RE: Move to JIRA [was: Re: [8u] API Request: RT-25613, ObservableValue should have a hasListener(listener) method]
Hi Anthony, Thanks for sending a request for applying the multi-threaded plug-in. As for the email plug-in mentioned below, using it does not mean sending all JIRA's automatically to a single mailing list.. On the contrary it can be used as a selective tool for sharing selective JIRAs to selective people/groups. This might be handy in case an issue was previously discussed in JIRA and at a point you want to share with specific recipients. Anyway if the current watch list mechanism functionality suffices that's it's fine with me.. :) Regards, Seeon -Original Message- From: Anthony Petrov Sent: Friday, January 24, 2014 1:07 PM To: Seeon Birger; Steve Northover Cc: John Hendrikx; openjfx-dev@openjdk.java.net Subject: Re: Move to JIRA [was: Re: [8u] API Request: RT-25613, ObservableValue should have a hasListener(listener) method] Hi Seeon, I don't think that posting updates from all our JIRAs to a single mailing list is a good idea because the messages traffic would be huge. Each JIRA provides a watch list, so people can follow specific bugs that they're interested in. I believe that this functionality is sufficient for all reasonable purposes. However, your suggestion about using threaded comments sounds interesting. We've sent a request to our Infrastructure team to evaluate the plugin and consider adding it to our JIRA instance. Note that this is not going to happen overnight and could take some time, but the request is filed at least. Thanks for the suggestion! -- best regards, Anthony On 1/23/2014 9:09 PM, Seeon Birger wrote: > Steve, > > I wonder if we could take advantage of available plug-ins for JIRA. > > I particular I found this one which enables threaded comments for JIRA: > https://marketplace.atlassian.com/plugins/com.atlassian.jira.threadedc > omments.threaded-comments > > Also interesting is the following which make it easy to put JIRA updates on > mailing lists in a flexible and customizable way: > https://marketplace.atlassian.com/plugins/com.metainf.jira.plugin.emai > lissue > > What do you think? > > Seeon > > > > -Original Message- > From: Stephen F Northover > Sent: Thursday, January 23, 2014 12:45 AM > To: John Hendrikx; openjfx-dev@openjdk.java.net > Subject: Re: Move to JIRA [was: Re: [8u] API Request: RT-25613, > ObservableValue should have a hasListener(listener) method] > > Hi John, > > The goal is not to end the discussion! > > It's a trade off. Mailing lists are good because they provide a threaded > discussion. JIRA is bad because it is not threaded. JIRA has the advantage > that it captures data in a single place and provides a good history of why a > decision was made. > > There's no right answer here but the policy that the FX committers is using > is to try to capture as much as possible in JIRA. > > Steve > > On 2014-01-22 5:29 PM, John Hendrikx wrote: >> Unfortunately, "discussing" things in JIRA works very poorly and is a >> good way to end a productive discussion IMHO. Mailinglists are much >> better suited to the task, as thousands of interesting mailinglists >> accross many developer communities will atest to. >> >> Keeping a record is good, aren't these mailinglists archived? >> >> --John >> >> On 22/01/2014 18:47, Daniel Blaukopf wrote: >>> Hi Martin, Randahl, Tom, Richard, Tomas and Ali, >>> >>> This is a productive discussion, but once we get to this level of >>> detail JIRA is the place to have it, so that we don't lose our >>> record of it. Would you continue the discussion on >>> https://javafx-jira.kenai.com/browse/RT-25613 ? >>> >>> See >>> https://wiki.openjdk.java.net/display/OpenJFX/Code+Reviews#CodeRevie >>> w s-TechnicalDiscussionsandCodeReviews >>> >>> Thanks, >>> Daniel >>> >>> On Jan 22, 2014, at 7:23 PM, Stephen F >>> Northover wrote: >>> >>>> If we add this API, I like addListener(InvalidationListener, >>>> boolean) better than ensureListener(). >>>> >>>> Steve >>>> >>>> On 2014-01-22 8:20 AM, Ali Ebrahimi wrote: >>>>> I suggest adding another overload for addListener method taking >>>>> boolean parameter "duplicateAllowed" or "duplicateNotAllowed". >>>>> >>>>> >>>>> On Wed, Jan 22, 2014 at 3:00 PM, Richard >>>>> Bairwrote: >>>>> >>>>>>>> The default implementation (for Observable) would look like this: >>>>>>>> >>>>>>>> public def
Re: Move to JIRA [was: Re: [8u] API Request: RT-25613, ObservableValue should have a hasListener(listener) method]
Hi Seeon, I don't think that posting updates from all our JIRAs to a single mailing list is a good idea because the messages traffic would be huge. Each JIRA provides a watch list, so people can follow specific bugs that they're interested in. I believe that this functionality is sufficient for all reasonable purposes. However, your suggestion about using threaded comments sounds interesting. We've sent a request to our Infrastructure team to evaluate the plugin and consider adding it to our JIRA instance. Note that this is not going to happen overnight and could take some time, but the request is filed at least. Thanks for the suggestion! -- best regards, Anthony On 1/23/2014 9:09 PM, Seeon Birger wrote: Steve, I wonder if we could take advantage of available plug-ins for JIRA. I particular I found this one which enables threaded comments for JIRA: https://marketplace.atlassian.com/plugins/com.atlassian.jira.threadedcomments.threaded-comments Also interesting is the following which make it easy to put JIRA updates on mailing lists in a flexible and customizable way: https://marketplace.atlassian.com/plugins/com.metainf.jira.plugin.emailissue What do you think? Seeon -Original Message- From: Stephen F Northover Sent: Thursday, January 23, 2014 12:45 AM To: John Hendrikx; openjfx-dev@openjdk.java.net Subject: Re: Move to JIRA [was: Re: [8u] API Request: RT-25613, ObservableValue should have a hasListener(listener) method] Hi John, The goal is not to end the discussion! It's a trade off. Mailing lists are good because they provide a threaded discussion. JIRA is bad because it is not threaded. JIRA has the advantage that it captures data in a single place and provides a good history of why a decision was made. There's no right answer here but the policy that the FX committers is using is to try to capture as much as possible in JIRA. Steve On 2014-01-22 5:29 PM, John Hendrikx wrote: Unfortunately, "discussing" things in JIRA works very poorly and is a good way to end a productive discussion IMHO. Mailinglists are much better suited to the task, as thousands of interesting mailinglists accross many developer communities will atest to. Keeping a record is good, aren't these mailinglists archived? --John On 22/01/2014 18:47, Daniel Blaukopf wrote: Hi Martin, Randahl, Tom, Richard, Tomas and Ali, This is a productive discussion, but once we get to this level of detail JIRA is the place to have it, so that we don't lose our record of it. Would you continue the discussion on https://javafx-jira.kenai.com/browse/RT-25613 ? See https://wiki.openjdk.java.net/display/OpenJFX/Code+Reviews#CodeReview s-TechnicalDiscussionsandCodeReviews Thanks, Daniel On Jan 22, 2014, at 7:23 PM, Stephen F Northover wrote: If we add this API, I like addListener(InvalidationListener, boolean) better than ensureListener(). Steve On 2014-01-22 8:20 AM, Ali Ebrahimi wrote: I suggest adding another overload for addListener method taking boolean parameter "duplicateAllowed" or "duplicateNotAllowed". On Wed, Jan 22, 2014 at 3:00 PM, Richard Bairwrote: The default implementation (for Observable) would look like this: public default void ensureListener(InvalidationListener listener) { removeListener(listener); addListener(listener); } subclasses might do something more effective. The same would apply to ObservableValue and ChangeListener and Observable[List|Set|Map] and [List|Set|Map]ChangeListener. Well this would destroy the order! I expect listeners to be called in the correct order not? That's a good point :-( Why doing a remove and not simply check if the listener has already been added? Because there is no way to check, except in the implementation. From the Observable interface level, there is no way to a) force all implementations of the interface to implement the method correctly (without breaking source compatibility anyway), or b) to provide a reasonable default implementation. Maybe this is one of those things we can't fix on the Observable interface and just have to provide implementations of on our concrete properties. Richard
RE: Move to JIRA [was: Re: [8u] API Request: RT-25613, ObservableValue should have a hasListener(listener) method]
