adding valid signals to a CallbackRegistry?

Hello all,

I am working to make mplot3d feature-parity with regular axes objects. I have come across a possible design flaw with the CallbackRegistry.

Many of the Axes3D methods are merely wrappers around Axes methods letting it do the work for x and y axis and then let Axes3D do the same for the z axis. In Axes.cla(), self.callbacks gets a CallbackRegistry of signals named “xlim_changed” and “ylim_changed”. However, once that is set, it does not appear to be a way for me to add another signal of “zlim_changed” in Axes3D.cla(). I could replace self.callbacks with a new CallbackRegistry, but since there is a lot of interveaning code between that first initialization of self.callbacks and coming back out of Axes.cla(), I worry that some callbacks may have been registered by then and would get eliminated in the process.

Is there a recommended way around this? Or maybe this is more evidence that we should move towards the use of a dictionary of axis objects and make Axes functions more agnostic to the number of axis?

Thanks,

Ben Root

I don’t know if there is a way around it as the code currently
stands. Your proposal here:

seems like one way out.  Then the code that creates the

CallbackRegistry in Axes.cla() could iterate through all the axes
and create a callback type for each of them.
However, to step back from this, I’ve never quite understood why it
was necessary to have a fixed set of callback types in the
CallbackRegistry to begin with. It seems to me it comes from a
history of callback registry classes I’ve seen in more static
languages (such as libsigc++). We could just as easily create new
signal types on the fly as they are registered. You lose some error
checking, I suppose, if the caller and receiver don’t agree on the
names, but seems like a small price to pay.
Mike

···

https://github.com/matplotlib/matplotlib/issues/379

-- Michael Droettboom
Science Software Branch
Space Telescope Science Institute
Baltimore, Maryland, USA

CallbackRegistry.signals is a "public" attribute, so is there anything
preventing you Ben from just doing

  self.callbacks.signals.add('zlim_changed')

and then connecting your desired callback?

JDH

···

On Thu, Jun 23, 2011 at 3:11 PM, Michael Droettboom <mdroe@...31...> wrote:

On 06/23/2011 03:49 PM, Benjamin Root wrote:

Hello all,

I am working to make mplot3d feature-parity with regular axes objects. I
have come across a possible design flaw with the CallbackRegistry.

Many of the Axes3D methods are merely wrappers around Axes methods letting
it do the work for x and y axis and then let Axes3D do the same for the z
axis. In Axes.cla(), self.callbacks gets a CallbackRegistry of signals
named "xlim_changed" and "ylim_changed". However, once that is set, it does
not appear to be a way for me to add another signal of "zlim_changed" in
Axes3D.cla(). I could replace self.callbacks with a new CallbackRegistry,
but since there is a lot of interveaning code between that first
initialization of self.callbacks and coming back out of Axes.cla(), I worry
that some callbacks may have been registered by then and would get
eliminated in the process.

Is there a recommended way around this? Or maybe this is more evidence that
we should move towards the use of a dictionary of axis objects and make Axes
functions more agnostic to the number of axis?

I don't know if there is a way around it as the code currently stands. Your
proposal here:

https://github.com/matplotlib/matplotlib/issues/379

seems like one way out. Then the code that creates the CallbackRegistry in
Axes.cla() could iterate through all the axes and create a callback type for
each of them.

However, to step back from this, I've never quite understood why it was
necessary to have a fixed set of callback types in the CallbackRegistry to
begin with. It seems to me it comes from a history of callback registry
classes I've seen in more static languages (such as libsigc++). We could
just as easily create new signal types on the fly as they are registered.
You lose some error checking, I suppose, if the caller and receiver don't
agree on the names, but seems like a small price to pay.

Hello all,

I am working to make mplot3d feature-parity with regular axes objects. I
have come across a possible design flaw with the CallbackRegistry.

Many of the Axes3D methods are merely wrappers around Axes methods letting
it do the work for x and y axis and then let Axes3D do the same for the z
axis. In Axes.cla(), self.callbacks gets a CallbackRegistry of signals
named "xlim_changed" and "ylim_changed". However, once that is set, it does
not appear to be a way for me to add another signal of "zlim_changed" in
Axes3D.cla(). I could replace self.callbacks with a new CallbackRegistry,
but since there is a lot of interveaning code between that first
initialization of self.callbacks and coming back out of Axes.cla(), I worry
that some callbacks may have been registered by then and would get
eliminated in the process.

Is there a recommended way around this? Or maybe this is more evidence that
we should move towards the use of a dictionary of axis objects and make Axes
functions more agnostic to the number of axis?

I don't know if there is a way around it as the code currently stands. Your
proposal here:

https://github.com/matplotlib/matplotlib/issues/379

seems like one way out. Then the code that creates the CallbackRegistry in
Axes.cla() could iterate through all the axes and create a callback type for
each of them.

