feature = "std"), not(test)))]
|
|
#[macro_use]
|
|
extern crate core as std;
|
|
|
|
#[cfg(all(uuid_unstable, feature = "zerocopy"))]
|
|
use zerocopy::{AsBytes, FromBytes, Unaligned};
|
|
|
|
mod builder;
|
|
mod error;
|
|
mod parser;
|
|
|
|
pub mod fmt;
|
|
pub mod timestamp;
|
|
|
|
pub use timestamp::{context::NoContext, ClockSequence, Timestamp};
|
|
|
|
#[cfg(any(feature = "v1", feature = "v6"))]
|
|
pub use timestamp::context::Context;
|
|
|
|
#[cfg(feature = "v1")]
|
|
#[doc(hidden)]
|
|
// Soft-deprecated (Rust doesn't support deprecating re-exports)
|
|
// Use `Context` from the crate root instead
|
|
pub mod v1;
|
|
#[cfg(feature = "v3")]
|
|
mod v3;
|
|
#[cfg(feature = "v4")]
|
|
mod v4;
|
|
#[cfg(feature = "v5")]
|
|
mod v5;
|
|
#[cfg(feature = "v6")]
|
|
mod v6;
|
|
#[cfg(feature = "v7")]
|
|
mod v7;
|
|
#[cfg(feature = "v8")]
|
|
mod v8;
|
|
|
|
#[cfg(feature = "md5")]
|
|
mod md5;
|
|
#[cfg(feature = "rng")]
|
|
mod rng;
|
|
#[cfg(feature = "sha1")]
|
|
mod sha1;
|
|
|
|
mod external;
|
|
|
|
#[macro_use]
|
|
mod macros;
|
|
|
|
#[doc(hidden)]
|
|
#[cfg(feature = "macro-diagnostics")]
|
|
pub extern crate uuid_macro_internal;
|
|
|
|
#[doc(hidden)]
|
|
pub mod __macro_support {
|
|
pub use crate::std::result::Result::{Err, Ok};
|
|
}
|
|
|
|
use crate::std::convert;
|
|
|
|
pub use crate::{builder::Builder, error::Error};
|
|
|
|
/// A 128-bit (16 byte) buffer containing the UUID.
|
|
///
|
|
/// # ABI
|
|
///
|
|
/// The `Bytes` type is always guaranteed to be have the same ABI as [`Uuid`].
|
|
pub type Bytes = [u8; 16];
|
|
|
|
/// The version of the UUID, denoting the generating algorithm.
|
|
///
|
|
/// # References
|
|
///
|
|
/// * [Version in RFC4122](https://datatracker.ietf.org/doc/html/rfc4122#section-4.1.3)
|
|
#[derive(Clone, Copy, Debug, PartialEq)]
|
|
#[non_exhaustive]
|
|
#[repr(u8)]
|
|
pub enum Version {
|
|
/// The "nil" (all zeros) UUID.
|
|
Nil = 0u8,
|
|
/// Version 1: Timestamp and node ID.
|
|
Mac = 1,
|
|
/// Version 2: DCE Security.
|
|
Dce = 2,
|
|
/// Version 3: MD5 hash.
|
|
Md5 = 3,
|
|
/// Version 4: Random.
|
|
Random = 4,
|
|
/// Version 5: SHA-1 hash.
|
|
Sha1 = 5,
|
|
/// Version 6: Sortable Timestamp and node ID.
|
|
SortMac = 6,
|
|
/// Version 7: Timestamp and random.
|
|
SortRand = 7,
|
|
/// Version 8: Custom.
|
|
Custom = 8,
|
|
/// The "max" (all ones) UUID.
|
|
Max = 0xff,
|
|
}
|
|
|
|
/// The reserved variants of UUIDs.
|
|
///
|
|
/// # References
|
|
///
|
|
/// * [Variant in RFC4122](http://tools.ietf.org/html/rfc4122#section-4.1.1)
|
|
#[derive(Clone, Copy, Debug, PartialEq)]
|
|
#[non_exhaustive]
|
|
#[repr(u8)]
|
|
pub enum Variant {
|
|
/// Reserved by the NCS for backward compatibility.
|
|
NCS = 0u8,
|
|
/// As described in the RFC4122 Specification (default).
|
|
RFC4122,
|
|
/// Reserved by Microsoft for backward compatibility.
|
|
Microsoft,
|
|
/// Reserved for future expansion.
|
|
Future,
|
|
}
|
|
|
|
/// A Universally Unique Identifier (UUID).
|
|
///
|
|
/// # Examples
|
|
///
|
|
/// Parse a UUID given in the simple format and print it as a urn:
|
|
///
|
|
/// ```
|
|
/// # use uuid::Uuid;
|
|
/// # fn main() -> Result<(), uuid::Error> {
|
|
/// let my_uuid = Uuid::parse_str("a1a2a3a4b1b2c1c2d1d2d3d4d5d6d7d8")?;
|
|
///
|
|
/// println!("{}", my_uuid.urn());
|
|
/// # Ok(())
|
|
/// # }
|
|
/// ```
|
|
///
|
|
/// Create a new random (V4) UUID and print it out in hexadecimal form:
|
|
///
|
|
/// ```
|
|
/// // Note that this requires the `v4` feature enabled in the uuid crate.
|
|
/// # use uuid::Uuid;
|
|
/// # fn main() {
|
|
/// # #[cfg(feature = "v4")] {
|
|
/// let my_uuid = Uuid::new_v4();
|
|
///
|
|
/// println!("{}", my_uuid);
|
|
/// # }
|
|
/// # }
|
|
/// ```
|
|
///
|
|
/// # Formatting
|
|
///
|
|
/// A UUID can be formatted in one of a few ways:
|
|
///
|
|
/// * [`simple`](#method.simple): `a1a2a3a4b1b2c1c2d1d2d3d4d5d6d7d8`.
|
|
/// * [`hyphenated`](#method.hyphenated):
|
|
/// `a1a2a3a4-b1b2-c1c2-d1d2-d3d4d5d6d7d8`.
|
|
/// * [`urn`](#method.urn): `urn:uuid:A1A2A3A4-B1B2-C1C2-D1D2-D3D4D5D6D7D8`.
|
|
/// * [`braced`](#method.braced): `{a1a2a3a4-b1b2-c1c2-d1d2-d3d4d5d6d7d8}`.
|
|
///
|
|
/// The default representation when formatting a UUID with `Display` is
|
|
/// hyphenated:
|
|
///
|
|
/// ```
|
|
/// # use uuid::Uuid;
|
|
/// # fn main() -> Result<(), uuid::Error> {
|
|
/// let my_uuid = Uuid::parse_str("a1a2a3a4b1b2c1c2d1d2d3d4d5d6d7d8")?;
|
|
///
|
|
/// assert_eq!(
|
|
/// "a1a2a3a4-b1b2-c1c2-d1d2-d3d4d5d6d7d8",
|
|
/// my_uuid.to_string(),
|
|
/// );
|
|
/// # Ok(())
|
|
/// # }
|
|
/// ```
|
|
///
|
|
/// Other formats can be specified using adapter methods on the UUID:
|
|
///
|
|
/// ```
|
|
/// # use uuid::Uuid;
|
|
/// # fn main() -> Result<(), uuid::Error> {
|
|
/// let my_uuid = Uuid::parse_str("a1a2a3a4b1b2c1c2d1d2d3d4d5d6d7d8")?;
|
|
///
|
|
/// assert_eq!(
|
|
/// "urn:uuid:a1a2a3a4-b1b2-c1c2-d1d2-d3d4d5d6d7d8",
|
|
/// my_uuid.urn().to_string(),
|
|
/// );
|
|
/// # Ok(())
|
|
/// # }
|
|
/// ```
|
|
///
|
|
/// # Endianness
|
|
///
|
|
/// The specification for UUIDs encodes the integer fields that make up the
|
|
/// value in big-endian order. This crate assumes integer inputs are already in
|
|
/// the correct order by default, regardless of the endianness of the
|
|
/// environment. Most methods that accept integers have a `_le` variant (such as
|
|
/// `from_fields_le`) that assumes any integer values will need to have their
|
|
/// bytes flipped, regardless of the endianness of the environment.