Steve, I wonder if we could take advantage of available plug-ins for JIRA. I particular I found this one which enables threaded comments for JIRA: https://marketplace.atlassian.com/plugins/com.atlassian.jira.threadedcomments.threaded-comments Also interesting is the following which make it easy to put JIRA updates on mailing lists in a flexible and customizable way: https://marketplace.atlassian.com/plugins/com.metainf.jira.plugin.emailissue What do you think? Seeon -Original Message- From: Stephen F Northover Sent: Thursday, January 23, 2014 12:45 AM To: John Hendrikx; openjfx-dev@openjdk.java.net Subject: Re: Move to JIRA [was: Re: [8u] API Request: RT-25613, ObservableValue should have a hasListener(listener) method] Hi John, The goal is not to end the discussion! It's a trade off. Mailing lists are good because they provide a threaded discussion. JIRA is bad because it is not threaded. JIRA has the advantage that it captures data in a single place and provides a good history of why a decision was made. There's no right answer here but the policy that the FX committers is using is to try to capture as much as possible in JIRA. Steve On 2014-01-22 5:29 PM, John Hendrikx wrote: > Unfortunately, "discussing" things in JIRA works very poorly and is a > good way to end a productive discussion IMHO. Mailinglists are much > better suited to the task, as thousands of interesting mailinglists > accross many developer communities will atest to. > > Keeping a record is good, aren't these mailinglists archived? > > --John > > On 22/01/2014 18:47, Daniel Blaukopf wrote: >> Hi Martin, Randahl, Tom, Richard, Tomas and Ali, >> >> This is a productive discussion, but once we get to this level of >> detail JIRA is the place to have it, so that we don't lose our record >> of it. Would you continue the discussion on >> https://javafx-jira.kenai.com/browse/RT-25613 ? >> >> See >> https://wiki.openjdk.java.net/display/OpenJFX/Code+Reviews#CodeReview >> s-TechnicalDiscussionsandCodeReviews >> >> Thanks, >> Daniel >> >> On Jan 22, 2014, at 7:23 PM, Stephen F >> Northover wrote: >> >>> If we add this API, I like addListener(InvalidationListener, >>> boolean) better than ensureListener(). >>> >>> Steve >>> >>> On 2014-01-22 8:20 AM, Ali Ebrahimi wrote: >>>> I suggest adding another overload for addListener method taking >>>> boolean parameter "duplicateAllowed" or "duplicateNotAllowed". >>>> >>>> >>>> On Wed, Jan 22, 2014 at 3:00 PM, Richard >>>> Bairwrote: >>>> >>>>>>> The default implementation (for Observable) would look like this: >>>>>>> >>>>>>> public default void ensureListener(InvalidationListener listener) { >>>>>>> removeListener(listener); >>>>>>> addListener(listener); >>>>>>> } >>>>>>> >>>>>>> subclasses might do something more effective. The same would >>>>>>> apply to >>>>>>> ObservableValue and ChangeListener and Observable[List|Set|Map] and >>>>>>> [List|Set|Map]ChangeListener. >>>>>> Well this would destroy the order! I expect listeners to be >>>>>> called in >>>>>> the correct order not? >>>>> That's a good point :-( >>>>> >>>>>> Why doing a remove and not simply check if the >>>>>> listener has already been added? >>>>> Because there is no way to check, except in the implementation. >>>>> From the >>>>> Observable interface level, there is no way to a) force all >>>>> implementations >>>>> of the interface to implement the method correctly (without >>>>> breaking source >>>>> compatibility anyway), or b) to provide a reasonable default >>>>> implementation. >>>>> >>>>> Maybe this is one of those things we can't fix on the Observable >>>>> interface >>>>> and just have to provide implementations of on our concrete >>>>> properties. >>>>> >>>>> Richard >
Re: Move to JIRA [was: Re: [8u] API Request: RT-25613, ObservableValue should have a hasListener(listener) method]
Hi John, The goal is not to end the discussion! It's a trade off. Mailing lists are good because they provide a threaded discussion. JIRA is bad because it is not threaded. JIRA has the advantage that it captures data in a single place and provides a good history of why a decision was made. There's no right answer here but the policy that the FX committers is using is to try to capture as much as possible in JIRA. Steve On 2014-01-22 5:29 PM, John Hendrikx wrote: Unfortunately, "discussing" things in JIRA works very poorly and is a good way to end a productive discussion IMHO. Mailinglists are much better suited to the task, as thousands of interesting mailinglists accross many developer communities will atest to. Keeping a record is good, aren't these mailinglists archived? --John On 22/01/2014 18:47, Daniel Blaukopf wrote: Hi Martin, Randahl, Tom, Richard, Tomas and Ali, This is a productive discussion, but once we get to this level of detail JIRA is the place to have it, so that we don’t lose our record of it. Would you continue the discussion on https://javafx-jira.kenai.com/browse/RT-25613 ? See https://wiki.openjdk.java.net/display/OpenJFX/Code+Reviews#CodeReviews-TechnicalDiscussionsandCodeReviews Thanks, Daniel On Jan 22, 2014, at 7:23 PM, Stephen F Northover wrote: If we add this API, I like addListener(InvalidationListener, boolean) better than ensureListener(). Steve On 2014-01-22 8:20 AM, Ali Ebrahimi wrote: I suggest adding another overload for addListener method taking boolean parameter "duplicateAllowed" or "duplicateNotAllowed". On Wed, Jan 22, 2014 at 3:00 PM, Richard Bairwrote: The default implementation (for Observable) would look like this: public default void ensureListener(InvalidationListener listener) { removeListener(listener); addListener(listener); } subclasses might do something more effective. The same would apply to ObservableValue and ChangeListener and Observable[List|Set|Map] and [List|Set|Map]ChangeListener. Well this would destroy the order! I expect listeners to be called in the correct order not? That’s a good point :-( Why doing a remove and not simply check if the listener has already been added? Because there is no way to check, except in the implementation. From the Observable interface level, there is no way to a) force all implementations of the interface to implement the method correctly (without breaking source compatibility anyway), or b) to provide a reasonable default implementation. Maybe this is one of those things we can’t fix on the Observable interface and just have to provide implementations of on our concrete properties. Richard
Re: Move to JIRA [was: Re: [8u] API Request: RT-25613, ObservableValue should have a hasListener(listener) method]
The point is that we'd rather have quick reference to the historical discussion without having to cross-reference between jira and the mailing list archives (which isn't automatically done). Once you add yourself as a watcher to the jira issue you are notified of every change, in much the same way as you are emailed of the discussion on this mailing list. A good case in point came up just today with Martin referencing historical choices made with the Observable API, but not being able to reference them as they are not recorded anywhere (and if they are, being able to search for them is not trivial - the only hope is to find them in your email client and then go off to the openjfx-dev archives and manually search through the discussions until the right thread is found). Whilst Jira search isn't great, it is better than this! In other words, I can see an upside without any downside. What troubles do you have with Jira specifically? -- Jonathan On 23/01/2014 11:29 a.m., John Hendrikx wrote: > Unfortunately, "discussing" things in JIRA works very poorly and is a > good way to end a productive discussion IMHO. Mailinglists are much > better suited to the task, as thousands of interesting mailinglists > accross many developer communities will atest to. > > Keeping a record is good, aren't these mailinglists archived? > > --John > > On 22/01/2014 18:47, Daniel Blaukopf wrote: >> Hi Martin, Randahl, Tom, Richard, Tomas and Ali, >> >> This is a productive discussion, but once we get to this level of >> detail JIRA is the place to have it, so that we don’t lose our record >> of it. Would you continue the discussion on >> https://javafx-jira.kenai.com/browse/RT-25613 ? >> >> See >> https://wiki.openjdk.java.net/display/OpenJFX/Code+Reviews#CodeReviews-TechnicalDiscussionsandCodeReviews >> >> Thanks, >> Daniel >> >> On Jan 22, 2014, at 7:23 PM, Stephen F >> Northover wrote: >> >>> If we add this API, I like addListener(InvalidationListener, >>> boolean) better than ensureListener(). >>> >>> Steve >>> >>> On 2014-01-22 8:20 AM, Ali Ebrahimi wrote: I suggest adding another overload for addListener method taking boolean parameter "duplicateAllowed" or "duplicateNotAllowed". On Wed, Jan 22, 2014 at 3:00 PM, Richard Bairwrote: >>> The default implementation (for Observable) would look like this: >>> >>> public default void ensureListener(InvalidationListener listener) { >>> removeListener(listener); >>> addListener(listener); >>> } >>> >>> subclasses might do something more effective. The same would >>> apply to >>> ObservableValue and ChangeListener and Observable[List|Set|Map] and >>> [List|Set|Map]ChangeListener. >> Well this would destroy the order! I expect listeners to be >> called in >> the correct order not? > That’s a good point :-( > >> Why doing a remove and not simply check if the >> listener has already been added? > Because there is no way to check, except in the implementation. > From the > Observable interface level, there is no way to a) force all > implementations > of the interface to implement the method correctly (without > breaking source > compatibility anyway), or b) to provide a reasonable default > implementation. > > Maybe this is one of those things we can’t fix on the Observable > interface > and just have to provide implementations of on our concrete > properties. > > Richard >
Re: Move to JIRA [was: Re: [8u] API Request: RT-25613, ObservableValue should have a hasListener(listener) method]
Unfortunately, "discussing" things in JIRA works very poorly and is a good way to end a productive discussion IMHO. Mailinglists are much better suited to the task, as thousands of interesting mailinglists accross many developer communities will atest to. Keeping a record is good, aren't these mailinglists archived? --John On 22/01/2014 18:47, Daniel Blaukopf wrote: Hi Martin, Randahl, Tom, Richard, Tomas and Ali, This is a productive discussion, but once we get to this level of detail JIRA is the place to have it, so that we don’t lose our record of it. Would you continue the discussion on https://javafx-jira.kenai.com/browse/RT-25613 ? See https://wiki.openjdk.java.net/display/OpenJFX/Code+Reviews#CodeReviews-TechnicalDiscussionsandCodeReviews Thanks, Daniel On Jan 22, 2014, at 7:23 PM, Stephen F Northover wrote: If we add this API, I like addListener(InvalidationListener, boolean) better than ensureListener(). Steve On 2014-01-22 8:20 AM, Ali Ebrahimi wrote: I suggest adding another overload for addListener method taking boolean parameter "duplicateAllowed" or "duplicateNotAllowed". On Wed, Jan 22, 2014 at 3:00 PM, Richard Bairwrote: The default implementation (for Observable) would look like this: public default void ensureListener(InvalidationListener listener) { removeListener(listener); addListener(listener); } subclasses might do something more effective. The same would apply to ObservableValue and ChangeListener and Observable[List|Set|Map] and [List|Set|Map]ChangeListener. Well this would destroy the order! I expect listeners to be called in the correct order not? That’s a good point :-( Why doing a remove and not simply check if the listener has already been added? Because there is no way to check, except in the implementation. From the Observable interface level, there is no way to a) force all implementations of the interface to implement the method correctly (without breaking source compatibility anyway), or b) to provide a reasonable default implementation. Maybe this is one of those things we can’t fix on the Observable interface and just have to provide implementations of on our concrete properties. Richard
