call site read strategy learning

[Chris Muller]

That's a neat idea, and there is already MagmaProxyMaterialization
notification which *may* help to implement it.

However, I'm wondering whether it is necessary to be so precise.  At
most, only two strategy's for most kinds of objects are really needed;
one for a "summary" (as in displaying in a list) and one for "detail".
 I would think err'ing on the side of being liberal would optimise
performance.

In fact, I have been thinking whether the minimumDepth should be
bigger; at least 2, maybe 3, and 4 to be agressive..?  Anyone have any
thoughts on that?

I know everyone wants Magma to be faster, but ReadStrategy's will only
take us so far, I hope to find time to look at other areas that can be
improved..  (and welcome ideas from who has themselves looked into it,
of course).



On 8/15/07, Igor Stasenko <siguctua at gmail.com> wrote:
> I wonder, is it possible to implement a something which will learn
> till code runs?
> Lets say i have in some method a block, which invokes code for
> accessing some db objects  and/or does some manipulations with them.
> What if at first invocation startup magma will learn from this
> process, so the next time it enters this block of code, it will
> already know how deep particular object types must be read to be
> optimal for speed/space?
>
> something as easy as:
>
> handle := session captureReadStrategy: [ some code ].
>
> and then:
>
> session withReadStrategy: handle do: [ some code ].
>
> or ever:
> session useAndCapture: handle do: [some code].
>
> I think it can be clever enough to use as default approach for
> creating read strategies, so developers will be freed from making tons
> of tests and supposes for making static read strategies work fine with
> their code.
>
> Even, if its not or hard to adopt, i think some analysis can be done
> while running user code (like messagetally), so developer can then
> make decisions based on it.
>
> --
> Best regards,
> Igor Stasenko AKA sig.
> _______________________________________________
> Magma mailing list
> Magma at lists.squeakfoundation.org
> http://lists.squeakfoundation.org/mailman/listinfo/magma
>