However, to step back from this, I've never quite understood why it was
necessary to have a fixed set of callback types in the CallbackRegistry to
begin with. It seems to me it comes from a history of callback registry
classes I've seen in more static languages (such as libsigc++). We could
just as easily create new signal types on the fly as they are registered.
You lose some error checking, I suppose, if the caller and receiver don't
agree on the names, but seems like a small price to pay.

CallbackRegistry.signals is a "public" attribute, so is there anything
preventing you Ben from just doing

   self.callbacks.signals.add('zlim_changed')

and then connecting your desired callback?

Yes, it requires some modification to the CallbackRegistry:

     def __init__(self, signals):
         '*signals* is a sequence of valid signals'
         self.signals = set(signals)
         self.callbacks = dict([(s, dict()) for s in signals])
         self._cid = 0

So adding to the set of signals is not enough. It would be easy to made an add_signal() method to take care of the two necessary steps. It would seem simpler, however, to just let the signals be added automatically as needed, as I believe Mike is suggesting.

Actually, it looks like self.signals could be replaced by a property that, upon reading, would return self.callbacks.keys(). Why use two data structures if one will do? Of course, since signals is public, that would represent API breakage--but of a rather obscure sort.

(Shooting from the hip; I haven't looked closely.)

Eric

···

On 06/23/2011 10:53 AM, John Hunter wrote:

On Thu, Jun 23, 2011 at 3:11 PM, Michael Droettboom<mdroe@...31...> wrote:

On 06/23/2011 03:49 PM, Benjamin Root wrote:

JDH

------------------------------------------------------------------------------
Simplify data backup and recovery for your virtual environment with vRanger.
Installation's a snap, and flexible recovery options mean your data is safe,
secure and there when you need it. Data protection magic?
Nope - It's vRanger. Get your free trial download today.
http://p.sf.net/sfu/quest-sfdev2dev
_______________________________________________
Matplotlib-devel mailing list
Matplotlib-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/matplotlib-devel

I am willing to go with whatever makes everyone happy. I have a limited amount of time, and my goal with the feature-parity branch (found here: https://github.com/WeatherGod/matplotlib/tree/mplot3d/consistency ) is to get a z-version of every single axis-related function into Axes3D, and make sure that most other functions that operate on arbitrary axis are capable of acting on the z-axis.

However, I do not intend to make sure that everything works (only that there are no regressions). For example, setting axis label properties (‘right’, ‘left’, etc.) make little sense in the current context, and working with minor ticks do nothing in mplot3d. The idea is that if the added functions work, then that is good news. If the added functions do not work, then that can be a bug report.

Therefore, anything that gets me to that goal would be fine. Heck, doing nothing about this might also work because there never was a callback for zlim_changed before, so not having it now will be no loss of function.

Sorry for rambling,
Ben Root

···

On Thu, Jun 23, 2011 at 4:29 PM, Eric Firing <efiring@…229…> wrote:

On 06/23/2011 10:53 AM, John Hunter wrote:

On Thu, Jun 23, 2011 at 3:11 PM, Michael Droettboom<mdroe@…31…> wrote:

On 06/23/2011 03:49 PM, Benjamin Root wrote:

Hello all,

I am working to make mplot3d feature-parity with regular axes objects. I

have come across a possible design flaw with the CallbackRegistry.

Many of the Axes3D methods are merely wrappers around Axes methods letting

it do the work for x and y axis and then let Axes3D do the same for the z

axis. In Axes.cla(), self.callbacks gets a CallbackRegistry of signals

named “xlim_changed” and “ylim_changed”. However, once that is set, it does

not appear to be a way for me to add another signal of “zlim_changed” in

Axes3D.cla(). I could replace self.callbacks with a new CallbackRegistry,

but since there is a lot of interveaning code between that first

initialization of self.callbacks and coming back out of Axes.cla(), I worry

that some callbacks may have been registered by then and would get

eliminated in the process.

Is there a recommended way around this? Or maybe this is more evidence that

we should move towards the use of a dictionary of axis objects and make Axes

functions more agnostic to the number of axis?

I don’t know if there is a way around it as the code currently stands. Your

proposal here:

https://github.com/matplotlib/matplotlib/issues/379

seems like one way out. Then the code that creates the CallbackRegistry in

Axes.cla() could iterate through all the axes and create a callback type for

each of them.

However, to step back from this, I’ve never quite understood why it was

necessary to have a fixed set of callback types in the CallbackRegistry to

begin with. It seems to me it comes from a history of callback registry

classes I’ve seen in more static languages (such as libsigc++). We could

just as easily create new signal types on the fly as they are registered.

You lose some error checking, I suppose, if the caller and receiver don’t

agree on the names, but seems like a small price to pay.

CallbackRegistry.signals is a “public” attribute, so is there anything

preventing you Ben from just doing

self.callbacks.signals.add(‘zlim_changed’)

and then connecting your desired callback?