Move to JIRA [was: Re: [8u] API Request: RT-25613, ObservableValue should have a hasListener(listener) method]
Hi Martin, Randahl, Tom, Richard, Tomas and Ali, This is a productive discussion, but once we get to this level of detail JIRA is the place to have it, so that we don’t lose our record of it. Would you continue the discussion on https://javafx-jira.kenai.com/browse/RT-25613 ? See https://wiki.openjdk.java.net/display/OpenJFX/Code+Reviews#CodeReviews-TechnicalDiscussionsandCodeReviews Thanks, Daniel On Jan 22, 2014, at 7:23 PM, Stephen F Northover wrote: > If we add this API, I like addListener(InvalidationListener, boolean) better > than ensureListener(). > > Steve > > On 2014-01-22 8:20 AM, Ali Ebrahimi wrote: >> I suggest adding another overload for addListener method taking boolean >> parameter "duplicateAllowed" or "duplicateNotAllowed". >> >> >> On Wed, Jan 22, 2014 at 3:00 PM, Richard Bair wrote: >> > The default implementation (for Observable) would look like this: > > public default void ensureListener(InvalidationListener listener) { >removeListener(listener); >addListener(listener); > } > > subclasses might do something more effective. The same would apply to > ObservableValue and ChangeListener and Observable[List|Set|Map] and > [List|Set|Map]ChangeListener. Well this would destroy the order! I expect listeners to be called in the correct order not? >>> That’s a good point :-( >>> Why doing a remove and not simply check if the listener has already been added? >>> Because there is no way to check, except in the implementation. From the >>> Observable interface level, there is no way to a) force all implementations >>> of the interface to implement the method correctly (without breaking source >>> compatibility anyway), or b) to provide a reasonable default implementation. >>> >>> Maybe this is one of those things we can’t fix on the Observable interface >>> and just have to provide implementations of on our concrete properties. >>> >>> Richard >
Re: [8u] API Request: RT-25613, ObservableValue should have a hasListener(listener) method
If we add this API, I like addListener(InvalidationListener, boolean) better than ensureListener(). Steve On 2014-01-22 8:20 AM, Ali Ebrahimi wrote: I suggest adding another overload for addListener method taking boolean parameter "duplicateAllowed" or "duplicateNotAllowed". On Wed, Jan 22, 2014 at 3:00 PM, Richard Bair wrote: The default implementation (for Observable) would look like this: public default void ensureListener(InvalidationListener listener) { removeListener(listener); addListener(listener); } subclasses might do something more effective. The same would apply to ObservableValue and ChangeListener and Observable[List|Set|Map] and [List|Set|Map]ChangeListener. Well this would destroy the order! I expect listeners to be called in the correct order not? That’s a good point :-( Why doing a remove and not simply check if the listener has already been added? Because there is no way to check, except in the implementation. From the Observable interface level, there is no way to a) force all implementations of the interface to implement the method correctly (without breaking source compatibility anyway), or b) to provide a reasonable default implementation. Maybe this is one of those things we can’t fix on the Observable interface and just have to provide implementations of on our concrete properties. Richard
Re: [8u] API Request: RT-25613, ObservableValue should have a hasListener(listener) method
Well it is just an fictive example. But for instance, there could be game rules that required us to listen to the energyPoints of all spaceships that had ever been part of the group. Perhaps the group would dissolve if more than 10 spaceships that had ever been part of the group where shot down. There could be many reasons why, and I certainly don’t think JavaFX should keep us from avoiding duplicate listeners. Yours Randahl On 22 Jan 2014, at 14:29, Tomas Mikula wrote: > Thanks for an example. I imagined something analogous to your example, but > the part I'm curious about is this > > "We still wanted to listen to a property of the spaceship, but now it was no > longer part of the group." > > Why would you want to keep listening to that spaceship after removal from the > group? > > Regards, > Tomas > > > On Wed, Jan 22, 2014 at 2:11 PM, Randahl Fink Isaksen > wrote: > Hi Tomas > > About the example you request: Say you and I implemented a computer game with > spaceships on screen, and lets imagine that these spaceships attacked in > groups. Everytime a spaceship was added to a group, we wanted to listen to a > property of that spaceship from that point in time and until eternity. Then, > imagine we took out a spaceship from the group. We still wanted to listen to > a property of the spaceship, but now it was no longer part of the group. > Then, imagine we added the same spaceship to the same group once more. At > this point, we don’t know if we have added the spaceship before, so when we > write > > spaceship.someProperty.addListener(ourListener); > > we suddenly have ourListener added twice to the property. > > I know we can write removeListener(ourListener) followed by > addListener(ourListener), and while that would work, I would much rather have > the listener list ensure that there are no duplicates, since I believe that > it is always a bug to register the same listener twice. > > Yours > > Randahl > > > > On 22 Jan 2014, at 13:34, Tomas Mikula wrote: > >> Hi Randahl, >> >> I'm curious about an example where you would take advantage of the behavior >> where multiple addListener(listener) calls add the listener just once. >> >> Anyway, here [1] are helper classes InvalidationSubscriber and >> ChangeSubscriber that allow you to do that: >> >> InvalidationSubscriber subscriber = new InvalidationSubscriber(observable, >> listener); >> >> subscriber.subscribe(); // registers the listener >> subscriber.subscribe(); // no-op >> >> Cheers, >> Tomas >> >> [1] https://gist.github.com/TomasMikula/8557825 >> >> On Wed, Jan 22, 2014 at 11:23 AM, Randahl Fink Isaksen >> wrote: >> Hi Martin >> >> While I agree your proposed solution would work, I still don’t understand >> why JavaFX should keep on supporting duplicates in listener collections. Can >> anyone come up with just 1 example of an application that might be depending >> on having two listeners on the same Observable? E.g. this kind of code: >> >> myObservable.addListener(myChangeListener); //add it >> myObservable.addListener(myChangeListener); //add it again >> >> In what kind of situation would this sort of code make any sense? >> >> If we all feel confident that the presence of duplicates listeners is always >> an error, I warmly recommend changing the API to be duplicate free. >> >> Yours >> >> Randahl >> >> >> >> >> On 22 Jan 2014, at 11:07, Martin Sladecek wrote: >> >> > Hi all, >> > I would like to start discussion about an addition to API in Observable, >> > ObservableValue and all Observable collections. >> > There were multiple requests for a way how to avoid duplicates in >> > listeners lists. The way RT-25613 solves this is that it introduces public >> > boolean hasListener(ListenerType listener) which would return true if the >> > provided listener is already registered. >> > >> > This has one significant drawback that all of Observable* are actually >> > interfaces. Means we can only add hasListener as a defender method. The >> > problem is with the default implementation. We cannot return anything >> > meaningful, so we have to throw an UnsupportedOperationException. The >> > problem is that this might blow up unexpectedly when some "older" >> > Observable implementation is used. Also, it might be easy to miss when >> > implementing the interface, since the IDE might not force you to implement >> > it. >> > >> > So as an alternative solution, I propose adding something like: >> > >> > ensureListener(ListenerType listener) >> > >> > which would make sure the listener is on the list and if a listener is >> > already present, the number of times listener is registered on the >> > Observable will NOT grow after this call. >> > >> > The default implementation (for Observable) would look like this: >> > >> > public default void ensureListener(InvalidationListener listener) { >> >removeListener(listener); >> >addListener(listener); >> > } >> > >> > subclas
Re: [8u] API Request: RT-25613, ObservableValue should have a hasListener(listener) method
