Daft is a library to perform semantic diffs of Rust data structures.
Daft consists of a trait called Diffable
, along with [a derive
macro][macro@Diffable] by the same name. This trait represents the
notion of a type for which two members can be simultaneously compared.
- Recursive diffing of structs, sets, and maps
- Derive macro for automatically generating diff types
- Choose between eager and lazy diffing
- No-std compatible, both with and without
alloc
use daft::{Diffable, Leaf};
// Annotate your struct with `#[derive(Diffable)]`:
struct MyStruct {
a: i32,
b: String,
}
// This generates a type called MyStructDiff, which looks like:
struct MyStructDiff<'daft> {
a: Leaf<&'daft i32>,
b: Leaf<&'daft str>,
}
// Then, with two instances of MyStruct:
let before = MyStruct { a: 1, b: "hello".to_owned() };
let after = MyStruct { a: 2, b: "world".to_owned() };
// You can diff them like so:
let diff = before.diff(&after);
// And compare the results:
assert_eq!(*diff.a.before, 1);
assert_eq!(*diff.a.after, 2);
assert_eq!(diff.b.before, "hello");
assert_eq!(diff.b.after, "world");
This crate assigns one side the name before, and the other side after. These labels are arbitrary: if before and after are swapped, the diff is reversed.
Currently, daft comes with a few kinds of diff types:
Leaf
instances
A Leaf
represents a logical leaf node or base case in a diff, i.e. a
point at which diffing stops. Leaf
instances are used for:
- Scalar or primitive types like
i32
,String
,bool
, etc. - Enums, since diffing across variants is usually not meaningful.
- Vector and slice types, since there are several reasonable ways to diff vectors (e.g. set-like, ordered, etc.) and we don’t want to make assumptions.
- As an opt-in mechanism for struct fields: see Recursive diffs below for more.
A contrived example for integers:
use daft::{Diffable, Leaf};
let diff: Leaf<&i32> = 1_i32.diff(&2);
assert_eq!(*diff.before, 1);
assert_eq!(*diff.after, 2);
Enums also use Leaf
:
use daft::{Diffable, Leaf};
// Option<T> uses Leaf:
let diff: Leaf<Option<&i32>> = Some(1_i32).diff(&Some(2));
assert_eq!(diff.before, Some(&1));
assert_eq!(diff.after, Some(&2));
// Automatically derived enums also use Leaf:
enum MyEnum {
A(i32),
B(String),
}
let before = MyEnum::A(1);
let after = MyEnum::B("hello".to_string());
let diff: Leaf<&MyEnum> = before.diff(&after);
assert_eq!(diff.before, &before);
assert_eq!(diff.after, &after);
Vectors use Leaf
as well:
use daft::{Diffable, Leaf};
let before = vec![1, 2, 3];
let after = vec![4, 5, 6];
let diff: Leaf<&[i32]> = before.diff(&after);
assert_eq!(diff.before, &before);
assert_eq!(diff.after, &after);
For BTreeMap
and HashMap
, daft has corresponding BTreeMapDiff
and HashMapDiff
types. These types have fields for common, added,
and removed entries.
Map diffs are performed eagerly for keys, but values are stored as leaf nodes.
use daft::{Diffable, Leaf, BTreeMapDiff};
use std::collections::BTreeMap;
let mut a = BTreeMap::new();
a.insert(1, "one");
a.insert(2, "two");
a.insert(3, "three");
let mut b = BTreeMap::new();
b.insert(2, "two");
b.insert(3, "THREE");
b.insert(4, "four");
let diff: BTreeMapDiff<'_, i32, &str> = a.diff(&b);
// Added and removed entries are stored as maps:
assert_eq!(diff.added, [(&4, &"four")].into_iter().collect());
assert_eq!(diff.removed, [(&1, &"one")].into_iter().collect());
// Common entries are stored as leaf nodes.
assert_eq!(
diff.common,
[
(&2, Leaf { before: &"two", after: &"two" }),
(&3, Leaf { before: &"three", after: &"THREE" })
]
.into_iter().collect(),
);
// If `V` implements `Eq`, unchanged and modified iterators become
// available. `unchanged` and `modified` return key-value pairs;
// `unchanged_keys` and `modified_keys` return keys; and
// `unchanged_values` and `modified_values` return values.
//
// Here's `unchanged_keys` to get the keys of unchanged entries:
assert_eq!(diff.unchanged_keys().collect::<Vec<_>>(), [&2]);
// `modified_values` returns leaf nodes for modified entries.
assert_eq!(
diff.modified_values().collect::<Vec<_>>(),
[Leaf { before: &"three", after: &"THREE" }],
);
For BTreeSet
and HashSet
, daft has corresponding BTreeSetDiff
and HashSetDiff
types. These types have fields for common, added,
and removed entries.
Set diffs are performed eagerly.
use daft::{Diffable, Leaf, BTreeSetDiff};
use std::collections::BTreeSet;
let a: BTreeSet<i32> = [0, 1, 2, 3, 4, 5].into_iter().collect();
let b: BTreeSet<i32> = [3, 4, 5, 6, 7, 8].into_iter().collect();
let diff: BTreeSetDiff<'_, i32> = a.diff(&b);
assert_eq!(diff.common, [&3, &4, &5].into_iter().collect());
assert_eq!(diff.added, [&6, &7, &8].into_iter().collect());
assert_eq!(diff.removed, [&0, &1, &2].into_iter().collect());
For structs, the [Diffable
][macro@Diffable] derive macro generates
a diff type with a field corresponding to each field type. Each field must
implement Diffable
.
A struct Foo
gets a corresponding FooDiff
struct, which has fields
corresponding to each field in Foo
.
