[Cuis-dev] Possible improvements for method lookup simulation

[Juan Vuletich]

Hi Facu,

Just ntegrated and pushed to GitHub. Thanks!

Cheers,

On 2026-09-21 2:18 PM, Facundo Javier Gelatti via Cuis-dev wrote:
> Hello!
>
> I've been experimenting with some improvements for the messages we use 
> to simulate the method lookup (i.e. Behavior>>#lookupSelector:), 
> motivated by the OOP courses I teach (where I like to show the 
> implementation of the method lookup). I figured the changes might be 
> of interest for integration on the Cuis base image, so I'm sending 
> some proposals!
>
> The changes I propose are twofold:
> 1. Change sets 1-2x are a proposal to have a new 
> #lookupSelector:ifAbsent: message, to be able to specify what to do 
> when the selector is not found (instead of always returning nil). The 
> change set labelled 1 introduces it, and the ones labelled as 2x use 
> it in some of the methods we already have. I tried to name the files 
> as descriptively as possible, to help during review.
> 2. Change set 3 is a proposal for a new implementation of 
> #lookupSelector:ifAbsent:. The implementation I propose avoids using a 
> while loop, it's a recursive implementation which I find a bit less 
> imperative (and, might I say, "more object-oriented"). Surprisingly, 
> the implementation has also better performance compared with the 
> current one (it runs about 30% faster)¹.
>
> To be able to evaluate the second proposal, I attach some useful scripts:
> * Some benchmarks for the average case and two "worst cases".
> * A (partial) correctness test, which asserts that the methods found 
> by both implementations are the same (traversing all the methods in 
> the image).
> * A script to try it out with very deep class hierarchies.
> These scripts should be evaluated manually in a workspace, while 
> switching between the two implementations.
>
> I hope you find these changes, if not useful, at least interesting!
> Cheers!
> Facu
>
> ______
> ¹ Of course, being a recursive implementation, it limits the class 
> hierarchy depth by the stack size. But since the depth of all 
> inheritance hierarchies on the image is less than 20, and having a 
> deep inheritance hierarchy is already a design smell, we might 
> consider taking the design + performance improvement. Also, currently 
> the maximum stack size is limited by RAM (at least by default), so the 
> change works with very large hierarchies too (see the attached 
> very_deep_hierarchy_test.st <http://very_deep_hierarchy_test.st> script).
>
-- 
Juan Vuletich
www.cuis.st
github.com/jvuletich
researchgate.net/profile/Juan-Vuletich
independent.academia.edu/JuanVuletich
patents.justia.com/inventor/juan-manuel-vuletich
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.cuis.st/mailman/archives/cuis-dev/attachments/20260924/8bd88fd4/attachment.htm>