Thanks for an example. I imagined something analogous to your example, but the part I'm curious about is this "We still wanted to listen to a property of the spaceship, but now it was no longer part of the group." Why would you want to keep listening to that spaceship after removal from the group? Regards, Tomas On Wed, Jan 22, 2014 at 2:11 PM, Randahl Fink Isaksen wrote: > Hi Tomas > > About the example you request: Say you and I implemented a computer game > with spaceships on screen, and lets imagine that these spaceships attacked > in groups. Everytime a spaceship was added to a group, we wanted to listen > to a property of that spaceship from that point in time and until eternity. > Then, imagine we took out a spaceship from the group. We still wanted to > listen to a property of the spaceship, but now it was no longer part of the > group. Then, imagine we added the same spaceship to the same group once > more. At this point, we don’t know if we have added the spaceship before, > so when we write > > spaceship.someProperty.addListener(ourListener); > > we suddenly have ourListener added twice to the property. > > I know we can write removeListener(ourListener) followed by > addListener(ourListener), and while that would work, I would much rather > have the listener list ensure that there are no duplicates, since I believe > that it is always a bug to register the same listener twice. > > Yours > > Randahl > > > > On 22 Jan 2014, at 13:34, Tomas Mikula wrote: > > Hi Randahl, > > I'm curious about an example where you would take advantage of the > behavior where multiple addListener(listener) calls add the listener just > once. > > Anyway, here [1] are helper classes InvalidationSubscriber and > ChangeSubscriber that allow you to do that: > > InvalidationSubscriber subscriber = new InvalidationSubscriber(observable, > listener); > > subscriber.subscribe(); // registers the listener > subscriber.subscribe(); // no-op > > Cheers, > Tomas > > [1] https://gist.github.com/TomasMikula/8557825 > > On Wed, Jan 22, 2014 at 11:23 AM, Randahl Fink Isaksen > wrote: > >> Hi Martin >> >> While I agree your proposed solution would work, I still don’t understand >> why JavaFX should keep on supporting duplicates in listener collections. >> Can anyone come up with just 1 example of an application that might be >> depending on having two listeners on the same Observable? E.g. this kind of >> code: >> >> myObservable.addListener(myChangeListener); //add it >> myObservable.addListener(myChangeListener); //add it again >> >> In what kind of situation would this sort of code make any sense? >> >> If we all feel confident that the presence of duplicates listeners is >> always an error, I warmly recommend changing the API to be duplicate free. >> >> Yours >> >> Randahl >> >> >> >> >> On 22 Jan 2014, at 11:07, Martin Sladecek >> wrote: >> >> > Hi all, >> > I would like to start discussion about an addition to API in >> Observable, ObservableValue and all Observable collections. >> > There were multiple requests for a way how to avoid duplicates in >> listeners lists. The way RT-25613 solves this is that it introduces public >> boolean hasListener(ListenerType listener) which would return true if the >> provided listener is already registered. >> > >> > This has one significant drawback that all of Observable* are actually >> interfaces. Means we can only add hasListener as a defender method. The >> problem is with the default implementation. We cannot return anything >> meaningful, so we have to throw an UnsupportedOperationException. The >> problem is that this might blow up unexpectedly when some "older" >> Observable implementation is used. Also, it might be easy to miss when >> implementing the interface, since the IDE might not force you to implement >> it. >> > >> > So as an alternative solution, I propose adding something like: >> > >> > ensureListener(ListenerType listener) >> > >> > which would make sure the listener is on the list and if a listener is >> already present, the number of times listener is registered on the >> Observable will NOT grow after this call. >> > >> > The default implementation (for Observable) would look like this: >> > >> > public default void ensureListener(InvalidationListener listener) { >> >removeListener(listener); >> >addListener(listener); >> > } >> > >> > subclasses might do something more effective. The same would apply to >> ObservableValue and ChangeListener and Observable[List|Set|Map] and >> [List|Set|Map]ChangeListener. >> > >> > What do you think? >> > >> > JIRA link: https://javafx-jira.kenai.com/browse/RT-25613 >> > >> > -Martin >> >> > >
Re: [8u] API Request: RT-25613, ObservableValue should have a hasListener(listener) method
There is no official API. You are free to copy-paste the code from the link I posted to your project. On Wed, Jan 22, 2014 at 1:59 PM, Randahl Fink Isaksen wrote: > Hi Tomas > > Great idea! I can see how the InvalidationSubscriber idea would work – but > it is just a concept right? There is no official API for this, right? > > Randahl > > > > > > On 22 Jan 2014, at 13:34, Tomas Mikula wrote: > > Hi Randahl, > > I'm curious about an example where you would take advantage of the > behavior where multiple addListener(listener) calls add the listener just > once. > > Anyway, here [1] are helper classes InvalidationSubscriber and > ChangeSubscriber that allow you to do that: > > InvalidationSubscriber subscriber = new InvalidationSubscriber(observable, > listener); > > subscriber.subscribe(); // registers the listener > subscriber.subscribe(); // no-op > > Cheers, > Tomas > > [1] https://gist.github.com/TomasMikula/8557825 > > On Wed, Jan 22, 2014 at 11:23 AM, Randahl Fink Isaksen > wrote: > >> Hi Martin >> >> While I agree your proposed solution would work, I still don’t understand >> why JavaFX should keep on supporting duplicates in listener collections. >> Can anyone come up with just 1 example of an application that might be >> depending on having two listeners on the same Observable? E.g. this kind of >> code: >> >> myObservable.addListener(myChangeListener); //add it >> myObservable.addListener(myChangeListener); //add it again >> >> In what kind of situation would this sort of code make any sense? >> >> If we all feel confident that the presence of duplicates listeners is >> always an error, I warmly recommend changing the API to be duplicate free. >> >> Yours >> >> Randahl >> >> >> >> >> On 22 Jan 2014, at 11:07, Martin Sladecek >> wrote: >> >> > Hi all, >> > I would like to start discussion about an addition to API in >> Observable, ObservableValue and all Observable collections. >> > There were multiple requests for a way how to avoid duplicates in >> listeners lists. The way RT-25613 solves this is that it introduces public >> boolean hasListener(ListenerType listener) which would return true if the >> provided listener is already registered. >> > >> > This has one significant drawback that all of Observable* are actually >> interfaces. Means we can only add hasListener as a defender method. The >> problem is with the default implementation. We cannot return anything >> meaningful, so we have to throw an UnsupportedOperationException. The >> problem is that this might blow up unexpectedly when some "older" >> Observable implementation is used. Also, it might be easy to miss when >> implementing the interface, since the IDE might not force you to implement >> it. >> > >> > So as an alternative solution, I propose adding something like: >> > >> > ensureListener(ListenerType listener) >> > >> > which would make sure the listener is on the list and if a listener is >> already present, the number of times listener is registered on the >> Observable will NOT grow after this call. >> > >> > The default implementation (for Observable) would look like this: >> > >> > public default void ensureListener(InvalidationListener listener) { >> >removeListener(listener); >> >addListener(listener); >> > } >> > >> > subclasses might do something more effective. The same would apply to >> ObservableValue and ChangeListener and Observable[List|Set|Map] and >> [List|Set|Map]ChangeListener. >> > >> > What do you think? >> > >> > JIRA link: https://javafx-jira.kenai.com/browse/RT-25613 >> > >> > -Martin >> >> > >
Re: [8u] API Request: RT-25613, ObservableValue should have a hasListener(listener) method
I suggest adding another overload for addListener method taking boolean parameter "duplicateAllowed" or "duplicateNotAllowed". On Wed, Jan 22, 2014 at 3:00 PM, Richard Bair wrote: > >> The default implementation (for Observable) would look like this: > >> > >> public default void ensureListener(InvalidationListener listener) { > >>removeListener(listener); > >>addListener(listener); > >> } > >> > >> subclasses might do something more effective. The same would apply to > >> ObservableValue and ChangeListener and Observable[List|Set|Map] and > >> [List|Set|Map]ChangeListener. > > > > Well this would destroy the order! I expect listeners to be called in > > the correct order not? > > That’s a good point :-( > > > Why doing a remove and not simply check if the > > listener has already been added? > > Because there is no way to check, except in the implementation. From the > Observable interface level, there is no way to a) force all implementations > of the interface to implement the method correctly (without breaking source > compatibility anyway), or b) to provide a reasonable default implementation. > > Maybe this is one of those things we can’t fix on the Observable interface > and just have to provide implementations of on our concrete properties. > > Richard
Re: [8u] API Request: RT-25613, ObservableValue should have a hasListener(listener) method
