Defered.setFinal

1 view
Skip to first unread message

Amit Mendapara

unread,
Feb 10, 2009, 8:30:36 AM2/10/09
to moch...@googlegroups.com, Per Cederberg
Hi Per,

I have just started again improving the MochiKit Extensions. While
creating tests for the Ajax module, I found one problem (not bug, but
specific to the feature I'm trying to implement). I registered a
callback which should be fired at the end of all registered callbacks.

I achieved by modifying Async.js with a new method `setFinal` (not
addFinal as there should be only one finalizer) which gets fired when
`chain.length == 0`. It's simple. I would we happy if you add one such
function in MochiKit.Async.Defered...

Regards
--
Amit

Async.patch

Per Cederberg

unread,
Feb 10, 2009, 11:04:56 AM2/10/09
to MochiKit
I think this is a good idea. I needed something similar too, so I
ended up writing an ugly hack that worked most of the time...

d.addBoth(function (res) {d.addBoth(finalFunc); return res; });

It adds new callback once the first deferred result drops in,
hopefully after all the other callbacks have been added...

But a more formally correct solution would probably be a good idea.

Cheers,

/Per

Bob Ippolito

unread,
Feb 10, 2009, 11:48:44 AM2/10/09
to Per Cederberg, MochiKit
Finalizing a Deferred should ensure that no further callback/errbacks
are registered and it should attach a default error handler (success
should be no-op). The most common problem I've seen with Deferreds is
that an error occurs but nobody attached an error handler that far
down the stack. In Python they work around this by having a finalizer
so that you see the error when the object gets GC'ed, but that's not
really possible in JS since you don't have finalizers or weak
references.

Amit Mendapara

unread,
Feb 10, 2009, 12:58:20 PM2/10/09
to Bob Ippolito, Per Cederberg, moch...@googlegroups.com
Per,

Your hack won't work with my case as it would mark the deferred as
`chained` so consecutive addCallback/addErrback won't work. Considering
Bob's suggestion, I modified my patch to ensure no further
callback/errbacks can be registered.

Regards
--
Amit

Async.patch

Per Cederberg

unread,
Feb 10, 2009, 2:52:30 PM2/10/09
to MochiKit
You mean like this?

setFinalizer: function(cb) {
this._finalizer = cb;
}

...

if (this.chain.length == 0 && this._finalizer) {
this.finalized = true;
if (res instanceof Error) {
this._finalizer(res);
}
}

Would that be acceptable to Amit as well? In that case we would add a
default error handler (called a "finalizer"), which would also be what
I'm planning to use this for.

Other opinions?

/Per

Amit Mendapara

unread,
Feb 11, 2009, 12:58:06 AM2/11/09
to MochiKit
It's not just error handler, it should always be called at the end
once result is available and all callbacks/errback (including default
errback) are fired.

setFinalizer: function(cb) {
this._finalizer = cb;
}

if (this.chain.length == 0 && this._finalizer) {
this.finalized = true;
this._finalizer(res);
}

--Amit

On Feb 11, 12:52 am, Per Cederberg <cederb...@gmail.com> wrote:
> You mean like this?
>
>     setFinalizer: function(cb) {
>         this._finalizer = cb;
>     }
>
>     ...
>
>     if (this.chain.length == 0 && this._finalizer) {
>         this.finalized = true;
>         if (res instanceof Error) {
>             this._finalizer(res);
>         }
>     }
>
> Would that be acceptable to Amit as well? In that case we would add a
> default error handler (called a "finalizer"), which would also be what
> I'm planning to use this for.
>
> Other opinions?
>
> /Per
>
> On Tue, Feb 10, 2009 at 5:48 PM, Bob Ippolito <b...@redivi.com> wrote:
> > Finalizing a Deferred should ensure that no further callback/errbacks
> > are registered and it should attach a default error handler (success
> > should be no-op). The most common problem I've seen with Deferreds is
> > that an error occurs but nobody attached an error handler that far
> > down the stack. In Python they work around this by having a finalizer
> > so that you see the error when the object gets GC'ed, but that's not
> > really possible in JS since you don't have finalizers or weak
> > references.
>

Amit Mendapara

unread,
Feb 11, 2009, 7:52:21 AM2/11/09
to MochiKit
Even better...

if (this.chain.length == 0 && this._finalizer) {
this._finalized = this._finalizer(res);
}

This will allow us to finalize the deferred based on some
conditions...