|
|
///
|
|
/// Most users won't need to worry about endianness unless they need to operate
|
|
/// on individual fields (such as when converting between Microsoft GUIDs). The
|
|
/// important things to remember are:
|
|
///
|
|
/// - The endianness is in terms of the fields of the UUID, not the environment.
|
|
/// - The endianness is assumed to be big-endian when there's no `_le` suffix
|
|
/// somewhere.
|
|
/// - Byte-flipping in `_le` methods applies to each integer.
|
|
/// - Endianness roundtrips, so if you create a UUID with `from_fields_le`
|
|
/// you'll get the same values back out with `to_fields_le`.
|
|
///
|
|
/// # ABI
|
|
///
|
|
/// The `Uuid` type is always guaranteed to be have the same ABI as [`Bytes`].
|
|
#[derive(Clone, Copy, Eq, Hash, Ord, PartialEq, PartialOrd)]
|
|
#[cfg_attr(
|
|
all(uuid_unstable, feature = "zerocopy"),
|
|
derive(AsBytes, FromBytes, Unaligned)
|
|
)]
|
|
#[cfg_attr(
|
|
all(uuid_unstable, feature = "borsh"),
|
|
derive(borsh::BorshDeserialize, borsh::BorshSerialize)
|
|
)]
|
|
#[repr(transparent)]
|
|
#[cfg_attr(
|
|
feature = "bytemuck",
|
|
derive(bytemuck::Zeroable, bytemuck::Pod, bytemuck::TransparentWrapper)
|
|
)]
|
|
pub struct Uuid(Bytes);
|
|
|
|
impl Uuid {
|
|
/// UUID namespace for Domain Name System (DNS).
|
|
pub const NAMESPACE_DNS: Self = Uuid([
|
|
0x6b, 0xa7, 0xb8, 0x10, 0x9d, 0xad, 0x11, 0xd1, 0x80, 0xb4, 0x00, 0xc0, 0x4f, 0xd4, 0x30,
|
|
0xc8,
|
|
]);
|
|
|
|
/// UUID namespace for ISO Object Identifiers (OIDs).
|
|
pub const NAMESPACE_OID: Self = Uuid([
|
|
0x6b, 0xa7, 0xb8, 0x12, 0x9d, 0xad, 0x11, 0xd1, 0x80, 0xb4, 0x00, 0xc0, 0x4f, 0xd4, 0x30,
|
|
0xc8,
|
|
]);
|
|
|
|
/// UUID namespace for Uniform Resource Locators (URLs).
|
|
pub const NAMESPACE_URL: Self = Uuid([
|
|
0x6b, 0xa7, 0xb8, 0x11, 0x9d, 0xad, 0x11, 0xd1, 0x80, 0xb4, 0x00, 0xc0, 0x4f, 0xd4, 0x30,
|
|
0xc8,
|
|
]);
|
|
|
|
/// UUID namespace for X.500 Distinguished Names (DNs).
|
|
pub const NAMESPACE_X500: Self = Uuid([
|
|
0x6b, 0xa7, 0xb8, 0x14, 0x9d, 0xad, 0x11, 0xd1, 0x80, 0xb4, 0x00, 0xc0, 0x4f, 0xd4, 0x30,
|
|
0xc8,
|
|
]);
|
|
|
|
/// Returns the variant of the UUID structure.
|
|
///
|
|
/// This determines the interpretation of the structure of the UUID.
|
|
/// This method simply reads the value of the variant byte. It doesn't
|
|
/// validate the rest of the UUID as conforming to that variant.
|
|
///
|
|
/// # Examples
|
|
///
|
|
/// Basic usage:
|
|
///
|
|
/// ```
|
|
/// # use uuid::{Uuid, Variant};
|
|
/// # fn main() -> Result<(), uuid::Error> {
|
|
/// let my_uuid = Uuid::parse_str("02f09a3f-1624-3b1d-8409-44eff7708208")?;
|
|
///
|
|
/// assert_eq!(Variant::RFC4122, my_uuid.get_variant());
|
|
/// # Ok(())
|
|
/// # }
|
|
/// ```
|
|
///
|
|
/// # References
|
|
///
|
|
/// * [Variant in RFC4122](http://tools.ietf.org/html/rfc4122#section-4.1.1)
|
|
pub const fn get_variant(&self) -> Variant {
|
|
match self.as_bytes()[8] {
|
|
x if x & 0x80 == 0x00 => Variant::NCS,
|
|
x if x & 0xc0 == 0x80 => Variant::RFC4122,
|
|
x if x & 0xe0 == 0xc0 => Variant::Microsoft,
|
|
x if x & 0xe0 == 0xe0 => Variant::Future,
|
|
// The above match arms are actually exhaustive
|
|
// We just return `Future` here because we can't
|
|
// use `unreachable!()` in a `const fn`
|
|
_ => Variant::Future,
|
|
}
|
|
}
|
|
|
|
/// Returns the version number of the UUID.
|
|
///
|
|
/// This represents the algorithm used to generate the value.
|
|
/// This method is the future-proof alternative to [`Uuid::get_version`].
|
|
///
|
|
/// # Examples
|
|
///
|
|
/// Basic usage:
|
|
///
|
|
/// ```
|
|
/// # use uuid::Uuid;
|
|
/// # fn main() -> Result<(), uuid::Error> {
|
|
/// let my_uuid = Uuid::parse_str("02f09a3f-1624-3b1d-8409-44eff7708208")?;
|
|
///
|
|
/// assert_eq!(3, my_uuid.get_version_num());
|
|
/// # Ok(())
|
|
/// # }
|
|
/// ```
|
|
///
|
|
/// # References
|
|
///
|
|
/// * [Version in RFC4122](https://datatracker.ietf.org/doc/html/rfc4122#section-4.1.3)
|
|
pub const fn get_version_num(&self) -> usize {
|
|
(self.as_bytes()[6] >> 4) as usize
|
|
}
|
|
|
|
/// Returns the version of the UUID.
|
|
///
|
|
/// This represents the algorithm used to generate the value.
|
|
/// If the version field doesn't contain a recognized version then `None`
|
|
/// is returned. If you're trying to read the version for a future extension
|
|
/// you can also use [`Uuid::get_version_num`] to unconditionally return a
|
|
/// number. Future extensions may start to return `Some` once they're
|
|
/// standardized and supported.
|
|
///
|
|
/// # Examples
|
|
///
|
|
/// Basic usage:
|
|
///
|
|
/// ```
|
|
/// # use uuid::{Uuid, Version};
|
|
/// # fn main() -> Result<(), uuid::Error> {
|
|
/// let my_uuid = Uuid::parse_str("02f09a3f-1624-3b1d-8409-44eff7708208")?;
|
|
///
|
|
/// assert_eq!(Some(Version::Md5), my_uuid.get_version());
|
|
/// # Ok(())
|
|
/// # }
|
|
/// ```
|
|
///
|
|
/// # References
|
|
///
|
|
/// * [Version in RFC4122](https://datatracker.ietf.org/doc/html/rfc4122#section-4.1.3)
|
|
pub const fn get_version(&self) -> Option<Version> {
|
|
match self.get_version_num() {
|
|
0 if self.is_nil() => Some(Version::Nil),
|
|
1 => Some(Version::Mac),
|
|
2 => Some(Version::Dce),
|
|
3 => Some(Version::Md5),
|
|
4 => Some(Version::Random),
|
|
5 => Some(Version::Sha1),
|
|
6 => Some(Version::SortMac),
|
|
7 => Some(Version::SortRand),
|
|
8 => Some(Version::Custom),
|
|
0xf => Some(Version::Max),
|
|
_ => None,
|
|
}
|
|
}
|
|
|
|
/// Returns the four field values of the UUID.
|
|
///
|
|
/// These values can be passed to the [`Uuid::from_fields`] method to get
|
|
/// the original `Uuid` back.
|
|
///
|
|
/// * The first field value represents the first group of (eight) hex
|
|
/// digits, taken as a big-endian `u32` value. For V1 UUIDs, this field
|
|
/// represents the low 32 bits of the timestamp.
|
|
/// * The second field value represents the second group of (four) hex
|
|
/// digits, taken as a big-endian `u16` value. For V1 UUIDs, this field
|
|
/// represents the middle 16 bits of the timestamp.