Hi Tomas About the example you request: Say you and I implemented a computer game with spaceships on screen, and lets imagine that these spaceships attacked in groups. Everytime a spaceship was added to a group, we wanted to listen to a property of that spaceship from that point in time and until eternity. Then, imagine we took out a spaceship from the group. We still wanted to listen to a property of the spaceship, but now it was no longer part of the group. Then, imagine we added the same spaceship to the same group once more. At this point, we don’t know if we have added the spaceship before, so when we write spaceship.someProperty.addListener(ourListener); we suddenly have ourListener added twice to the property. I know we can write removeListener(ourListener) followed by addListener(ourListener), and while that would work, I would much rather have the listener list ensure that there are no duplicates, since I believe that it is always a bug to register the same listener twice. Yours Randahl On 22 Jan 2014, at 13:34, Tomas Mikula wrote: > Hi Randahl, > > I'm curious about an example where you would take advantage of the behavior > where multiple addListener(listener) calls add the listener just once. > > Anyway, here [1] are helper classes InvalidationSubscriber and > ChangeSubscriber that allow you to do that: > > InvalidationSubscriber subscriber = new InvalidationSubscriber(observable, > listener); > > subscriber.subscribe(); // registers the listener > subscriber.subscribe(); // no-op > > Cheers, > Tomas > > [1] https://gist.github.com/TomasMikula/8557825 > > On Wed, Jan 22, 2014 at 11:23 AM, Randahl Fink Isaksen > wrote: > Hi Martin > > While I agree your proposed solution would work, I still don’t understand why > JavaFX should keep on supporting duplicates in listener collections. Can > anyone come up with just 1 example of an application that might be depending > on having two listeners on the same Observable? E.g. this kind of code: > > myObservable.addListener(myChangeListener); //add it > myObservable.addListener(myChangeListener); //add it again > > In what kind of situation would this sort of code make any sense? > > If we all feel confident that the presence of duplicates listeners is always > an error, I warmly recommend changing the API to be duplicate free. > > Yours > > Randahl > > > > > On 22 Jan 2014, at 11:07, Martin Sladecek wrote: > > > Hi all, > > I would like to start discussion about an addition to API in Observable, > > ObservableValue and all Observable collections. > > There were multiple requests for a way how to avoid duplicates in listeners > > lists. The way RT-25613 solves this is that it introduces public boolean > > hasListener(ListenerType listener) which would return true if the provided > > listener is already registered. > > > > This has one significant drawback that all of Observable* are actually > > interfaces. Means we can only add hasListener as a defender method. The > > problem is with the default implementation. We cannot return anything > > meaningful, so we have to throw an UnsupportedOperationException. The > > problem is that this might blow up unexpectedly when some "older" > > Observable implementation is used. Also, it might be easy to miss when > > implementing the interface, since the IDE might not force you to implement > > it. > > > > So as an alternative solution, I propose adding something like: > > > > ensureListener(ListenerType listener) > > > > which would make sure the listener is on the list and if a listener is > > already present, the number of times listener is registered on the > > Observable will NOT grow after this call. > > > > The default implementation (for Observable) would look like this: > > > > public default void ensureListener(InvalidationListener listener) { > >removeListener(listener); > >addListener(listener); > > } > > > > subclasses might do something more effective. The same would apply to > > ObservableValue and ChangeListener and Observable[List|Set|Map] and > > [List|Set|Map]ChangeListener. > > > > What do you think? > > > > JIRA link: https://javafx-jira.kenai.com/browse/RT-25613 > > > > -Martin > >
Re: [8u] API Request: RT-25613, ObservableValue should have a hasListener(listener) method
Hi Tomas Great idea! I can see how the InvalidationSubscriber idea would work – but it is just a concept right? There is no official API for this, right? Randahl On 22 Jan 2014, at 13:34, Tomas Mikula wrote: > Hi Randahl, > > I'm curious about an example where you would take advantage of the behavior > where multiple addListener(listener) calls add the listener just once. > > Anyway, here [1] are helper classes InvalidationSubscriber and > ChangeSubscriber that allow you to do that: > > InvalidationSubscriber subscriber = new InvalidationSubscriber(observable, > listener); > > subscriber.subscribe(); // registers the listener > subscriber.subscribe(); // no-op > > Cheers, > Tomas > > [1] https://gist.github.com/TomasMikula/8557825 > > On Wed, Jan 22, 2014 at 11:23 AM, Randahl Fink Isaksen > wrote: > Hi Martin > > While I agree your proposed solution would work, I still don’t understand why > JavaFX should keep on supporting duplicates in listener collections. Can > anyone come up with just 1 example of an application that might be depending > on having two listeners on the same Observable? E.g. this kind of code: > > myObservable.addListener(myChangeListener); //add it > myObservable.addListener(myChangeListener); //add it again > > In what kind of situation would this sort of code make any sense? > > If we all feel confident that the presence of duplicates listeners is always > an error, I warmly recommend changing the API to be duplicate free. > > Yours > > Randahl > > > > > On 22 Jan 2014, at 11:07, Martin Sladecek wrote: > > > Hi all, > > I would like to start discussion about an addition to API in Observable, > > ObservableValue and all Observable collections. > > There were multiple requests for a way how to avoid duplicates in listeners > > lists. The way RT-25613 solves this is that it introduces public boolean > > hasListener(ListenerType listener) which would return true if the provided > > listener is already registered. > > > > This has one significant drawback that all of Observable* are actually > > interfaces. Means we can only add hasListener as a defender method. The > > problem is with the default implementation. We cannot return anything > > meaningful, so we have to throw an UnsupportedOperationException. The > > problem is that this might blow up unexpectedly when some "older" > > Observable implementation is used. Also, it might be easy to miss when > > implementing the interface, since the IDE might not force you to implement > > it. > > > > So as an alternative solution, I propose adding something like: > > > > ensureListener(ListenerType listener) > > > > which would make sure the listener is on the list and if a listener is > > already present, the number of times listener is registered on the > > Observable will NOT grow after this call. > > > > The default implementation (for Observable) would look like this: > > > > public default void ensureListener(InvalidationListener listener) { > >removeListener(listener); > >addListener(listener); > > } > > > > subclasses might do something more effective. The same would apply to > > ObservableValue and ChangeListener and Observable[List|Set|Map] and > > [List|Set|Map]ChangeListener. > > > > What do you think? > > > > JIRA link: https://javafx-jira.kenai.com/browse/RT-25613 > > > > -Martin > >
Re: [8u] API Request: RT-25613, ObservableValue should have a hasListener(listener) method
[...] > Actually even when you would rely on the order, in situations when you > call ensureListener, you don't really know if the listener is already > there. So you might really expect that listener would be added at this > point as the last. It's just that will always be the outcome if the > default implementation is used. It would be even worse if this would be different in none default implementations. Tom
Re: [8u] API Request: RT-25613, ObservableValue should have a hasListener(listener) method
On 01/22/2014 12:30 PM, Richard Bair wrote: The default implementation (for Observable) would look like this: public default void ensureListener(InvalidationListener listener) { removeListener(listener); addListener(listener); } subclasses might do something more effective. The same would apply to ObservableValue and ChangeListener and Observable[List|Set|Map] and [List|Set|Map]ChangeListener. Well this would destroy the order! I expect listeners to be called in the correct order not? That’s a good point :-( Actually even when you would rely on the order, in situations when you call ensureListener, you don't really know if the listener is already there. So you might really expect that listener would be added at this point as the last. It's just that will always be the outcome if the default implementation is used. -Martin
Re: [8u] API Request: RT-25613, ObservableValue should have a hasListener(listener) method
