Memoizer turns an ordinary Elixir module into a transparent service that memoizes (aka caches) your function calls. The first function call may take a very long time:

iex> :tc.timer(Ultimate, :answer, [:life, :universe, :everything]) {7.5 million years in microseconds, 42}

but further calls happen almost instantaneously when the answer is already available:

iex> :tc.timer(Ultimate, :answer, [:life, :universe, :everything]) {0, 42}

Here’s a simple example how to us it. Let’s make the module below a memoized one:

defmodule Memo do
  def fact(0), do: 1
  def fact(n), do: n * fact(n - 1)
end

The modification is simple:

defmemo Memo do
  use Memoizer

  def     fact(0), do: 1
  defmemo fact(n), do: n * fact(n - 1)
end

It is also important that Memo has to be started before you start calculating values. The simplest way is to call Memo.start_link if you are just playing around, but you should consider putting it under supervision.

When the service is running, you can make calls to the functions just as they were defined in the first place: Memo.fact(100). This caches the return values of fact/1 for params between 2 and 100. Next time you call the function within this range, the stored value is returned.

With defmemo you can also use when gueards, and it’s legal to mix simple defs with defmemo definitions. This enables you to have a fine grained controll over what makes it into the cache and what gets calculated over and over again with every function call. For instance, it does not make much sense to memoize fact(0) in the example above, so we can keep this case in a normal function clause. Here’s an example of calculating the Fibonacci sequence with only memoizing fib(n) when n or n-1 can be divided by 7. This reduces the memory footprint to 2/7 of the version memoizing every value:

def     fib(0),                     do: 0
def     fib(1),                     do: 1
def     fib(n) when rem(n, 7) == 2, do:  1 * fib(n - 1) +  1 * fib(n - 2)
def     fib(n) when rem(n, 7) == 3, do:  2 * fib(n - 2) +  1 * fib(n - 3)
def     fib(n) when rem(n, 7) == 4, do:  3 * fib(n - 3) +  2 * fib(n - 4)
def     fib(n) when rem(n, 7) == 5, do:  5 * fib(n - 4) +  3 * fib(n - 5)
def     fib(n) when rem(n, 7) == 6, do:  8 * fib(n - 5) +  5 * fib(n - 6)
defmemo fib(n) when rem(n, 7) == 0, do: 13 * fib(n - 6) +  8 * fib(n - 7)
defmemo fib(n) when rem(n, 7) == 1, do: 21 * fib(n - 7) + 13 * fib(n - 8)

You may want to normalize the arguments of a function call before making the expensive function call, or post-process the returned value before returning it to the caller. In this case you can use defmemop to define your calculation, and wrap the call in a public function that makes the necessary transformations and normalizations. defmemop can be mixed with defp caluses just like defmemo and def.

While memoizing pure functions does not change the semantics of your program, nothings stops you applying the technique to non-pure functions. This may make sense in certain situations. In this case functions like Memoizer.forget, Memoizer.memoized? and Memoizer.learn can come in handy. It is a deliberate decision that these functions are kept in the Memoizer module and are not injected into the modules that use Memoizer.

If you are after more performance, Memoizer.compile can give you some more speed by compiling the currently cached function clauses into a separate module.