- Amit

On Feb 11, 12:52 am, Per Cederberg <cederb...@gmail.com> wrote:
> You mean like this?
>
>     setFinalizer: function(cb) {
>         this._finalizer = cb;
>     }
>
>     ...
>
>     if (this.chain.length == 0 && this._finalizer) {
>         this.finalized = true;
>         if (res instanceof Error) {
>             this._finalizer(res);
>         }
>     }
>
> Would that be acceptable to Amit as well? In that case we would add a
> default error handler (called a "finalizer"), which would also be what
> I'm planning to use this for.
>
> Other opinions?
>
> /Per
>
> On Tue, Feb 10, 2009 at 5:48 PM, Bob Ippolito <b...@redivi.com> wrote:
> > Finalizing a Deferred should ensure that no further callback/errbacks
> > are registered and it should attach a default error handler (success
> > should be no-op). The most common problem I've seen with Deferreds is
> > that an error occurs but nobody attached an error handler that far
> > down the stack. In Python they work around this by having a finalizer
> > so that you see the error when the object gets GC'ed, but that's not
> > really possible in JS since you don't have finalizers or weak
> > references.
>

Per Cederberg

unread,
Jul 20, 2009, 5:26:12 AM7/20/09
to MochiKit
Slow response here, but I finally got around to committing this to
Subversion (r1533):

http://trac.mochikit.com/changeset/1533

Modified the patch a bit further to account for some additional cases.
Also added tests and documentation.

@Bob: Can you please review this change? I have the feeling that I've
still missed one or two corner cases... Also, the mochikit.com and
trac.mochikit.com site seems slow from here... High server load
recently?

@Amit: Please check that my additional controls doesn't break you use case.

Cheers,

/Per

Amit Mendapara

unread,
Jul 20, 2009, 6:03:55 AM7/20/09
to MochiKit
Hi Per,

I have just tested the changes with the Ajax extension module and it
works as expected. Thanks for considering adding this feature. This
would help me creating jQuery compatible Ajax module for Mochikit
extensions project.

Currently, I'm quite busy with some another work but supposed to come
back on MochiKit Extensions soon...

Regards
--
Amit Mendapara

On Jul 20, 2:26 pm, Per Cederberg <cederb...@gmail.com> wrote:
> Slow response here, but I finally got around to committing this to
> Subversion (r1533):
>
>  http://trac.mochikit.com/changeset/1533
>
> Modified the patch a bit further to account for some additional cases.
> Also added tests and documentation.
>
> @Bob: Can you please review this change? I have the feeling that I've
> still missed one or two corner cases... Also, the mochikit.com and
> trac.mochikit.com site seems slow from here... High server load
> recently?
>
> @Amit: Please check that my additional controls doesn't break you use case.
>
> Cheers,
>
> /Per
>

Giulio Cesare Solaroli

unread,
Jul 25, 2009, 6:51:59 PM7/25/09
to Bob Ippolito, Per Cederberg, MochiKit
Hello all,

On Tue, Feb 10, 2009 at 6:48 PM, Bob Ippolito<b...@redivi.com> wrote:
>
> Finalizing a Deferred should ensure that no further callback/errbacks
> are registered and it should attach a default error handler (success
> should be no-op). The most common problem I've seen with Deferreds is
> that an error occurs but nobody attached an error handler that far

> down the stack. [...]

In order to handle this case, we have overwrote the 'callback' method
in our subclass of the Deferred class, to add a default Errback that
at least logs the error.

A similar solution could be probably used also in MochiKit, allowing
the user to optionally specify a specific error handler / finalizer.
This will allow any further check, as once the Deferred is triggered
it is already impossible to add further callback/errback.

I will check anyway the fix added to the repository to see how to
improve our current extension with the committed fix.

Giulio Cesare

Reply all
Reply to author
Forward
0 new messages