Hi Randahl, I'm curious about an example where you would take advantage of the behavior where multiple addListener(listener) calls add the listener just once. Anyway, here [1] are helper classes InvalidationSubscriber and ChangeSubscriber that allow you to do that: InvalidationSubscriber subscriber = new InvalidationSubscriber(observable, listener); subscriber.subscribe(); // registers the listener subscriber.subscribe(); // no-op Cheers, Tomas [1] https://gist.github.com/TomasMikula/8557825 On Wed, Jan 22, 2014 at 11:23 AM, Randahl Fink Isaksen wrote: > Hi Martin > > While I agree your proposed solution would work, I still don’t understand > why JavaFX should keep on supporting duplicates in listener collections. > Can anyone come up with just 1 example of an application that might be > depending on having two listeners on the same Observable? E.g. this kind of > code: > > myObservable.addListener(myChangeListener); //add it > myObservable.addListener(myChangeListener); //add it again > > In what kind of situation would this sort of code make any sense? > > If we all feel confident that the presence of duplicates listeners is > always an error, I warmly recommend changing the API to be duplicate free. > > Yours > > Randahl > > > > > On 22 Jan 2014, at 11:07, Martin Sladecek > wrote: > > > Hi all, > > I would like to start discussion about an addition to API in Observable, > ObservableValue and all Observable collections. > > There were multiple requests for a way how to avoid duplicates in > listeners lists. The way RT-25613 solves this is that it introduces public > boolean hasListener(ListenerType listener) which would return true if the > provided listener is already registered. > > > > This has one significant drawback that all of Observable* are actually > interfaces. Means we can only add hasListener as a defender method. The > problem is with the default implementation. We cannot return anything > meaningful, so we have to throw an UnsupportedOperationException. The > problem is that this might blow up unexpectedly when some "older" > Observable implementation is used. Also, it might be easy to miss when > implementing the interface, since the IDE might not force you to implement > it. > > > > So as an alternative solution, I propose adding something like: > > > > ensureListener(ListenerType listener) > > > > which would make sure the listener is on the list and if a listener is > already present, the number of times listener is registered on the > Observable will NOT grow after this call. > > > > The default implementation (for Observable) would look like this: > > > > public default void ensureListener(InvalidationListener listener) { > >removeListener(listener); > >addListener(listener); > > } > > > > subclasses might do something more effective. The same would apply to > ObservableValue and ChangeListener and Observable[List|Set|Map] and > [List|Set|Map]ChangeListener. > > > > What do you think? > > > > JIRA link: https://javafx-jira.kenai.com/browse/RT-25613 > > > > -Martin > >
Re: [8u] API Request: RT-25613, ObservableValue should have a hasListener(listener) method
On 01/22/2014 12:30 PM, Richard Bair wrote: The default implementation (for Observable) would look like this: public default void ensureListener(InvalidationListener listener) { removeListener(listener); addListener(listener); } subclasses might do something more effective. The same would apply to ObservableValue and ChangeListener and Observable[List|Set|Map] and [List|Set|Map]ChangeListener. Well this would destroy the order! I expect listeners to be called in the correct order not? That’s a good point :-( Why doing a remove and not simply check if the listener has already been added? Because there is no way to check, except in the implementation. From the Observable interface level, there is no way to a) force all implementations of the interface to implement the method correctly (without breaking source compatibility anyway), or b) to provide a reasonable default implementation. Maybe this is one of those things we can’t fix on the Observable interface and just have to provide implementations of on our concrete properties. Unfortunately, this is not possible (without requiring users to cast). Most properties are exposed in API through ReadOnly*Property and *Property classes, which are abstract and do not have access to the listeners. Same with FX collections, an interface is returned from the API, not the base class. -Martin
Re: [8u] API Request: RT-25613, ObservableValue should have a hasListener(listener) method
On Wed, Jan 22, 2014 at 11:27 AM, Tom Schindl wrote: > On 22.01.14 11:07, Martin Sladecek wrote: > > Hi all, > > I would like to start discussion about an addition to API in Observable, > > ObservableValue and all Observable collections. > > There were multiple requests for a way how to avoid duplicates in > > listeners lists. The way RT-25613 solves this is that it introduces > > public boolean hasListener(ListenerType listener) which would return > > true if the provided listener is already registered. > > > > This has one significant drawback that all of Observable* are actually > > interfaces. Means we can only add hasListener as a defender method. The > > problem is with the default implementation. We cannot return anything > > meaningful, so we have to throw an UnsupportedOperationException. The > > problem is that this might blow up unexpectedly when some "older" > > Observable implementation is used. Also, it might be easy to miss when > > implementing the interface, since the IDE might not force you to > > implement it. > > > > So as an alternative solution, I propose adding something like: > > > > ensureListener(ListenerType listener) > > > > which would make sure the listener is on the list and if a listener is > > already present, the number of times listener is registered on the > > Observable will NOT grow after this call. > > > > The default implementation (for Observable) would look like this: > > > > public default void ensureListener(InvalidationListener listener) { > > removeListener(listener); > > addListener(listener); > > } > > > > subclasses might do something more effective. The same would apply to > > ObservableValue and ChangeListener and Observable[List|Set|Map] and > > [List|Set|Map]ChangeListener. > > Well this would destroy the order! I expect listeners to be called in > the correct order not? Why doing a remove and not simply check if the > listener has already been added? I, too, rely on the implementation detail that listeners are called in the order of registration. My code doesn't break as long as the internal JavaFX implementation keeps using Lists. I will just not use ensureListener() where the order matters. Tomas
Re: [8u] API Request: RT-25613, ObservableValue should have a hasListener(listener) method
I respect your point of view. Thank you for a thorough response. Randahl On 22 Jan 2014, at 12:18, Martin Sladecek wrote: > On 01/22/2014 11:38 AM, Randahl Fink Isaksen wrote: >> Hi Martin >> >> Then I respectfully disagree with this design decision. In my point of view, >> choosing performance over ease of use is rarely a good idea. Here, the >> performance choice has put us in a situation where no one knows how many >> JavaFX apps have duplicate listener bugs, and such bugs can be very hard to >> debug. >> >> Have anyone done any tests that prove that avoiding listener duplicates >> would lead to a severe performance impact? > I'm not the original author of the Observable API, but I remember there were > many performance tests back at the days before 2.0 release, esp. when such > API decision was made. So likely, the tests were done, but the results were > not archived or anything. > >> >> I understand, that if all observables had thousands of listeners and we >> searched for a possible duplicate using an inefficient linear search, we >> would have a problem. But if reality is that very few observables have more >> than 20 listeners and these could be stored in a map for efficient duplicate >> checking, then what is the problem? > Well it's always a balance between performance, dynamic footprint (a set/map > for duplicates) and API. Having an API that extensively checks for all the > mistakes a developer can do is an extreme case in the same manner as an API > that does not check input at all. If a cleanup of some listeners is missing, > it's a developer's bug (memory leak) no matter if there a duplicate listener > check or not on the subsequent addListener call when the object becomes > "valid" again (or whatnot). > > I understand your position, there are APIs I would also like to behave > differently, but the decision was already made, it's too late for the > discussion. There might be 3rd implementations that use a list a don't > guarantee a duplicate check, we don't want to make them suddenly "broken". > > I also disagree that such bugs are hard to debug, a simple printout should do. > In many cases it would likely just make a computations run more times than > necessary. I can imagine doing a duplicate check on something like > -Djavafx.debug=true. > > Regards, > -Martin > > >> >> Yours >> >> Randahl >> >> >> On 22 Jan 2014, at 11:26, Martin Sladecek >> wrote: >> >>> The reason why this was decided this way is simple : performance. You >>> usually don't (try to) add a listener twice, so in most cases it doesn't >>> make sense to check for duplicates every time a listener is added. So we >>> currently leave the burden of avoiding duplicates on the developer. >>> >>> -Martin >>> >>> On 01/22/2014 11:23 AM, Randahl Fink Isaksen wrote: Hi Martin While I agree your proposed solution would work, I still don’t understand why JavaFX should keep on supporting duplicates in listener collections. Can anyone come up with just 1 example of an application that might be depending on having two listeners on the same Observable? E.g. this kind of code: myObservable.addListener(myChangeListener); //add it myObservable.addListener(myChangeListener); //add it again In what kind of situation would this sort of code make any sense? If we all feel confident that the presence of duplicates listeners is always an error, I warmly recommend changing the API to be duplicate free. Yours Randahl On 22 Jan 2014, at 11:07, Martin Sladecek wrote: > Hi all, > I would like to start discussion about an addition to API in Observable, > ObservableValue and all Observable collections. > There were multiple requests for a way how to avoid duplicates in > listeners lists. The way RT-25613 solves this is that it introduces > public boolean hasListener(ListenerType listener) which would return true > if the provided listener is already registered. > > This has one significant drawback that all of Observable* are actually > interfaces. Means we can only add hasListener as a defender method. The > problem is with the default implementation. We cannot return anything > meaningful, so we have to throw an UnsupportedOperationException. The > problem is that this might blow up unexpectedly when some "older" > Observable implementation is used. Also, it might be easy to miss when > implementing the interface, since the IDE might not force you to > implement it. > > So as an alternative solution, I propose adding something like: > > ensureListener(ListenerType listener) > > which would make sure the listener is on the list and if a listener is > already present, the number of times listener is registered on the > Observable will NOT grow after
