3 releases
0.1.2 | Aug 8, 2024 |
---|---|
0.1.1 | Aug 2, 2024 |
0.1.0 | Aug 2, 2024 |
#861 in Data structures
97 downloads per month
20KB
354 lines
hashrevise
A Rust library for hashing data structures and then caching results and avoiding work.
The main component of the library is RevisionHash
, which just an integer summary/digest of the contents of an arbitrary data structure through the Revisable
trait and its get_revision()
method. Through the use of hashing, when an object's RevisionHash
is changed, we know the contents have changed, and conversely, if the RevisionHash
is unchanged, with very high probability its contents also are unchanged. Thus, in a larger program, by storing a hash of some data model from a previous known good state, we can easily compare its new hash and accurately determine whether or not to perform an expensive update from the new data.
A basic implementation of the Revisable
trait is as follows. An implementation should hash all fields that are relevant to the identity of the object and the contents it represents.
struct Test(i32, u8);
impl Revisable for Test {
fn get_revision(&self) -> RevisionHash {
let mut hasher = RevisionHasher::new();
hasher.write_i32(self.0);
hasher.write_u8(self.1);
hasher.into_revision()
}
}
Recursion is encouraged wherever objects contain sub-objects that also implement Revisable
, simply with a line like hasher.write_revisable(&self.subobject);
.
Care should be taken when dealing with enums (you should probably hash the discriminator first). Care should also be taken when hashing variable-size data structures like Vec
and HashMap
, although some blanket implementations already exist which simplify this.
A helper struct Revised<T>
is provided which looks and acts like an object of type T
but caches the RevisionHash
of that object as needed whenever it isn't being mutably accessed. This is useful when composing larger data structure to ensure that not all contents need to be read and hashed during partial modifications or if no modifications have been made at all. Conceivably, this could also be used for lightweight equality comparisons of very large data structures.
Another helper struct RevisedProperty<T>
is provided which represents a lazily-computed result of calling some function which returns T
and whose arguments are all Revisable
. Its methods refreshN(f, arg0, arg1, ... argN)
will call f
on the provided arguments only if the arguments if different from the last call to a refresh
method. This can be used to avoid expensive computations while still ensuring that derived results are up-to-date. The result of calling f
is available through the get_cached()
method.
I may add other helpers when I feel clever and have a use for them.
Dependencies
~44KB