# fn_cache crate

This crate implements an easy way to cache values for
a function. If you have a slow running function, this
can be used to speed up successive runs dramatically.
It is also quite useful for memoization of recursive
functions, to prevent calculating the same function
twice in different calls.

Of particular note, this caching is done without
cloning or copying, allowing functions to return
large objects, while the cache only returns a reference
to them instead of copying them.

# Allowed functions
This crate attempts to remain fairly flexible with
the functions it accepts. All of the following should
be allowed:
  * [`fn`][fn primitive] types.
  * [`Fn`] types that have no references.
  * [`Fn`] + 'static types that take only static references.
  * [`Fn`] + 'a types that take references of lifetime 'a.

For obvious reasons, [`FnMut`] and [`FnOnce`] are not allowed,
as functions need to be rerunnable and pure.

# Examples

The caches can handle recursive functions, with a shortcut
for defining it with non-recursive functions. Each cache
has a `recursive` function to create a recursion capable
cache, but requires the function to accept the cache as the
first argument. Each `new` function takes a function that
does not require the cache as an argument.

## Non-recursive

Here is an example for a function that takes a while to
calculate. Instead of running the calculations each time
you'd like to do it just once, and recall the value. The
results are stored in a `HashCache` for random access.

```rust
use fn_cache::{FnCache, HashCache};
use std::{thread, time};

let sleep_time = time::Duration::from_secs(3);

let mut cache = HashCache::new(|&x| {
    thread::sleep(sleep_time);
    x
});

let start = time::Instant::now();
assert_eq!(cache.get(100), &100);
assert_eq!(cache.get(100), &100);

// time elapsed is only slightly longer than the sleep time
// far less than twice.
assert!(time::Instant::now() - start < sleep_time.mul_f32(1.1));
```

## Recursive

The following example shows a recursive fibonacci
implementation, which would be O(2ⁿ) without
memoization (caching). With memoization, it becomes
O(n), and can easily be calculated.

```rust
use fn_cache::{FnCache, HashCache};

let mut cache = HashCache::<u8,u128>::recursive(|cache, x|
    match x {
        0 => 0,
        1 => 1,
        _ => *cache.get(x - 1) + *cache.get(x - 2),
    }
);

assert_eq!(
    *cache.get(186),
    332_825_110_087_067_562_321_196_029_789_634_457_848
);
```

For even bigger results, the [num] crate might be employed.
In order to avoid copying the `BigUint`s while accessing the
cache twice, using [`FnCacheMany::get_many`] can be used to get
multiple values at once, to avoid trying to take a reference and
then mutating again. Additionally, since the inputs start at 0
and each value must be filled before the next is calculated, you
might use a [`VecCache`] as an optimization.

```rust
use fn_cache::{FnCache, FnCacheMany, VecCache};
use num_bigint::BigUint;

let mut cache = VecCache::recursive(|cache, x|
    match x {
        0 => BigUint::new(vec![0]),
        1 => BigUint::new(vec![1]),
        _ => cache.get_many([x - 1, x - 2]).into_iter().sum(),
    }
);

assert_eq!(
    cache.get(999),
    &BigUint::parse_bytes(b"26863810024485359386146727202142923967616609318986952340123175997617981700247881689338369654483356564191827856161443356312976673642210350324634850410377680367334151172899169723197082763985615764450078474174626", 10).unwrap()
);
```

[fn primitive]: https://doc.rust-lang.org/std/primitive.fn.html
[`Fn`]: https://doc.rust-lang.org/std/ops/trait.Fn.html
[`FnMut`]: https://doc.rust-lang.org/std/ops/trait.FnMut.html
[`FnOnce`]: https://doc.rust-lang.org/std/ops/trait.FnOnce.html
[`Rc`]: https://doc.rust-lang.org/std/rc/struct.Rc.html
[num]: https://docs.rs/num/