|
|
/// * The third field value represents the third group of (four) hex digits,
|
|
/// taken as a big-endian `u16` value. The 4 most significant bits give
|
|
/// the UUID version, and for V1 UUIDs, the last 12 bits represent the
|
|
/// high 12 bits of the timestamp.
|
|
/// * The last field value represents the last two groups of four and twelve
|
|
/// hex digits, taken in order. The first 1-3 bits of this indicate the
|
|
/// UUID variant, and for V1 UUIDs, the next 13-15 bits indicate the clock
|
|
/// sequence and the last 48 bits indicate the node ID.
|
|
///
|
|
/// # Examples
|
|
///
|
|
/// ```
|
|
/// # use uuid::Uuid;
|
|
/// # fn main() -> Result<(), uuid::Error> {
|
|
/// let uuid = Uuid::nil();
|
|
///
|
|
/// assert_eq!(uuid.as_fields(), (0, 0, 0, &[0u8; 8]));
|
|
///
|
|
/// let uuid = Uuid::parse_str("a1a2a3a4-b1b2-c1c2-d1d2-d3d4d5d6d7d8")?;
|
|
///
|
|
/// assert_eq!(
|
|
/// uuid.as_fields(),
|
|
/// (
|
|
/// 0xa1a2a3a4,
|
|
/// 0xb1b2,
|
|
/// 0xc1c2,
|
|
/// &[0xd1, 0xd2, 0xd3, 0xd4, 0xd5, 0xd6, 0xd7, 0xd8],
|
|
/// )
|
|
/// );
|
|
/// # Ok(())
|
|
/// # }
|
|
/// ```
|
|
pub fn as_fields(&self) -> (u32, u16, u16, &[u8; 8]) {
|
|
let bytes = self.as_bytes();
|
|
|
|
let d1 = (bytes[0] as u32) << 24
|
|
| (bytes[1] as u32) << 16
|
|
| (bytes[2] as u32) << 8
|
|
| (bytes[3] as u32);
|
|
|
|
let d2 = (bytes[4] as u16) << 8 | (bytes[5] as u16);
|
|
|
|
let d3 = (bytes[6] as u16) << 8 | (bytes[7] as u16);
|
|
|
|
let d4: &[u8; 8] = convert::TryInto::try_into(&bytes[8..16]).unwrap();
|
|
(d1, d2, d3, d4)
|
|
}
|
|
|
|
/// Returns the four field values of the UUID in little-endian order.
|
|
///
|
|
/// The bytes in the returned integer fields will be converted from
|
|
/// big-endian order. This is based on the endianness of the UUID,
|
|
/// rather than the target environment so bytes will be flipped on both
|
|
/// big and little endian machines.
|
|
///
|
|
/// # Examples
|
|
///
|
|
/// ```
|
|
/// use uuid::Uuid;
|
|
///
|
|
/// # fn main() -> Result<(), uuid::Error> {
|
|
/// let uuid = Uuid::parse_str("a1a2a3a4-b1b2-c1c2-d1d2-d3d4d5d6d7d8")?;
|
|
///
|
|
/// assert_eq!(
|
|
/// uuid.to_fields_le(),
|
|
/// (
|
|
/// 0xa4a3a2a1,
|
|
/// 0xb2b1,
|
|
/// 0xc2c1,
|
|
/// &[0xd1, 0xd2, 0xd3, 0xd4, 0xd5, 0xd6, 0xd7, 0xd8],
|
|
/// )
|
|
/// );
|
|
/// # Ok(())
|
|
/// # }
|
|
/// ```
|
|
pub fn to_fields_le(&self) -> (u32, u16, u16, &[u8; 8]) {
|
|
let d1 = (self.as_bytes()[0] as u32)
|
|
| (self.as_bytes()[1] as u32) << 8
|
|
| (self.as_bytes()[2] as u32) << 16
|
|
| (self.as_bytes()[3] as u32) << 24;
|
|
|
|
let d2 = (self.as_bytes()[4] as u16) | (self.as_bytes()[5] as u16) << 8;
|
|
|
|
let d3 = (self.as_bytes()[6] as u16) | (self.as_bytes()[7] as u16) << 8;
|
|
|
|
let d4: &[u8; 8] = convert::TryInto::try_into(&self.as_bytes()[8..16]).unwrap();
|
|
(d1, d2, d3, d4)
|
|
}
|
|
|
|
/// Returns a 128bit value containing the value.
|
|
///
|
|
/// The bytes in the UUID will be packed directly into a `u128`.
|
|
///
|
|
/// # Examples
|
|
///
|
|
/// ```
|
|
/// # use uuid::Uuid;
|
|
/// # fn main() -> Result<(), uuid::Error> {
|
|
/// let uuid = Uuid::parse_str("a1a2a3a4-b1b2-c1c2-d1d2-d3d4d5d6d7d8")?;
|
|
///
|
|
/// assert_eq!(
|
|
/// uuid.as_u128(),
|
|
/// 0xa1a2a3a4b1b2c1c2d1d2d3d4d5d6d7d8,
|
|
/// );
|
|
/// # Ok(())
|
|
/// # }
|
|
/// ```
|
|
pub const fn as_u128(&self) -> u128 {
|
|
u128::from_be_bytes(*self.as_bytes())
|
|
}
|
|
|
|
/// Returns a 128bit little-endian value containing the value.
|
|
///
|
|
/// The bytes in the `u128` will be flipped to convert into big-endian
|
|
/// order. This is based on the endianness of the UUID, rather than the
|
|
/// target environment so bytes will be flipped on both big and little
|
|
/// endian machines.
|
|
///
|
|
/// Note that this will produce a different result than
|
|
/// [`Uuid::to_fields_le`], because the entire UUID is reversed, rather
|
|
/// than reversing the individual fields in-place.
|
|
///
|
|
/// # Examples
|
|
///
|
|
/// ```
|
|
/// # use uuid::Uuid;
|
|
/// # fn main() -> Result<(), uuid::Error> {
|
|
/// let uuid = Uuid::parse_str("a1a2a3a4-b1b2-c1c2-d1d2-d3d4d5d6d7d8")?;
|
|
///
|
|
/// assert_eq!(
|
|
/// uuid.to_u128_le(),
|
|
/// 0xd8d7d6d5d4d3d2d1c2c1b2b1a4a3a2a1,
|
|
/// );
|
|
/// # Ok(())
|
|
/// # }
|
|
/// ```
|
|
pub const fn to_u128_le(&self) -> u128 {
|
|
u128::from_le_bytes(*self.as_bytes())
|
|
}
|
|
|
|
/// Returns two 64bit values containing the value.
|
|
///
|
|
/// The bytes in the UUID will be split into two `u64`.
|
|
/// The first u64 represents the 64 most significant bits,
|
|
/// the second one represents the 64 least significant.
|
|
///
|
|
/// # Examples
|
|
///
|
|
/// ```
|
|
/// # use uuid::Uuid;
|
|
/// # fn main() -> Result<(), uuid::Error> {
|
|
/// let uuid = Uuid::parse_str("a1a2a3a4-b1b2-c1c2-d1d2-d3d4d5d6d7d8")?;
|
|
/// assert_eq!(
|
|
/// uuid.as_u64_pair(),
|
|
/// (0xa1a2a3a4b1b2c1c2, 0xd1d2d3d4d5d6d7d8),
|
|
/// );
|
|
/// # Ok(())
|
|
/// # }
|
|
/// ```
|
|
pub const fn as_u64_pair(&self) -> (u64, u64) {
|
|
let value = self.as_u128();
|
|
((value >> 64) as u64, value as u64)
|
|
}
|
|
|
|
/// Returns a slice of 16 octets containing the value.
|
|
///
|
|
/// This method borrows the underlying byte value of the UUID.
|
|
///
|
|
/// # Examples
|
|
///
|
|
/// ```
|
|
/// # use uuid::Uuid;
|
|
/// let bytes1 = [
|
|
/// 0xa1, 0xa2, 0xa3, 0xa4,
|
|
/// 0xb1, 0xb2,
|
|
/// 0xc1, 0xc2,
|
|
/// 0xd1, 0xd2, 0xd3, 0xd4, 0xd5, 0xd6, 0xd7, 0xd8,
|
|
/// ];
|
|
/// let uuid1 = Uuid::from_bytes_ref(&bytes1);
|
|
///
|
|
/// let bytes2 = uuid1.as_bytes();
|
|
/// let uuid2 = Uuid::from_bytes_ref(bytes2);
|
|
///
|
|
/// assert_eq!(uuid1, uuid2);
|
|
///
|
|
/// assert!(std::ptr::eq(
|
|
/// uuid2 as *const Uuid as *const u8,
|
|
/// &bytes1 as *const [u8; 16] as *const u8,
|
|
/// ));
|
|
/// ```
|
|
#[inline]
|
|
pub const fn as_bytes(&self) -> &Bytes {
|
|
&self.0
|
|
}
|
|
|
|
/// Consumes self and returns the underlying byte value of the UUID.
