[PEAK] Re: trellis.Set.discard

[Phillip J. Eby]

At 09:39 PM 11/3/2008 +0200, Sergey Schetinin wrote:
>We also need to consider that cells can be passed to
>Component.__init__. Consider this:
>
>class C1(Component):
>     val = attr()
>
>     @maintain
>     def rule(self):
>         self.val = True
>
>val = Cell()
>
>@atomically
>def test():
>     c1 = C1(val=val)
>     assert val.value is True
>
>Not just the old values could be set, but also old rules run etc. In
>some cases this is invalid, but in some, when the running rule didn't
>run in current transaction yet it should be OK. But then, the old rule
>should be undoable, the new ones shouldn't.

I think that we need to stop thinking of component initialization as 
a side transaction.  It really doesn't help in understanding what 
should be done.


>...
>This is almost true, but consider the case when we create a component
>and immidiately set one of its discrete cell, this would mean that the
>dependent rules will see the init value, the newly set value, but
>never the reset value in between.

There are more problems with this than just that.  See the email 
before this one.


>I have a number of small reports / suggestions, would you prefer if I
>created a few new threads for them (they have nothing to do with
>Set.discard anyway =) or is it ok just to list them all in one?

All in one would be fine; might as well start a new thread, 
also.  The sooner you can do it the better, because this is probably 
the only day this week I will be working on any of this.