Re: [8u] API Request: RT-25613, ObservableValue should have a hasListener(listener) method
>> The default implementation (for Observable) would look like this: >> >> public default void ensureListener(InvalidationListener listener) { >>removeListener(listener); >>addListener(listener); >> } >> >> subclasses might do something more effective. The same would apply to >> ObservableValue and ChangeListener and Observable[List|Set|Map] and >> [List|Set|Map]ChangeListener. > > Well this would destroy the order! I expect listeners to be called in > the correct order not? That’s a good point :-( > Why doing a remove and not simply check if the > listener has already been added? Because there is no way to check, except in the implementation. From the Observable interface level, there is no way to a) force all implementations of the interface to implement the method correctly (without breaking source compatibility anyway), or b) to provide a reasonable default implementation. Maybe this is one of those things we can’t fix on the Observable interface and just have to provide implementations of on our concrete properties. Richard
Re: [8u] API Request: RT-25613, ObservableValue should have a hasListener(listener) method
On 01/22/2014 11:38 AM, Randahl Fink Isaksen wrote: Hi Martin Then I respectfully disagree with this design decision. In my point of view, choosing performance over ease of use is rarely a good idea. Here, the performance choice has put us in a situation where no one knows how many JavaFX apps have duplicate listener bugs, and such bugs can be very hard to debug. Have anyone done any tests that prove that avoiding listener duplicates would lead to a severe performance impact? I'm not the original author of the Observable API, but I remember there were many performance tests back at the days before 2.0 release, esp. when such API decision was made. So likely, the tests were done, but the results were not archived or anything. I understand, that if all observables had thousands of listeners and we searched for a possible duplicate using an inefficient linear search, we would have a problem. But if reality is that very few observables have more than 20 listeners and these could be stored in a map for efficient duplicate checking, then what is the problem? Well it's always a balance between performance, dynamic footprint (a set/map for duplicates) and API. Having an API that extensively checks for all the mistakes a developer can do is an extreme case in the same manner as an API that does not check input at all. If a cleanup of some listeners is missing, it's a developer's bug (memory leak) no matter if there a duplicate listener check or not on the subsequent addListener call when the object becomes "valid" again (or whatnot). I understand your position, there are APIs I would also like to behave differently, but the decision was already made, it's too late for the discussion. There might be 3rd implementations that use a list a don't guarantee a duplicate check, we don't want to make them suddenly "broken". I also disagree that such bugs are hard to debug, a simple printout should do. In many cases it would likely just make a computations run more times than necessary. I can imagine doing a duplicate check on something like -Djavafx.debug=true. Regards, -Martin Yours Randahl On 22 Jan 2014, at 11:26, Martin Sladecek wrote: The reason why this was decided this way is simple : performance. You usually don't (try to) add a listener twice, so in most cases it doesn't make sense to check for duplicates every time a listener is added. So we currently leave the burden of avoiding duplicates on the developer. -Martin On 01/22/2014 11:23 AM, Randahl Fink Isaksen wrote: Hi Martin While I agree your proposed solution would work, I still don’t understand why JavaFX should keep on supporting duplicates in listener collections. Can anyone come up with just 1 example of an application that might be depending on having two listeners on the same Observable? E.g. this kind of code: myObservable.addListener(myChangeListener); //add it myObservable.addListener(myChangeListener); //add it again In what kind of situation would this sort of code make any sense? If we all feel confident that the presence of duplicates listeners is always an error, I warmly recommend changing the API to be duplicate free. Yours Randahl On 22 Jan 2014, at 11:07, Martin Sladecek wrote: Hi all, I would like to start discussion about an addition to API in Observable, ObservableValue and all Observable collections. There were multiple requests for a way how to avoid duplicates in listeners lists. The way RT-25613 solves this is that it introduces public boolean hasListener(ListenerType listener) which would return true if the provided listener is already registered. This has one significant drawback that all of Observable* are actually interfaces. Means we can only add hasListener as a defender method. The problem is with the default implementation. We cannot return anything meaningful, so we have to throw an UnsupportedOperationException. The problem is that this might blow up unexpectedly when some "older" Observable implementation is used. Also, it might be easy to miss when implementing the interface, since the IDE might not force you to implement it. So as an alternative solution, I propose adding something like: ensureListener(ListenerType listener) which would make sure the listener is on the list and if a listener is already present, the number of times listener is registered on the Observable will NOT grow after this call. The default implementation (for Observable) would look like this: public default void ensureListener(InvalidationListener listener) { removeListener(listener); addListener(listener); } subclasses might do something more effective. The same would apply to ObservableValue and ChangeListener and Observable[List|Set|Map] and [List|Set|Map]ChangeListener. What do you think? JIRA link: https://javafx-jira.kenai.com/browse/RT-25613 -Martin
Re: [8u] API Request: RT-25613, ObservableValue should have a hasListener(listener) method
On 01/22/2014 11:27 AM, Tom Schindl wrote: On 22.01.14 11:07, Martin Sladecek wrote: Hi all, I would like to start discussion about an addition to API in Observable, ObservableValue and all Observable collections. There were multiple requests for a way how to avoid duplicates in listeners lists. The way RT-25613 solves this is that it introduces public boolean hasListener(ListenerType listener) which would return true if the provided listener is already registered. This has one significant drawback that all of Observable* are actually interfaces. Means we can only add hasListener as a defender method. The problem is with the default implementation. We cannot return anything meaningful, so we have to throw an UnsupportedOperationException. The problem is that this might blow up unexpectedly when some "older" Observable implementation is used. Also, it might be easy to miss when implementing the interface, since the IDE might not force you to implement it. So as an alternative solution, I propose adding something like: ensureListener(ListenerType listener) which would make sure the listener is on the list and if a listener is already present, the number of times listener is registered on the Observable will NOT grow after this call. The default implementation (for Observable) would look like this: public default void ensureListener(InvalidationListener listener) { removeListener(listener); addListener(listener); } subclasses might do something more effective. The same would apply to ObservableValue and ChangeListener and Observable[List|Set|Map] and [List|Set|Map]ChangeListener. Well this would destroy the order! I expect listeners to be called in the correct order not? Why doing a remove and not simply check if the listener has already been added? Tom Because there's no way to do it in the interface, hence the problem with hasListener default implementation. Yes, the order would be broken, but it's actually not guaranteed. Although FX internally uses a List, 3rd party implementations of Observable or ObservableValue might use a Set for listeners for example. This was the idea, but funnily enough, the current Observable javadoc is quite strict on the duplicates which would rule out the Set: /** * Adds an {@link InvalidationListener} which will be notified whenever the * {@code Observable} becomes invalid. If the same * listener is added more than once, then it will be notified more than * once. That is, no check is made to ensure uniqueness. Personally, I would rather make this requirement less strict and allow both List backed and Set backed implementations. -Martin
Re: [8u] API Request: RT-25613, ObservableValue should have a hasListener(listener) method
