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
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. Mike -- Michael Droettboom Science Software Branch Space Telescope Science Institute Baltimore, Maryland, USA
On Thu, Jun 23, 2011 at 3:11 PM, Michael Droettboom <md...@st...> 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? JDH
On 06/23/2011 10:53 AM, John Hunter wrote: > On Thu, Jun 23, 2011 at 3:11 PM, Michael Droettboom<md...@st...> 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 > > 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 > Mat...@li... > https://lists.sourceforge.net/lists/listinfo/matplotlib-devel
On Thu, Jun 23, 2011 at 4:29 PM, Eric Firing <ef...@ha...> wrote: > On 06/23/2011 10:53 AM, John Hunter wrote: > > On Thu, Jun 23, 2011 at 3:11 PM, Michael Droettboom<md...@st...> > 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
On Thu, Jun 23, 2011 at 6:55 PM, Benjamin Root <ben...@ou...> wrote: > On Thu, Jun 23, 2011 at 4:29 PM, Eric Firing <ef...@ha...> wrote: > >> On 06/23/2011 10:53 AM, John Hunter wrote: >> > On Thu, Jun 23, 2011 at 3:11 PM, Michael Droettboom<md...@st...> >> 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 a pull request for adding the method "add_signal" to the CallbackRegistry here: https://github.com/matplotlib/matplotlib/pull/381 Ben Root
On 06/24/2011 10:24 AM, Benjamin Root wrote: > > > > 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 > 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