This is cool indeed. I find an unlimited cache a bit coarse, would not really work in practice. What about memoizedUsing: cache ^ [ :x | cache at: x ifAbsentPut: [ self value: x ] ] then you could pass in a tuned LRUCache instance, among others. Sven

On 11 Nov 2014, at 16:24, Torsten Bergmann <astares@gmx.de> wrote:

Phil wrote:

...

thanks for the pointer. Now, how do I use that?

First: create a method BlockClosure>>#memoized in your Pharo image (see https://gist.github.com/jcromartie/2705526)

memoized "returns memoized version of an unary function" | cache | cache := Dictionary new. ^ [ :x | cache at: x ifAbsentPut: [ self value: x ] ]!

So the idea of #memoized is to return a new "outer" block who evaluates the original block only when the computation of the inner block was not yet done/the result of the inner evaluation is not yet cached.

As you see from the comment it works for "unary functions" which would mean it works for "one argument blocks". Simple example:

|square memo | square:= [:x | Transcript show: 'Processing'. x * x]. memo := square memoized. memo value: 2. memo value: 3. memo value: 2.

The Transcript only shows "Processing" two times instead of three times, so only for the first two #value: calls we have to calculate x * x.

For the last evaluation on the memo block (with argument 2 again) the result is already known from the cache, so the square block is not evaluated again. Instead the cached result could be returned which could be faster than doing the computation again.

x*x is only an example - this "memoization" can usually speed up when a block is costly (performancewise) and is often called with the same arguments. A precondition (for this caching) is that the block evalutes the same way/returns the same result for the same input argument.

You also try the two prepared workspaces in Pharo:

Non-memoized version: http://ws.stfx.eu/5ZE87XSFL9IH Memoized Version: http://ws.stfx.eu/QLQLSLT7PZ4C

Hope this makes it more clear. This simple example is a nice performance trick and shows again the power of Smalltalk and block closures.

Dont know about the license - but we could ask author John Cromartie for MIT license to include this simple method into the standard image. I will try to contact him.

Bye T.