Hi Martin Then I respectfully disagree with this design decision. In my point of view, choosing performance over ease of use is rarely a good idea. Here, the performance choice has put us in a situation where no one knows how many JavaFX apps have duplicate listener bugs, and such bugs can be very hard to debug. Have anyone done any tests that prove that avoiding listener duplicates would lead to a severe performance impact? I understand, that if all observables had thousands of listeners and we searched for a possible duplicate using an inefficient linear search, we would have a problem. But if reality is that very few observables have more than 20 listeners and these could be stored in a map for efficient duplicate checking, then what is the problem? Yours Randahl On 22 Jan 2014, at 11:26, Martin Sladecek wrote: > The reason why this was decided this way is simple : performance. You usually > don't (try to) add a listener twice, so in most cases it doesn't make sense > to check for duplicates every time a listener is added. So we currently leave > the burden of avoiding duplicates on the developer. > > -Martin > > On 01/22/2014 11:23 AM, Randahl Fink Isaksen wrote: >> Hi Martin >> >> While I agree your proposed solution would work, I still don’t understand >> why JavaFX should keep on supporting duplicates in listener collections. Can >> anyone come up with just 1 example of an application that might be depending >> on having two listeners on the same Observable? E.g. this kind of code: >> >> myObservable.addListener(myChangeListener); //add it >> myObservable.addListener(myChangeListener); //add it again >> >> In what kind of situation would this sort of code make any sense? >> >> If we all feel confident that the presence of duplicates listeners is always >> an error, I warmly recommend changing the API to be duplicate free. >> >> Yours >> >> Randahl >> >> >> >> >> On 22 Jan 2014, at 11:07, Martin Sladecek wrote: >> >>> Hi all, >>> I would like to start discussion about an addition to API in Observable, >>> ObservableValue and all Observable collections. >>> There were multiple requests for a way how to avoid duplicates in listeners >>> lists. The way RT-25613 solves this is that it introduces public boolean >>> hasListener(ListenerType listener) which would return true if the provided >>> listener is already registered. >>> >>> This has one significant drawback that all of Observable* are actually >>> interfaces. Means we can only add hasListener as a defender method. The >>> problem is with the default implementation. We cannot return anything >>> meaningful, so we have to throw an UnsupportedOperationException. The >>> problem is that this might blow up unexpectedly when some "older" >>> Observable implementation is used. Also, it might be easy to miss when >>> implementing the interface, since the IDE might not force you to implement >>> it. >>> >>> So as an alternative solution, I propose adding something like: >>> >>> ensureListener(ListenerType listener) >>> >>> which would make sure the listener is on the list and if a listener is >>> already present, the number of times listener is registered on the >>> Observable will NOT grow after this call. >>> >>> The default implementation (for Observable) would look like this: >>> >>> public default void ensureListener(InvalidationListener listener) { >>>removeListener(listener); >>>addListener(listener); >>> } >>> >>> subclasses might do something more effective. The same would apply to >>> ObservableValue and ChangeListener and Observable[List|Set|Map] and >>> [List|Set|Map]ChangeListener. >>> >>> What do you think? >>> >>> JIRA link: https://javafx-jira.kenai.com/browse/RT-25613 >>> >>> -Martin >
Re: [8u] API Request: RT-25613, ObservableValue should have a hasListener(listener) method
On 22.01.14 11:07, Martin Sladecek wrote: > Hi all, > I would like to start discussion about an addition to API in Observable, > ObservableValue and all Observable collections. > There were multiple requests for a way how to avoid duplicates in > listeners lists. The way RT-25613 solves this is that it introduces > public boolean hasListener(ListenerType listener) which would return > true if the provided listener is already registered. > > This has one significant drawback that all of Observable* are actually > interfaces. Means we can only add hasListener as a defender method. The > problem is with the default implementation. We cannot return anything > meaningful, so we have to throw an UnsupportedOperationException. The > problem is that this might blow up unexpectedly when some "older" > Observable implementation is used. Also, it might be easy to miss when > implementing the interface, since the IDE might not force you to > implement it. > > So as an alternative solution, I propose adding something like: > > ensureListener(ListenerType listener) > > which would make sure the listener is on the list and if a listener is > already present, the number of times listener is registered on the > Observable will NOT grow after this call. > > The default implementation (for Observable) would look like this: > > public default void ensureListener(InvalidationListener listener) { > removeListener(listener); > addListener(listener); > } > > subclasses might do something more effective. The same would apply to > ObservableValue and ChangeListener and Observable[List|Set|Map] and > [List|Set|Map]ChangeListener. Well this would destroy the order! I expect listeners to be called in the correct order not? Why doing a remove and not simply check if the listener has already been added? Tom
Re: [8u] API Request: RT-25613, ObservableValue should have a hasListener(listener) method
The reason why this was decided this way is simple : performance. You usually don't (try to) add a listener twice, so in most cases it doesn't make sense to check for duplicates every time a listener is added. So we currently leave the burden of avoiding duplicates on the developer. -Martin On 01/22/2014 11:23 AM, Randahl Fink Isaksen wrote: Hi Martin While I agree your proposed solution would work, I still don’t understand why JavaFX should keep on supporting duplicates in listener collections. Can anyone come up with just 1 example of an application that might be depending on having two listeners on the same Observable? E.g. this kind of code: myObservable.addListener(myChangeListener); //add it myObservable.addListener(myChangeListener); //add it again In what kind of situation would this sort of code make any sense? If we all feel confident that the presence of duplicates listeners is always an error, I warmly recommend changing the API to be duplicate free. Yours Randahl On 22 Jan 2014, at 11:07, Martin Sladecek wrote: Hi all, I would like to start discussion about an addition to API in Observable, ObservableValue and all Observable collections. There were multiple requests for a way how to avoid duplicates in listeners lists. The way RT-25613 solves this is that it introduces public boolean hasListener(ListenerType listener) which would return true if the provided listener is already registered. This has one significant drawback that all of Observable* are actually interfaces. Means we can only add hasListener as a defender method. The problem is with the default implementation. We cannot return anything meaningful, so we have to throw an UnsupportedOperationException. The problem is that this might blow up unexpectedly when some "older" Observable implementation is used. Also, it might be easy to miss when implementing the interface, since the IDE might not force you to implement it. So as an alternative solution, I propose adding something like: ensureListener(ListenerType listener) which would make sure the listener is on the list and if a listener is already present, the number of times listener is registered on the Observable will NOT grow after this call. The default implementation (for Observable) would look like this: public default void ensureListener(InvalidationListener listener) { removeListener(listener); addListener(listener); } subclasses might do something more effective. The same would apply to ObservableValue and ChangeListener and Observable[List|Set|Map] and [List|Set|Map]ChangeListener. What do you think? JIRA link: https://javafx-jira.kenai.com/browse/RT-25613 -Martin
Re: [8u] API Request: RT-25613, ObservableValue should have a hasListener(listener) method
Hi Martin While I agree your proposed solution would work, I still don’t understand why JavaFX should keep on supporting duplicates in listener collections. Can anyone come up with just 1 example of an application that might be depending on having two listeners on the same Observable? E.g. this kind of code: myObservable.addListener(myChangeListener); //add it myObservable.addListener(myChangeListener); //add it again In what kind of situation would this sort of code make any sense? If we all feel confident that the presence of duplicates listeners is always an error, I warmly recommend changing the API to be duplicate free. Yours Randahl On 22 Jan 2014, at 11:07, Martin Sladecek wrote: > Hi all, > I would like to start discussion about an addition to API in Observable, > ObservableValue and all Observable collections. > There were multiple requests for a way how to avoid duplicates in listeners > lists. The way RT-25613 solves this is that it introduces public boolean > hasListener(ListenerType listener) which would return true if the provided > listener is already registered. > > This has one significant drawback that all of Observable* are actually > interfaces. Means we can only add hasListener as a defender method. The > problem is with the default implementation. We cannot return anything > meaningful, so we have to throw an UnsupportedOperationException. The problem > is that this might blow up unexpectedly when some "older" Observable > implementation is used. Also, it might be easy to miss when implementing the > interface, since the IDE might not force you to implement it. > > So as an alternative solution, I propose adding something like: > > ensureListener(ListenerType listener) > > which would make sure the listener is on the list and if a listener is > already present, the number of times listener is registered on the Observable > will NOT grow after this call. > > The default implementation (for Observable) would look like this: > > public default void ensureListener(InvalidationListener listener) { >removeListener(listener); >addListener(listener); > } > > subclasses might do something more effective. The same would apply to > ObservableValue and ChangeListener and Observable[List|Set|Map] and > [List|Set|Map]ChangeListener. > > What do you think? > > JIRA link: https://javafx-jira.kenai.com/browse/RT-25613 > > -Martin