Structs can be annotated with #[daft(leaf)]
to treat the field as a leaf
node, regardless of the field’s Diff
type or even whether it implements
Diffable
.
For an example of structs with named fields, see Usage above.
Tuple-like structs produce tuple-like diff structs:
use daft::Diffable;
use std::collections::BTreeMap;
struct MyTuple(BTreeMap<i32, &'static str>, i32);
let before = MyTuple(BTreeMap::new(), 1);
let after = MyTuple([(1, "hello")].into_iter().collect(), 2);
let diff = before.diff(&after);
// The generated type is MyTupleDiff(BTreeMapDiff<i32, &str>, Leaf<i32>).
assert_eq!(**diff.0.added.get(&1).unwrap(), "hello");
assert_eq!(*diff.1.before, 1);
assert_eq!(*diff.1.after, 2);
An example with #[daft(leaf)]
:
use daft::{Diffable, Leaf};
// A simple struct that implements Diffable.
struct InnerStruct {
text: &'static str,
}
// A struct that does not implement Diffable.
struct PlainStruct(usize);
struct OuterStruct {
// Ordinarily, InnerStruct would be diffed recursively, but
// with #[daft(leaf)], it is treated as a leaf node.
#[daft(leaf)]
inner: InnerStruct,
// PlainStruct does not implement Diffable, but using
// daft(leaf) allows it to be diffed anyway.
#[daft(leaf)]
plain: PlainStruct,
}
let before = OuterStruct { inner: InnerStruct { text: "hello" }, plain: PlainStruct(1) };
let after = OuterStruct { inner: InnerStruct { text: "world" }, plain: PlainStruct(2) };
let diff = before.diff(&after);
// `OuterStructDiff` does *not* recursively diff `InnerStruct`, but instead
// returns a leaf node.
assert_eq!(
diff.inner,
Leaf { before: &InnerStruct { text: "hello" }, after: &InnerStruct { text: "world" } },
);
// But you can continue the recursion anyway, since `InnerStruct` implements
// `Diffable`:
let inner_diff = diff.inner.diff_pair();
assert_eq!(
inner_diff,
InnerStructDiff { text: Leaf { before: "hello", after: "world" } },
);
// `PlainStruct` can also be compared even though it doesn't implement `Diffable`.
assert_eq!(diff.plain, Leaf { before: &PlainStruct(1), after: &PlainStruct(2) });
The Diffable
trait can also be implemented manually for custom behavior.
In general, most custom implementations will likely use one of the built-in diff types directly.
Some structs like identifiers should be treated as leaf nodes:
use daft::{Diffable, Leaf};
struct Identifier(String);
impl Diffable for Identifier {
type Diff<'daft> = Leaf<&'daft Self>;
fn diff<'daft>(&'daft self, other: &'daft Self) -> Self::Diff<'daft> {
Leaf {
before: self,
after: other,
}
}
}
If a type parameter is specified, the [Diffable
][macro@Diffable] derive
macro for structs normally requires that the type parameter implement
Diffable
. This is not required if the field is annotated with
#[daft(leaf)]
.
Daft fully supports types with arbitrary lifetimes. Automatically generated
diff structs will have an additional 'daft
lifetime parameter at the
beginning, with the requirement that all other lifetime and type parameters
outlive it.
use daft::Diffable;
struct BorrowedData<'a, 'b, T: Diffable + ?Sized> {
a: &'a str,
b: &'b T,
// TODO: example with daft(leaf)
}
// This generates a struct that looks like:
struct BorrowedDataDiff<'daft, 'a: 'daft, 'b: 'daft, T: ?Sized + 'daft> {
a: Leaf<'daft, &'a str>,
b: T::Diff<'daft>,
}
Implementations for standard library types, all enabled by default:
alloc
: Enable diffing for types from thealloc
crate.std
: Enable diffing for types from thestd
crate.
(With default-features = false
, daft is no-std compatible.)
Implementations for third-party types, all disabled by default:
uuid1
: Enable diffing foruuid::Uuid
.oxnet01
: Enable diffing for network types from the [oxnet
] crate.newtype-uuid1
: Enable diffing fornewtype_uuid::TypedUuid
.
The minimum supported Rust version is 1.81.0. At any time, at least the last three stable versions of Rust are supported.
While this crate is a pre-release (0.x.x) it may have its MSRV bumped in a patch release. Once this crate has reached 1.x, any MSRV bump will be accompanied with a new minor version.
Diffus is the original inspiration for this crate and a great alternative. Daft diverges from diffus in a few ways:
-
Daft’s derive macro does not attempt to diff enums with different variants. In practice, we’ve found that diffing enums across different variants is less useful than it first appears.
-
Daft has the notion of a
Leaf
type, which represents an atomic unit. (For example, theDiffable
implementation fori32
is aLeaf
.)Leaf
s are also used for enums, as well as in any other place where lazy diffing is desired. -
Diffus has a
Same
trait, which is likeEq
except it’s also implemented for floats. Daft doesn’t have theSame
trait, and its core functionality forgoes the need forEq
entirely.For a primitive scalar like
f64
, you’ll get aLeaf
struct which you can compare with whatever notion of equality you want. -
Daft uses a generic associated type (GAT) so that the
Diffable
trait no longer needs a lifetime parameter. This leads to simpler usage. (Diffus was written before GATs were available in stable Rust.) -
Daft uses fewer types in general. For example, diffus wraps its return values in an outer
Edit
type, but daft does not. -
Daft is no-std-compatible, while diffus requires std.
This project is available under the terms of either the Apache 2.0 license or the MIT license.