|
|
///
|
|
/// # Examples
|
|
///
|
|
/// ```
|
|
/// # use uuid::Uuid;
|
|
/// let bytes = [
|
|
/// 0xa1, 0xa2, 0xa3, 0xa4,
|
|
/// 0xb1, 0xb2,
|
|
/// 0xc1, 0xc2,
|
|
/// 0xd1, 0xd2, 0xd3, 0xd4, 0xd5, 0xd6, 0xd7, 0xd8,
|
|
/// ];
|
|
/// let uuid = Uuid::from_bytes(bytes);
|
|
/// assert_eq!(bytes, uuid.into_bytes());
|
|
/// ```
|
|
#[inline]
|
|
pub const fn into_bytes(self) -> Bytes {
|
|
self.0
|
|
}
|
|
|
|
/// Returns the bytes of the UUID in little-endian order.
|
|
///
|
|
/// The bytes will be flipped to convert into little-endian order. This is
|
|
/// based on the endianness of the UUID, rather than the target environment
|
|
/// so bytes will be flipped on both big and little endian machines.
|
|
///
|
|
/// # Examples
|
|
///
|
|
/// ```
|
|
/// use uuid::Uuid;
|
|
///
|
|
/// # fn main() -> Result<(), uuid::Error> {
|
|
/// let uuid = Uuid::parse_str("a1a2a3a4-b1b2-c1c2-d1d2-d3d4d5d6d7d8")?;
|
|
///
|
|
/// assert_eq!(
|
|
/// uuid.to_bytes_le(),
|
|
/// ([
|
|
/// 0xa4, 0xa3, 0xa2, 0xa1, 0xb2, 0xb1, 0xc2, 0xc1, 0xd1, 0xd2,
|
|
/// 0xd3, 0xd4, 0xd5, 0xd6, 0xd7, 0xd8
|
|
/// ])
|
|
/// );
|
|
/// # Ok(())
|
|
/// # }
|
|
/// ```
|
|
pub const fn to_bytes_le(&self) -> Bytes {
|
|
[
|
|
self.0[3], self.0[2], self.0[1], self.0[0], self.0[5], self.0[4], self.0[7], self.0[6],
|
|
self.0[8], self.0[9], self.0[10], self.0[11], self.0[12], self.0[13], self.0[14],
|
|
self.0[15],
|
|
]
|
|
}
|
|
|
|
/// Tests if the UUID is nil (all zeros).
|
|
pub const fn is_nil(&self) -> bool {
|
|
self.as_u128() == u128::MIN
|
|
}
|
|
|
|
/// Tests if the UUID is max (all ones).
|
|
pub const fn is_max(&self) -> bool {
|
|
self.as_u128() == u128::MAX
|
|
}
|
|
|
|
/// A buffer that can be used for `encode_...` calls, that is
|
|
/// guaranteed to be long enough for any of the format adapters.
|
|
///
|
|
/// # Examples
|
|
///
|
|
/// ```
|
|
/// # use uuid::Uuid;
|
|
/// let uuid = Uuid::nil();
|
|
///
|
|
/// assert_eq!(
|
|
/// uuid.simple().encode_lower(&mut Uuid::encode_buffer()),
|
|
/// "00000000000000000000000000000000"
|
|
/// );
|
|
///
|
|
/// assert_eq!(
|
|
/// uuid.hyphenated()
|
|
/// .encode_lower(&mut Uuid::encode_buffer()),
|
|
/// "00000000-0000-0000-0000-000000000000"
|
|
/// );
|
|
///
|
|
/// assert_eq!(
|
|
/// uuid.urn().encode_lower(&mut Uuid::encode_buffer()),
|
|
/// "urn:uuid:00000000-0000-0000-0000-000000000000"
|
|
/// );
|
|
/// ```
|
|
pub const fn encode_buffer() -> [u8; fmt::Urn::LENGTH] {
|
|
[0; fmt::Urn::LENGTH]
|
|
}
|
|
|
|
/// If the UUID is the correct version (v1, v6, or v7) this will return
|
|
/// the timestamp and counter portion parsed from a V1 UUID.
|
|
///
|
|
/// Returns `None` if the supplied UUID is not V1.
|
|
///
|
|
/// The V1 timestamp format defined in RFC4122 specifies a 60-bit
|
|
/// integer representing the number of 100-nanosecond intervals
|
|
/// since 00:00:00.00, 15 Oct 1582.
|
|
///
|
|
/// [`Timestamp`] offers several options for converting the raw RFC4122
|
|
/// value into more commonly-used formats, such as a unix timestamp.
|
|
///
|
|
/// # Roundtripping
|
|
///
|
|
/// This method is unlikely to roundtrip a timestamp in a UUID due to the way
|
|
/// UUIDs encode timestamps. The timestamp returned from this method will be truncated to
|
|
/// 100ns precision for version 1 and 6 UUIDs, and to millisecond precision for version 7 UUIDs.
|
|
///
|
|
/// [`Timestamp`]: v1/struct.Timestamp.html
|
|
pub const fn get_timestamp(&self) -> Option<Timestamp> {
|
|
match self.get_version() {
|
|
Some(Version::Mac) => {
|
|
let (ticks, counter) = timestamp::decode_rfc4122_timestamp(self);
|
|
|
|
Some(Timestamp::from_rfc4122(ticks, counter))
|
|
}
|
|
Some(Version::SortMac) => {
|
|
let (ticks, counter) = timestamp::decode_sorted_rfc4122_timestamp(self);
|
|
|
|
Some(Timestamp::from_rfc4122(ticks, counter))
|
|
}
|
|
Some(Version::SortRand) => {
|
|
let millis = timestamp::decode_unix_timestamp_millis(self);
|
|
|
|
let seconds = millis / 1000;
|
|
let nanos = ((millis % 1000) * 1_000_000) as u32;
|
|
|
|
Some(Timestamp {
|
|
seconds,
|
|
nanos,
|
|
#[cfg(any(feature = "v1", feature = "v6"))]
|
|
counter: 0,
|
|
})
|
|
}
|
|
_ => None,
|
|
}
|
|
}
|
|
}
|
|
|
|
impl Default for Uuid {
|
|
#[inline]
|
|
fn default() -> Self {
|
|
Uuid::nil()
|
|
}
|
|
}
|
|
|
|
impl AsRef<[u8]> for Uuid {
|
|
#[inline]
|
|
fn as_ref(&self) -> &[u8] {
|
|
&self.0
|
|
}
|
|
}
|
|
|
|
#[cfg(feature = "std")]
|
|
impl From<Uuid> for std::vec::Vec<u8> {
|
|
fn from(value: Uuid) -> Self {
|
|
value.0.to_vec()
|
|
}
|
|
}
|
|
|
|
#[cfg(feature = "std")]
|
|
impl std::convert::TryFrom<std::vec::Vec<u8>> for Uuid {
|
|
type Error = Error;
|
|
|
|
fn try_from(value: std::vec::Vec<u8>) -> Result<Self, Self::Error> {
|
|
Uuid::from_slice(&value)
|
|
}
|
|
}
|
|
|
|
#[cfg(feature = "serde")]
|
|
pub mod serde {
|
|
//! Adapters for alternative `serde` formats.
|
|
//!
|
|
//! This module contains adapters you can use with [`#[serde(with)]`](https://serde.rs/field-attrs.html#with)
|
|
//! to change the way a [`Uuid`](../struct.Uuid.html) is serialized
|
|
//! and deserialized.
