Crates.io | bcount |
lib.rs | bcount |
version | 0.1.0 |
source | src |
created_at | 2015-11-25 20:56:15.541304 |
updated_at | 2015-12-12 23:55:19.918092 |
description | Borrow counted types |
homepage | https://github.com/luthaf/bcount |
repository | https://github.com/luthaf/bcount |
max_upload_size | |
id | 3501 |
size | 11,001 |
This crate provide type for counting mutable borrow of a value. The Bc<T>
type is a small wrapper on top of a value of type T
which count the number
of time the value has been mutably borrowed since it's creation.
If you want to cache an expensive computation result, you need to have information about wether the parameters of the computation have changed or not. You can use a hash for that, but this have two shortcomings:
Hash
trait. Some useful
types like f64
do not;If computing the hash is harder or more expensive than doing the
computation, you are doomed. Or you can use Bc<T>
which will give you
information about the number of borrow since the last computation. If this
number have changed, then it is very likely that the value have changed,
and that you need to redo your computation.
This can not be used a real hash algorithm, because the number of borrow can change even if the value do not.
If more than usize::MAX
borrow occurs, the borrow counter will be wrapped
around to 0, and will not panic
because of the overflow.
The Bc
type do not introduce notable overhead when borrowing. Here are the
benchmark results comparing a raw value and a borrow counted value:
running 2 tests
test counted ... bench: 1,061 ns/iter (+/- 12)
test raw ... bench: 1,059 ns/iter (+/- 16)
Using the number of borrow as hash value is way faster than doing the real
hash. Here is a benchmark for [usize; 10000]
values:
running 2 tests
test counted ... bench: 22 ns/iter (+/- 1)
test raw ... bench: 49,011 ns/iter (+/- 755)
You can see the code for these benchmarks on Github.
extern crate bcount;
use bcount::Bc;
fn main() {
let mut a = Bc::new(vec![63, 67, 42]);
assert_eq!(a.count(), 0);
do_work(&mut a);
do_work(&mut a);
do_work(&mut a);
do_work(&mut a);
assert_eq!(a.count(), 4);
*a = vec![3, 4, 5];
assert_eq!(a.count(), 5);
}
fn do_work(_: &mut [usize]) {
// Whatever, nobody cares
}
MIT