[reactive] switching events = space leak?

[John Lato]

sodium and elerea both have garbage collection for dynamically created
events.  However, in practice I find we often need to call performGC to
make it happen, because ghc's runtime doesn't seem to schedule major gc's
often enough.  But that may only be because we run our reactive app at full
tilt, so there are very few unused cycles already.

I don't think it would be all that difficult to add support for this to
reactive-banana, but Apfelmus indicated there are a few design issues that
need to be resolved when I last discussed this with him.  That was a few
months ago though, so maybe he's made some progress.

John


On Fri, Feb 8, 2013 at 5:59 PM, Stefan Kersten <sk at k-hornz.de> wrote:

> On Feb 8, 2013, at 1:13 AM, John Lato wrote:
> > elerea itself doesn't follow the event/behavior distinction, so it looks
> quite a bit different from the others.  Also it doesn't have a particular
> switch combinator, rather switching behavior is enabled by the use of
> recursive-do bindings.
> >
> > I think reactive-banana's approach is expressive enough to do as you
> suggest, but I haven't actually tried it yet.
>
> one sentence in the package documentation is holding me back from using
> event switching:
>
> "There is currently no garbage collection for dynamically created events."
>
> any ideas how much work it would be to to fix this? how does sodium
> compare in this regard?
>
> (cc'ing reactive-banana's author in case he's interested in joining the
> discussion here ...)
>
> sk
>
> > From: Don Vincenze <don at tr80.com>
> >
> > Wow thank you for all these pointers ... I must take the time to look
> > into these libraries. Stuff like this doesn't come up when you Google
> > "Functional Reactive Programming" or at least doesn't rank very high.
> >
> >
> >
> > I'm not sure I understand the sodium approach (couldn't find switch in
> > elerea). The aging/'trimming' approach in banana seems to rely on the
> > events being switched into being known from the start, i.e. the point
> > in time 'switch' starts executing. This way you can dynamically switch
> > into a static set of events but not dynamically into a dynamic set of
> > events.
> >
> >
> >
> > In any case, considering such concerns seem to be transcend individual
> > libraries/frameworks, you wouldn't know of any mailing list/forum that
> > is more suitable for this discussion, i.e. a general FRP forum not tied
> > to any library or one purely focused on theory?
> >
> >
> >
> >
> >
> > On Wed, Feb 6, 2013, at 07:22 PM, John Lato wrote:
> >
> >
> > This can be a serious issue, however most modern FRP implementations
> > (by which I mean at least reactive-banana, elerea, and sodium) have
> > solutions.  I'm sure elm does as well, but I don't have experience with
> > it.
> >
> > I'm aware of two general approaches.  The first, used by elerea and
> > sodium, enforces that the creation of signaling and switching
> > constructs happen within a monad.  In this case the switching primitive
> > looks something like
> >
> > mSwitch :: Behavior (SGen a) -> SGen (Behavior a)
> >
> > this function is responsible for ensuring that signals and behaviors
> > are represented in the monad, and thus executed.
> >
> > The other approach was introduced in Grapefruit, and to my knowledge is
> > only used in Grapefruit and reactive-banana.  Instead of a signal
> > generation monad, switchable constructs have an extra type parameter
> > that's used to enforce aging.  apfelmus has written a bit about it
> > at [1]http://apfelmus.nfshost.com/blog/2012/09/03-frp-dynamic-event-swi
> > tching-0-7.html .  Conceptually it's very similar to how the "st"
> > parameter is used in the ST monad if you're familiar with that.
> >
> > as to scaling, it is possible for a reactive network to scale to fairly
> > large systems, but not all of them do.  I've been meaning to blog about
> > this soon, maybe over the next week or two.
> >
> > John L.
> >
> > References
> >
> > 1.
> http://apfelmus.nfshost.com/blog/2012/09/03-frp-dynamic-event-switching-0-7.html
> > -------------- next part --------------
> > An HTML attachment was scrubbed...
> > URL: <
> http://www.haskell.org/pipermail/reactive/attachments/20130207/d4748a02/attachment.html
> >
> >
> > ------------------------------
> >
> > _______________________________________________
> > Reactive mailing list
> > Reactive at haskell.org
> > http://www.haskell.org/mailman/listinfo/reactive
> >
> >
> > End of Reactive Digest, Vol 18, Issue 3
> > ***************************************
> >
> > _______________________________________________
> > Reactive mailing list
> > Reactive at haskell.org
> > http://www.haskell.org/mailman/listinfo/reactive
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.haskell.org/pipermail/reactive/attachments/20130211/ff8e3d2f/attachment.htm>