|
|
|
|
pub use crate::external::serde_support::compact;
|
|
}
|
|
|
|
#[cfg(test)]
|
|
mod tests {
|
|
use super::*;
|
|
|
|
use crate::std::string::{String, ToString};
|
|
|
|
#[cfg(all(
|
|
target_arch = "wasm32",
|
|
target_vendor = "unknown",
|
|
target_os = "unknown"
|
|
))]
|
|
use wasm_bindgen_test::*;
|
|
|
|
macro_rules! check {
|
|
($buf:ident, $format:expr, $target:expr, $len:expr, $cond:expr) => {
|
|
$buf.clear();
|
|
write!($buf, $format, $target).unwrap();
|
|
assert!($buf.len() == $len);
|
|
assert!($buf.chars().all($cond), "{}", $buf);
|
|
};
|
|
}
|
|
|
|
pub const fn new() -> Uuid {
|
|
Uuid::from_bytes([
|
|
0xF9, 0x16, 0x8C, 0x5E, 0xCE, 0xB2, 0x4F, 0xAA, 0xB6, 0xBF, 0x32, 0x9B, 0xF3, 0x9F,
|
|
0xA1, 0xE4,
|
|
])
|
|
}
|
|
|
|
pub const fn new2() -> Uuid {
|
|
Uuid::from_bytes([
|
|
0xF9, 0x16, 0x8C, 0x5E, 0xCE, 0xB2, 0x4F, 0xAB, 0xB6, 0xBF, 0x32, 0x9B, 0xF3, 0x9F,
|
|
0xA1, 0xE4,
|
|
])
|
|
}
|
|
|
|
#[test]
|
|
#[cfg_attr(
|
|
all(
|
|
target_arch = "wasm32",
|
|
target_vendor = "unknown",
|
|
target_os = "unknown"
|
|
),
|
|
wasm_bindgen_test
|
|
)]
|
|
fn test_uuid_compare() {
|
|
let uuid1 = new();
|
|
let uuid2 = new2();
|
|
|
|
assert_eq!(uuid1, uuid1);
|
|
assert_eq!(uuid2, uuid2);
|
|
|
|
assert_ne!(uuid1, uuid2);
|
|
assert_ne!(uuid2, uuid1);
|
|
}
|
|
|
|
#[test]
|
|
#[cfg_attr(
|
|
all(
|
|
target_arch = "wasm32",
|
|
target_vendor = "unknown",
|
|
target_os = "unknown"
|
|
),
|
|
wasm_bindgen_test
|
|
)]
|
|
fn test_uuid_default() {
|
|
let default_uuid = Uuid::default();
|
|
let nil_uuid = Uuid::nil();
|
|
|
|
assert_eq!(default_uuid, nil_uuid);
|
|
}
|
|
|
|
#[test]
|
|
#[cfg_attr(
|
|
all(
|
|
target_arch = "wasm32",
|
|
target_vendor = "unknown",
|
|
target_os = "unknown"
|
|
),
|
|
wasm_bindgen_test
|
|
)]
|
|
fn test_uuid_display() {
|
|
use crate::std::fmt::Write;
|
|
|
|
let uuid = new();
|
|
let s = uuid.to_string();
|
|
let mut buffer = String::new();
|
|
|
|
assert_eq!(s, uuid.hyphenated().to_string());
|
|
|
|
check!(buffer, "{}", uuid, 36, |c| c.is_lowercase()
|
|
|| c.is_digit(10)
|
|
|| c == '-');
|
|
}
|
|
|
|
#[test]
|
|
#[cfg_attr(
|
|
all(
|
|
target_arch = "wasm32",
|
|
target_vendor = "unknown",
|
|
target_os = "unknown"
|
|
),
|
|
wasm_bindgen_test
|
|
)]
|
|
fn test_uuid_lowerhex() {
|
|
use crate::std::fmt::Write;
|
|
|
|
let mut buffer = String::new();
|
|
let uuid = new();
|
|
|
|
check!(buffer, "{:x}", uuid, 36, |c| c.is_lowercase()
|
|
|| c.is_digit(10)
|
|
|| c == '-');
|
|
}
|
|
|
|
// noinspection RsAssertEqual
|
|
#[test]
|
|
#[cfg_attr(
|
|
all(
|
|
target_arch = "wasm32",
|
|
target_vendor = "unknown",
|
|
target_os = "unknown"
|
|
),
|
|
wasm_bindgen_test
|
|
)]
|
|
fn test_uuid_operator_eq() {
|
|
let uuid1 = new();
|
|
let uuid1_dup = uuid1.clone();
|
|
let uuid2 = new2();
|
|
|
|
assert!(uuid1 == uuid1);
|
|
assert!(uuid1 == uuid1_dup);
|
|
assert!(uuid1_dup == uuid1);
|
|
|
|
assert!(uuid1 != uuid2);
|
|
assert!(uuid2 != uuid1);
|
|
assert!(uuid1_dup != uuid2);
|
|
assert!(uuid2 != uuid1_dup);
|
|
}
|
|
|
|
#[test]
|
|
#[cfg_attr(
|
|
all(
|
|
target_arch = "wasm32",
|
|
target_vendor = "unknown",
|
|
target_os = "unknown"
|
|
),
|
|
wasm_bindgen_test
|
|
)]
|
|
fn test_uuid_to_string() {
|
|
use crate::std::fmt::Write;
|
|
|
|
let uuid = new();
|
|
let s = uuid.to_string();
|
|
let mut buffer = String::new();
|
|
|
|
assert_eq!(s.len(), 36);
|
|
|
|
check!(buffer, "{}", s, 36, |c| c.is_lowercase()
|
|
|| c.is_digit(10)
|
|
|| c == '-');
|
|
}
|
|
|
|
#[test]
|
|
#[cfg_attr(
|
|
all(
|
|
target_arch = "wasm32",
|
|
target_vendor = "unknown",
|
|
target_os = "unknown"
|
|
),
|
|
wasm_bindgen_test
|
|
)]
|
|
fn test_non_conforming() {
|
|
let from_bytes =
|
|
Uuid::from_bytes([4, 54, 67, 12, 43, 2, 2, 76, 32, 50, 87, 5, 1, 33, 43, 87]);
|
|
|
|
assert_eq!(from_bytes.get_version(), None);
|
|
}
|
|
|
|
#[test]
|
|
#[cfg_attr(
|
|
all(
|
|
target_arch = "wasm32",
|
|
target_vendor = "unknown",
|
|
target_os = "unknown"
|
|
),
|
|
wasm_bindgen_test
|
|
)]
|
|
fn test_nil() {
|
|
let nil = Uuid::nil();
|
|
let not_nil = new();
|
|
|
|
assert!(nil.is_nil());
|
|
assert!(!not_nil.is_nil());
|
|
|
|
assert_eq!(nil.get_version(), Some(Version::Nil));
|
|
assert_eq!(not_nil.get_version(), Some(Version::Random));
|
|
|
|
assert_eq!(
|
|
nil,
|
|
Builder::from_bytes([0; 16])
|
|
.with_version(Version::Nil)
|
|
.into_uuid()
|
|
);
|
|
}
|
|
|
|
#[test]
|
|
#[cfg_attr(
|
|
all(
|
|
target_arch = "wasm32",
|
|
target_vendor = "unknown",
|
|
target_os = "unknown"
|
|
),
|
|
wasm_bindgen_test
|
|
)]
|
|
fn test_max() {
|
|
let max = Uuid::max();
|
|
let not_max = new();
|
|
|
|
assert!(max.is_max());
|
|
assert!(!not_max.is_max());
|
|
|
|
assert_eq!(max.get_version(), Some(Version::Max));
|
|
assert_eq!(not_max.get_version(), Some(Version::Random));
|
|
|
|
assert_eq!(
|
|
max,
|
|
Builder::from_bytes([0xff; 16])
|
|
.with_version(Version::Max)
|
|
.into_uuid()
|
|
);
|
|
}
|
|
|
|
#[test]
|
|
#[cfg_attr(
|
|
all(
|
|
target_arch = "wasm32",
|
|
target_vendor = "unknown",