Yes, it requires some modification to the CallbackRegistry:

 def __init__(self, signals):

     '*signals* is a sequence of valid signals'

     self.signals = set(signals)

     self.callbacks = dict([(s, dict()) for s in signals])

     self._cid = 0

So adding to the set of signals is not enough. It would be easy to made

an add_signal() method to take care of the two necessary steps. It

would seem simpler, however, to just let the signals be added

automatically as needed, as I believe Mike is suggesting.

Actually, it looks like self.signals could be replaced by a property

that, upon reading, would return self.callbacks.keys(). Why use two

data structures if one will do? Of course, since signals is public,

that would represent API breakage–but of a rather obscure sort.

(Shooting from the hip; I haven’t looked closely.)

Eric

I put up a pull request for adding the method “add_signal” to the CallbackRegistry here: https://github.com/matplotlib/matplotlib/pull/381

Ben Root

···

On Thu, Jun 23, 2011 at 6:55 PM, Benjamin Root <ben.root@…553…> wrote:

On Thu, Jun 23, 2011 at 4:29 PM, Eric Firing <efiring@…229…> wrote:

On 06/23/2011 10:53 AM, John Hunter wrote:

On Thu, Jun 23, 2011 at 3:11 PM, Michael Droettboom<mdroe@…31…> wrote:

On 06/23/2011 03:49 PM, Benjamin Root wrote:

Hello all,

I am working to make mplot3d feature-parity with regular axes objects. I

have come across a possible design flaw with the CallbackRegistry.

Many of the Axes3D methods are merely wrappers around Axes methods letting

it do the work for x and y axis and then let Axes3D do the same for the z

axis. In Axes.cla(), self.callbacks gets a CallbackRegistry of signals

named “xlim_changed” and “ylim_changed”. However, once that is set, it does

not appear to be a way for me to add another signal of “zlim_changed” in

Axes3D.cla(). I could replace self.callbacks with a new CallbackRegistry,

but since there is a lot of interveaning code between that first

initialization of self.callbacks and coming back out of Axes.cla(), I worry

that some callbacks may have been registered by then and would get

eliminated in the process.

Is there a recommended way around this? Or maybe this is more evidence that

we should move towards the use of a dictionary of axis objects and make Axes

functions more agnostic to the number of axis?

I don’t know if there is a way around it as the code currently stands. Your

proposal here:

https://github.com/matplotlib/matplotlib/issues/379

seems like one way out. Then the code that creates the CallbackRegistry in

Axes.cla() could iterate through all the axes and create a callback type for

each of them.

However, to step back from this, I’ve never quite understood why it was

necessary to have a fixed set of callback types in the CallbackRegistry to

begin with. It seems to me it comes from a history of callback registry

classes I’ve seen in more static languages (such as libsigc++). We could

just as easily create new signal types on the fly as they are registered.

You lose some error checking, I suppose, if the caller and receiver don’t

agree on the names, but seems like a small price to pay.

CallbackRegistry.signals is a “public” attribute, so is there anything

preventing you Ben from just doing

self.callbacks.signals.add(‘zlim_changed’)

and then connecting your desired callback?

Yes, it requires some modification to the CallbackRegistry:

 def __init__(self, signals):

     '*signals* is a sequence of valid signals'

     self.signals = set(signals)

     self.callbacks = dict([(s, dict()) for s in signals])

     self._cid = 0

So adding to the set of signals is not enough. It would be easy to made

an add_signal() method to take care of the two necessary steps. It

would seem simpler, however, to just let the signals be added

automatically as needed, as I believe Mike is suggesting.

Actually, it looks like self.signals could be replaced by a property

that, upon reading, would return self.callbacks.keys(). Why use two

data structures if one will do? Of course, since signals is public,

that would represent API breakage–but of a rather obscure sort.

(Shooting from the hip; I haven’t looked closely.)

Eric

I am willing to go with whatever makes everyone happy. I have a limited amount of time, and my goal with the feature-parity branch (found here: https://github.com/WeatherGod/matplotlib/tree/mplot3d/consistency ) is to get a z-version of every single axis-related function into Axes3D, and make sure that most other functions that operate on arbitrary axis are capable of acting on the z-axis.

However, I do not intend to make sure that everything works (only that there are no regressions). For example, setting axis label properties (‘right’, ‘left’, etc.) make little sense in the current context, and working with minor ticks do nothing in mplot3d. The idea is that if the added functions work, then that is good news. If the added functions do not work, then that can be a bug report.

Therefore, anything that gets me to that goal would be fine. Heck, doing nothing about this might also work because there never was a callback for zlim_changed before, so not having it now will be no loss of function.

Sorry for rambling,
Ben Root

I put up an alternative request to just do away with the fixed set
of signals altogether. I personally think this is more Pythonic,
but I could be persuaded otherwise if there’s a good reason to
maintain the list of acceptable callbacks.

Mike
···
-- Michael Droettboom
Science Software Branch
Space Telescope Science Institute
Baltimore, Maryland, USA