DynamicVariable in a Continuation

> Hi All,
> It looks like the sematic of DynamicVariable changed not very long ago; it
> was based on Notification, but it now just uses a ProcessVariable. Here is a
> simple testcase:
>
> DynamicVariable subclass: Seed [
>      ].
> Eval [
>      |b g1 g2|
>      b := [:v | v + Seed value].
>      g1 := Generator on: [ :g |
>          |v|
>          v := 0.
>          Seed use: 2 during: [
>              [ g yield: v.
>                v := b value: v ] repeat ]].
>      g2 := Generator on: [ :g |
>          |v|
>          v := 0.
>          Seed use: 3 during: [
>              [ g yield: v.
>                v := b value: v ] repeat ]].
>      Transcript nextPutAll: '%1 %2 %3 %4' % {
>          g1 next.
>          g2 next.
>          g1 next.
>          g2 next}; nl.
> ]
>
> It used to yield "0 0 2 3" but yields "0 0 3 3" now. I'd think "0 0 2 3" is
> the correct answer; the documantion specifically mentioned stackframes so
> the stackframes hopping from the Continuation should affect the variable,
> right?

> On 05/24/2011 08:20 AM, Derek Zhou wrote:
> >Hi All,
> >It looks like the sematic of DynamicVariable changed not very long ago; it
> >was based on Notification, but it now just uses a ProcessVariable. Here is a
> >simple testcase:
> >
> >DynamicVariable subclass: Seed [
> >     ].
> >Eval [
> >     |b g1 g2|
> >     b := [:v | v + Seed value].
> >     g1 := Generator on: [ :g |
> >         |v|
> >         v := 0.
> >         Seed use: 2 during: [
> >             [ g yield: v.
> >               v := b value: v ] repeat ]].
> >     g2 := Generator on: [ :g |
> >         |v|
> >         v := 0.
> >         Seed use: 3 during: [
> >             [ g yield: v.
> >               v := b value: v ] repeat ]].
> >     Transcript nextPutAll: '%1 %2 %3 %4' % {
> >         g1 next.
> >         g2 next.
> >         g1 next.
> >         g2 next}; nl.
> >]
> >
> >It used to yield "0 0 2 3" but yields "0 0 3 3" now. I'd think "0 0 2 3" is
> >the correct answer; the documantion specifically mentioned stackframes so
> >the stackframes hopping from the Continuation should affect the variable,
> >right?
>
> You are correct.  The change was made because it was hundreds of
> time faster to use ProcessVariable, but I missed this use case.  I
> have to think about it.  Do you have any suggestion on how to name
> the classes to keep the current one available, and still provide
> "real" DynamicVariables?

> On 05/26/2011 12:29 AM, Derek Zhou wrote:
> >I am not very good at naming, but can we keep DynamicVariable a subclass of
> >Notification, and provide 2 methods:
> >use_during_: for the fast way by making use a ProcessVariable
> >use_inside_: for the slow way of stackframe backtracing.
>
> Interesting idea if it can be implemented!  But how would #value
> know which one to use?  The slow thing is not _setting_ the
> variable, it is _reading_ it, and it seems to me that without doing
> stack inspection you cannot know whether you are in a #use:inside:
> block.
>