|
|
target_os = "unknown"
|
|
),
|
|
wasm_bindgen_test
|
|
)]
|
|
fn test_predefined_namespaces() {
|
|
assert_eq!(
|
|
Uuid::NAMESPACE_DNS.hyphenated().to_string(),
|
|
"6ba7b810-9dad-11d1-80b4-00c04fd430c8"
|
|
);
|
|
assert_eq!(
|
|
Uuid::NAMESPACE_URL.hyphenated().to_string(),
|
|
"6ba7b811-9dad-11d1-80b4-00c04fd430c8"
|
|
);
|
|
assert_eq!(
|
|
Uuid::NAMESPACE_OID.hyphenated().to_string(),
|
|
"6ba7b812-9dad-11d1-80b4-00c04fd430c8"
|
|
);
|
|
assert_eq!(
|
|
Uuid::NAMESPACE_X500.hyphenated().to_string(),
|
|
"6ba7b814-9dad-11d1-80b4-00c04fd430c8"
|
|
);
|
|
}
|
|
|
|
#[cfg(feature = "v3")]
|
|
#[test]
|
|
#[cfg_attr(
|
|
all(
|
|
target_arch = "wasm32",
|
|
target_vendor = "unknown",
|
|
target_os = "unknown"
|
|
),
|
|
wasm_bindgen_test
|
|
)]
|
|
fn test_get_version_v3() {
|
|
let uuid = Uuid::new_v3(&Uuid::NAMESPACE_DNS, "rust-lang.org".as_bytes());
|
|
|
|
assert_eq!(uuid.get_version().unwrap(), Version::Md5);
|
|
assert_eq!(uuid.get_version_num(), 3);
|
|
}
|
|
|
|
#[test]
|
|
#[cfg_attr(
|
|
all(
|
|
target_arch = "wasm32",
|
|
target_vendor = "unknown",
|
|
target_os = "unknown"
|
|
),
|
|
wasm_bindgen_test
|
|
)]
|
|
fn test_get_variant() {
|
|
let uuid1 = new();
|
|
let uuid2 = Uuid::parse_str("550e8400-e29b-41d4-a716-446655440000").unwrap();
|
|
let uuid3 = Uuid::parse_str("67e55044-10b1-426f-9247-bb680e5fe0c8").unwrap();
|
|
let uuid4 = Uuid::parse_str("936DA01F9ABD4d9dC0C702AF85C822A8").unwrap();
|
|
let uuid5 = Uuid::parse_str("F9168C5E-CEB2-4faa-D6BF-329BF39FA1E4").unwrap();
|
|
let uuid6 = Uuid::parse_str("f81d4fae-7dec-11d0-7765-00a0c91e6bf6").unwrap();
|
|
|
|
assert_eq!(uuid1.get_variant(), Variant::RFC4122);
|
|
assert_eq!(uuid2.get_variant(), Variant::RFC4122);
|
|
assert_eq!(uuid3.get_variant(), Variant::RFC4122);
|
|
assert_eq!(uuid4.get_variant(), Variant::Microsoft);
|
|
assert_eq!(uuid5.get_variant(), Variant::Microsoft);
|
|
assert_eq!(uuid6.get_variant(), Variant::NCS);
|
|
}
|
|
|
|
#[test]
|
|
#[cfg_attr(
|
|
all(
|
|
target_arch = "wasm32",
|
|
target_vendor = "unknown",
|
|
target_os = "unknown"
|
|
),
|
|
wasm_bindgen_test
|
|
)]
|
|
fn test_to_simple_string() {
|
|
let uuid1 = new();
|
|
let s = uuid1.simple().to_string();
|
|
|
|
assert_eq!(s.len(), 32);
|
|
assert!(s.chars().all(|c| c.is_digit(16)));
|
|
}
|
|
|
|
#[test]
|
|
#[cfg_attr(
|
|
all(
|
|
target_arch = "wasm32",
|
|
target_vendor = "unknown",
|
|
target_os = "unknown"
|
|
),
|
|
wasm_bindgen_test
|
|
)]
|
|
fn test_hyphenated_string() {
|
|
let uuid1 = new();
|
|
let s = uuid1.hyphenated().to_string();
|
|
|
|
assert_eq!(36, s.len());
|
|
assert!(s.chars().all(|c| c.is_digit(16) || c == '-'));
|
|
}
|
|
|
|
#[test]
|
|
#[cfg_attr(
|
|
all(
|
|
target_arch = "wasm32",
|
|
target_vendor = "unknown",
|
|
target_os = "unknown"
|
|
),
|
|
wasm_bindgen_test
|
|
)]
|
|
fn test_upper_lower_hex() {
|
|
use std::fmt::Write;
|
|
|
|
let mut buf = String::new();
|
|
let u = new();
|
|
|
|
macro_rules! check {
|
|
($buf:ident, $format:expr, $target:expr, $len:expr, $cond:expr) => {
|
|
$buf.clear();
|
|
write!($buf, $format, $target).unwrap();
|
|
assert_eq!($len, buf.len());
|
|
assert!($buf.chars().all($cond), "{}", $buf);
|
|
};
|
|
}
|
|
|
|
check!(buf, "{:x}", u, 36, |c| c.is_lowercase()
|
|
|| c.is_digit(10)
|
|
|| c == '-');
|
|
check!(buf, "{:X}", u, 36, |c| c.is_uppercase()
|
|
|| c.is_digit(10)
|
|
|| c == '-');
|
|
check!(buf, "{:#x}", u, 36, |c| c.is_lowercase()
|
|
|| c.is_digit(10)
|
|
|| c == '-');
|
|
check!(buf, "{:#X}", u, 36, |c| c.is_uppercase()
|
|
|| c.is_digit(10)
|
|
|| c == '-');
|
|
|
|
check!(buf, "{:X}", u.hyphenated(), 36, |c| c.is_uppercase()
|
|
|| c.is_digit(10)
|
|
|| c == '-');
|
|
check!(buf, "{:X}", u.simple(), 32, |c| c.is_uppercase()
|
|
|| c.is_digit(10));
|
|
check!(buf, "{:#X}", u.hyphenated(), 36, |c| c.is_uppercase()
|
|
|| c.is_digit(10)
|
|
|| c == '-');
|
|
check!(buf, "{:#X}", u.simple(), 32, |c| c.is_uppercase()
|
|
|| c.is_digit(10));
|
|
|
|
check!(buf, "{:x}", u.hyphenated(), 36, |c| c.is_lowercase()
|
|
|| c.is_digit(10)
|
|
|| c == '-');
|
|
check!(buf, "{:x}", u.simple(), 32, |c| c.is_lowercase()
|
|
|| c.is_digit(10));
|
|
check!(buf, "{:#x}", u.hyphenated(), 36, |c| c.is_lowercase()
|
|
|| c.is_digit(10)
|
|
|| c == '-');
|
|
check!(buf, "{:#x}", u.simple(), 32, |c| c.is_lowercase()
|
|
|| c.is_digit(10));
|
|
}
|
|
|
|
#[test]
|
|
#[cfg_attr(
|
|
all(
|
|
target_arch = "wasm32",
|
|
target_vendor = "unknown",
|
|
target_os = "unknown"
|
|
),
|
|
wasm_bindgen_test
|
|
)]
|
|
fn test_to_urn_string() {
|
|
let uuid1 = new();
|
|
let ss = uuid1.urn().to_string();
|
|
let s = &ss[9..];
|
|
|
|
assert!(ss.starts_with("urn:uuid:"));
|
|
assert_eq!(s.len(), 36);
|
|
assert!(s.chars().all(|c| c.is_digit(16) || c == '-'));
|
|
}
|
|
|
|
#[test]
|
|
#[cfg_attr(
|
|
all(
|
|
target_arch = "wasm32",
|
|
target_vendor = "unknown",
|
|
target_os = "unknown"
|
|
),
|
|
wasm_bindgen_test
|
|
)]
|
|
fn test_to_simple_string_matching() {
|
|
let uuid1 = new();
|
|
|
|
