Hi Florin,

2010/10/26 Florin Mateoc <fmateoc@gmail.com>
Eliot wrote:

Hi Levente,

    in my experience playing with the Dan and Alan Borning's multiple
inheritance implementation getting things to work was one thing, reasonably
straighforward, but getting tool support right (file-out format so that
things filed back in correctly, etc) was another, an dth ebulk of the work.
Looking at your proposal below I see no problem getting it to work but see
lots of problems integrating it, e.g. with Monticello, etc.  Traits is
already integrated with Monticello, supported with tests, etc.  As such it
is a much less risky or effortful proposition than a new scheme.  So with
that in mind are there important functional benefits of your scheme's direct
state access as compared with traits that use accessors to access state that
each user of a trait must implement?  i.e. what are the key benefits you see
in direct inst var access?

best
Eliot

Hi Eliot,

I think there is a strong argument for stateful traits, especially since we are talking about collection smells.
One of the main selling points for traits (for me, at least) is that they would allow us to clean up the mess
created by inheritance for reuse (Dictionary is a Set) in existing libraries.
And obviously to create cleaner libraries going forward.
Well, inheritance for reuse is used not just for behavior, but for state as well.
Stateful traits would allow for a cleaner separation between inheritance and subtyping.
Stateful traits would even allow you to push the separation further and, as part of your design, you could have state-only traits
complementing behavior-only traits.
Let us not forget that clean concepts are also easier to learn. I think it is much more important to get the concepts right
than any existing integration with Monticello. Why does it matter that we have tests and integration if people are not
using traits?

The argument is not over �the utility of stateful traits. �The argument is over how to access that state. �The existing Traits implementation allows indirect access to state, where the Trait uses accessors that must be implemented by a class using the trait. �This has the disadvantages listed by Levente, possibility of accessor selectors clashing with existing protocol, lack of privacy and poorer performance. �One possible solution is to do a new traits implementation that supports state, but as discussed integration is hard and that could be a lot of work. �There are other solutions to explore that could localise the changes.

For example imagine adding some special illegal syntax that a variant of the compiler (a compiler mode or a subclass or...) can deal with and using that to reference trait state. Then changes are localised to�
- the compiler (to compile the syntax down onto the relevant inst vars)
- class definition/trait composition (to ensure the state inst vars exist in the class)
- shout (to highlight the new syntax)
- the decompiler (so decompiled trait code displays the new syntax for trait state access)

Another possibility might be a selector privacy hack. �The selectors used to access state in a trait are known not to clash in the trait (they ca be so chosen by the trait author not to conflict with anything in the trait) but they cannot be guaranteed not to clash with a user of the trait. �So post-compile the selectors for the trait's state could be changed to private values (keys that existed only in the method dictionary of the trait user). �This doesn't have the performance of direct inst var access, but it solves the clash and privacy issues.

And I expect there are other solutions too.

In summary don't argue for or against stateful traits, instead think about how to add state to traits whilst having the most minimal impact on the rest of the infrastructure.

best
Eliot


Cheers,
Florin

_______________________________________________
Pharo-project mailing list
Pharo-project@lists.gforge.inria.fr
http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project