let hs = uuid1.hyphenated().to_string();
|
|
let ss = uuid1.simple().to_string();
|
|
|
|
let hsn = hs.chars().filter(|&c| c != '-').collect::<String>();
|
|
|
|
assert_eq!(hsn, ss);
|
|
}
|
|
|
|
#[test]
|
|
#[cfg_attr(
|
|
all(
|
|
target_arch = "wasm32",
|
|
target_vendor = "unknown",
|
|
target_os = "unknown"
|
|
),
|
|
wasm_bindgen_test
|
|
)]
|
|
fn test_string_roundtrip() {
|
|
let uuid = new();
|
|
|
|
let hs = uuid.hyphenated().to_string();
|
|
let uuid_hs = Uuid::parse_str(&hs).unwrap();
|
|
assert_eq!(uuid_hs, uuid);
|
|
|
|
let ss = uuid.to_string();
|
|
let uuid_ss = Uuid::parse_str(&ss).unwrap();
|
|
assert_eq!(uuid_ss, uuid);
|
|
}
|
|
|
|
#[test]
|
|
#[cfg_attr(
|
|
all(
|
|
target_arch = "wasm32",
|
|
target_vendor = "unknown",
|
|
target_os = "unknown"
|
|
),
|
|
wasm_bindgen_test
|
|
)]
|
|
fn test_from_fields() {
|
|
let d1: u32 = 0xa1a2a3a4;
|
|
let d2: u16 = 0xb1b2;
|
|
let d3: u16 = 0xc1c2;
|
|
let d4 = [0xd1, 0xd2, 0xd3, 0xd4, 0xd5, 0xd6, 0xd7, 0xd8];
|
|
|
|
let u = Uuid::from_fields(d1, d2, d3, &d4);
|
|
|
|
let expected = "a1a2a3a4b1b2c1c2d1d2d3d4d5d6d7d8";
|
|
let result = u.simple().to_string();
|
|
assert_eq!(result, expected);
|
|
}
|
|
|
|
#[test]
|
|
#[cfg_attr(
|
|
all(
|
|
target_arch = "wasm32",
|
|
target_vendor = "unknown",
|
|
target_os = "unknown"
|
|
),
|
|
wasm_bindgen_test
|
|
)]
|
|
fn test_from_fields_le() {
|
|
let d1: u32 = 0xa4a3a2a1;
|
|
let d2: u16 = 0xb2b1;
|
|
let d3: u16 = 0xc2c1;
|
|
let d4 = [0xd1, 0xd2, 0xd3, 0xd4, 0xd5, 0xd6, 0xd7, 0xd8];
|
|
|
|
let u = Uuid::from_fields_le(d1, d2, d3, &d4);
|
|
|
|
let expected = "a1a2a3a4b1b2c1c2d1d2d3d4d5d6d7d8";
|
|
let result = u.simple().to_string();
|
|
assert_eq!(result, expected);
|
|
}
|
|
|
|
#[test]
|
|
#[cfg_attr(
|
|
all(
|
|
target_arch = "wasm32",
|
|
target_vendor = "unknown",
|
|
target_os = "unknown"
|
|
),
|
|
wasm_bindgen_test
|
|
)]
|
|
fn test_as_fields() {
|
|
let u = new();
|
|
let (d1, d2, d3, d4) = u.as_fields();
|
|
|
|
assert_ne!(d1, 0);
|
|
assert_ne!(d2, 0);
|
|
assert_ne!(d3, 0);
|
|
assert_eq!(d4.len(), 8);
|
|
assert!(!d4.iter().all(|&b| b == 0));
|
|
}
|
|
|
|
#[test]
|
|
#[cfg_attr(
|
|
all(
|
|
target_arch = "wasm32",
|
|
target_vendor = "unknown",
|
|
target_os = "unknown"
|
|
),
|
|
wasm_bindgen_test
|
|
)]
|
|
fn test_fields_roundtrip() {
|
|
let d1_in: u32 = 0xa1a2a3a4;
|
|
let d2_in: u16 = 0xb1b2;
|
|
let d3_in: u16 = 0xc1c2;
|
|
let d4_in = &[0xd1, 0xd2, 0xd3, 0xd4, 0xd5, 0xd6, 0xd7, 0xd8];
|
|
|
|
let u = Uuid::from_fields(d1_in, d2_in, d3_in, d4_in);
|
|
let (d1_out, d2_out, d3_out, d4_out) = u.as_fields();
|
|
|
|
assert_eq!(d1_in, d1_out);
|
|
assert_eq!(d2_in, d2_out);
|
|
assert_eq!(d3_in, d3_out);
|
|
assert_eq!(d4_in, d4_out);
|
|
}
|
|
|
|
#[test]
|
|
#[cfg_attr(
|
|
all(
|
|
target_arch = "wasm32",
|
|
target_vendor = "unknown",
|
|
target_os = "unknown"
|
|
),
|
|
wasm_bindgen_test
|
|
)]
|
|
fn test_fields_le_roundtrip() {
|
|
let d1_in: u32 = 0xa4a3a2a1;
|
|
let d2_in: u16 = 0xb2b1;
|
|
let d3_in: u16 = 0xc2c1;
|
|
let d4_in = &[0xd1, 0xd2, 0xd3, 0xd4, 0xd5, 0xd6, 0xd7, 0xd8];
|
|
|
|
let u = Uuid::from_fields_le(d1_in, d2_in, d3_in, d4_in);
|
|
let (d1_out, d2_out, d3_out, d4_out) = u.to_fields_le();
|
|
|
|
assert_eq!(d1_in, d1_out);
|
|
assert_eq!(d2_in, d2_out);
|
|
assert_eq!(d3_in, d3_out);
|
|
assert_eq!(d4_in, d4_out);
|
|
}
|
|
|
|
#[test]
|
|
#[cfg_attr(
|
|
all(
|
|
target_arch = "wasm32",
|
|
target_vendor = "unknown",
|
|
target_os = "unknown"
|
|
),
|
|
wasm_bindgen_test
|
|
)]
|
|
fn test_fields_le_are_actually_le() {
|
|
let d1_in: u32 = 0xa1a2a3a4;
|
|
let d2_in: u16 = 0xb1b2;
|
|
let d3_in: u16 = 0xc1c2;
|
|
let d4_in = &[0xd1, 0xd2, 0xd3, 0xd4, 0xd5, 0xd6, 0xd7, 0xd8];
|
|
|
|
let u = Uuid::from_fields(d1_in, d2_in, d3_in, d4_in);
|
|
let (d1_out, d2_out, d3_out, d4_out) = u.to_fields_le();
|
|
|
|
assert_eq!(d1_in, d1_out.swap_bytes());
|
|
assert_eq!(d2_in, d2_out.swap_bytes());
|
|
assert_eq!(d3_in, d3_out.swap_bytes());
|
|
assert_eq!(d4_in, d4_out);
|
|
}
|
|
|
|
#[test]
|
|
#[cfg_attr(
|
|
all(
|
|
target_arch = "wasm32",
|
|
target_vendor = "unknown",
|
|
target_os = "unknown"
|
|
),
|
|
wasm_bindgen_test
|
|
)]
|
|
fn test_from_u128() {
|
|
let v_in: u128 = 0xa1a2a3a4b1b2c1c2d1d2d3d4d5d6d7d8;
|
|
|
|
let u = Uuid::from_u128(v_in);
|
|
|
|
let expected = "a1a2a3a4b1b2c1c2d1d2d3d4d5d6d7d8";
|
|
let result = u.simple().to_string();
|
|
assert_eq!(result, expected);
|
|
}
|
|
|
|
#[test]
|
|
#[cfg_attr(
|
|
all(
|
|
target_arch = "wasm32",
|
|
target_vendor = "unknown",
|
|
target_os = "unknown"
|
|
),
|
|
wasm_bindgen_test
|
|
)]
|
|
fn test_from_u128_le() {
|
|
let v_in: u128 = 0xd8d7d6d5d4d3d2d1c2c1b2b1a4a3a2a1;
|
|
|
|
let u = Uuid::from_u128_le(v_in);
|
|
|
|
let expected = "a1a2a3a4b1b2c1c2d1d2d3d4d5d6d7d8";
|
|
let result = u.simple().to_string();
|
|
assert_eq!(result, expected);
|
|
}
|
|
|
|
#[test]
|
|
#[cfg_attr(
|
|
all(
|
|
target_arch = "wasm32",
|
|
target_vendor = "unknown",
|
|
target_os = "unknown"
|
|
),
|
|
wasm_bindgen_test
|
|
)]
|
|
fn test_from_u64_pair() {
|
|
let high_in: u64 = 0xa1a2a3a4b1b2c1c2;
|
|
let low_in: u64 = 0xd1d2d3d4d5d6d7d8;
|
|
|
|
let u = Uuid::from_u64_pair(high_in, low_in);
|
|
|
|
let expected = "a1a2a3a4b1b2c1c2d1d2d3d4d5d6d7d8";
|
|
let result = u.simple().to_string();
|
|
assert_eq!(result, expected);
|
|
}
|
|
|
|
#[test]
|
|
#[cfg_attr(
|
|
all(
|
|
target_arch = "wasm32",
|
|
target_vendor = "unknown",
|
|
target_os = "unknown"
|
|
),
|
|
wasm_bindgen_test
|
|
)]
|
|
fn test_u128_roundtrip() {
|
|
let v_in: u128 = 0xa1a2a3a4b1b2c1c2d1d2d3d4d5d6d7d8;
|
|
|
|
let u = Uuid::from_u128(v_in);
|
|
let v_out = u.as_u128();
|
|
|
|
assert_eq!(v_in, v_out);
|
|
}
|
|
|
|
#[test]
|
|
#[cfg_attr(
|
|
all(
|
|
target_arch = "wasm32",
|
|
target_vendor = "unknown",
|
|
target_os = "unknown"
|
|
),
|
|
wasm_bindgen_test
|
|
)]
|
|
fn test_u128_le_roundtrip() {
|
|
let v_in: u128 = 0xd8d7d6d5d4d3d2d1c2c1b2b1a4a3a2a1;
|
|
|
|
let u = Uuid::from_u128_le(v_in);
|
|
let v_out = u.to_u128_le();
|
|
|
|
assert_eq!(v_in, v_out);
|
|
}
|
|
|
|
#[test]
|
|
#[cfg_attr(
|
|
all(
|
|
target_arch = "wasm32",
|
|
target_vendor = "unknown",
|
|
target_os = "unknown"
|
|
),
|
|
wasm_bindgen_test
|
|
)]
|
|
fn test_u64_pair_roundtrip() {
|
|
let high_in: u64 = 0xa1a2a3a4b1b2c1c2;
|
|
let low_in: u64 = 0xd1d2d3d4d5d6d7d8;
|
|
|
|
let u = Uuid::from_u64_pair(high_in, low_in);
|
|
let (high_out, low_out) = u.as_u64_pair();
|
|
|
|
assert_eq!(high_in, high_out);
|
|
assert_eq!(low_in, low_out);
|
|
}
|
|
|
|
#[test]
|
|
#[cfg_attr(
|
|
all(
|
|
target_arch = "wasm32",
|
|
target_vendor = "unknown",
|
|
target_os = "unknown"
|
|
),
|
|
wasm_bindgen_test
|
|
)]
|
|
fn test_u128_le_is_actually_le() {
|
|
let v_in: u128 = 0xa1a2a3a4b1b2c1c2d1d2d3d4d5d6d7d8;
|
|
|
|
let u = Uuid::from_u128(v_in);
|
|
let v_out = u.to_u128_le();
|
|
|
|
assert_eq!(v_in, v_out.swap_bytes());
|
|
}
|
|
|
|
#[test]
|
|
#[cfg_attr(
|
|
all(
|
|
target_arch = "wasm32",
|
|
target_vendor = "unknown",
|
|
target_os = "unknown"
|
|
),
|
|
wasm_bindgen_test
|
|
)]
|
|
fn test_from_slice() {
|
|
let b = [
|
|
0xa1, 0xa2, 0xa3, 0xa4, 0xb1, 0xb2, 0xc1, 0xc2, 0xd1, 0xd2, 0xd3, 0xd4, 0xd5, 0xd6,
|
|
0xd7, 0xd8,
|
|
];
|
|
|
|
let u = Uuid::from_slice(&b).unwrap();
|
|
let expected = "a1a2a3a4b1b2c1c2d1d2d3d4d5d6d7d8";
|
|
|
|
assert_eq!(u.simple().to_string(), expected);
|
|
}
|
|
|
|
#[test]
|
|
#[cfg_attr(
|
|
all(
|
|
target_arch = "wasm32",
|
|
target_vendor = "unknown",
|
|
target_os = "unknown"
|
|
),
|
|
wasm_bindgen_test
|
|
)]
|
|
fn test_from_bytes() {
|
|
let b = [
|
|
0xa1, 0xa2, 0xa3, 0xa4, 0xb1, 0xb2, 0xc1, 0xc2, 0xd1, 0xd2, 0xd3, 0xd4, 0xd5, 0xd6,
|
|
0xd7, 0xd8,
|
|
];
|
|
|
|
let u = Uuid::from_bytes(b);
|
|
let expected = "a1a2a3a4b1b2c1c2d1d2d3d4d5d6d7d8";
|
|
|
|
assert_eq!(u.simple().to_string(), expected);
|
|
}
|
|
|
|
#[test]
|
|
#[cfg_attr(
|
|
all(
|
|
target_arch = "wasm32",
|
|
target_vendor = "unknown",
|
|
target_os = "unknown"
|
|
),
|
|
wasm_bindgen_test
|
|
)]
|
|
fn test_as_bytes() {
|
|
let u = new();
|
|
let ub = u.as_bytes();
|
|
let ur = u.as_ref();
|
|
|
|
assert_eq!(ub.len(), 16);
|
|
assert_eq!(ur.len(), 16);
|
|
assert!(!ub.iter().all(|&b| b == 0));
|
|
assert!(!ur.iter().all(|&b| b == 0));
|
|
}
|
|
|
|
#[test]
|
|
#[cfg(feature = "std")]
|
|
#[cfg_attr(
|
|
all(
|
|
target_arch = "wasm32",
|
|
target_vendor = "unknown",
|
|
target_os = "unknown"
|
|
),
|
|
wasm_bindgen_test
|
|
)]
|
|
fn test_convert_vec() {
|
|
use crate::std::{convert::TryInto, vec::Vec};
|
|
|
|
let u = new();
|
|
let ub = u.as_ref();
|
|
|
|
let v: Vec<u8> = u.into();
|
|
|
|
assert_eq!(&v, ub);
|
|
|
|
let uv: Uuid = v.try_into().unwrap();
|
|
|
|
assert_eq!(uv, u);
|
|
}
|
|
|
|
#[test]
|
|
#[cfg_attr(
|
|
all(
|
|
target_arch = "wasm32",
|
|
target_vendor = "unknown",
|
|
target_os = "unknown"
|
|
),
|
|
wasm_bindgen_test
|
|
)]
|
|
fn test_bytes_roundtrip() {
|
|
let b_in: crate::Bytes = [
|
|
0xa1, 0xa2, 0xa3, 0xa4, 0xb1, 0xb2, 0xc1, 0xc2, 0xd1, 0xd2, 0xd3, 0xd4, 0xd5, 0xd6,
|
|
0xd7, 0xd8,
|
|
];
|
|
|
|
let u = Uuid::from_slice(&b_in).unwrap();
|
|
|
|
let b_out = u.as_bytes();
|
|
|
|
assert_eq!(&b_in, b_out);
|
|
}
|
|
|
|
#[test]
|
|
#[cfg_attr(
|
|
all(
|
|
target_arch = "wasm32",
|
|
target_vendor = "unknown",
|
|
target_os = "unknown"
|
|
),
|
|
wasm_bindgen_test
|
|
)]
|
|
fn test_bytes_le_roundtrip() {
|
|
let b = [
|
|
0xa1, 0xa2, 0xa3, 0xa4, 0xb1, 0xb2, 0xc1, 0xc2, 0xd1, 0xd2, 0xd3, 0xd4, 0xd5, 0xd6,
|
|
0xd7, 0xd8,
|
|
];
|
|
|
|
let u1 = Uuid::from_bytes(b);
|
|
|
|
let b_le = u1.to_bytes_le();
|
|
|
|
let u2 = Uuid::from_bytes_le(b_le);
|
|
|
|
assert_eq!(u1, u2);
|
|
}
|
|
|
|
#[test]
|
|
#[cfg_attr(
|
|
all(
|
|
target_arch = "wasm32",
|
|
target_vendor = "unknown",
|
|
target_os = "unknown"
|
|
),
|
|
wasm_bindgen_test
|
|
)]
|
|
fn test_iterbytes_impl_for_uuid() {
|
|
let mut set = std::collections::HashSet::new();
|
|
let id1 = new();
|
|
let id2 = new2();
|
|
set.insert(id1.clone());
|
|
|
|
assert!(set.contains(&id1));
|
|
assert!(!set.contains(&id2));
|
|
}
|
|
}
|