Proof of concept for fluent bindings for ObservableValue
John Hendrikx
hjohn at xs4all.nl
Mon Sep 13 00:05:50 UTC 2021
On 12/09/2021 02:05, Nir Lisker wrote:
> I've gotten back to look at this.
>
> For now I'm dealing only with the nullableMapping method in Bindings so
> we can limit the amount of new classes to LazyObjectBinding
> (FlatMapBinding and ConditionalBinding can come later). This method is
> used by map, orElse and orElseGet in ObservableValue. Of these, map is
> the only fundamental one since the other 2 can be represented by it. I
> don't mind keeping them in the discussion, though I will be centered on
> the map method.
>
> The implementation of these methods rely on Bindings, LazyObjectBinding,
> and Subscription in the current implementation. I think that we can
> introduce these internally for now. The biggest hurdle left are the
> public changes to ObjectBinding. If we add protected methods, we need to
> be sure that by the end of this large task they would have been the
> right ones to add and at the right place. This is why I recommend adding
> them at the package visibility level and add LazyObjectBinding (and
> friends) in its package so they can extend it. I understand that this
> can look ugly, but moving internal implementation is cheap, and in this
> case, since the coupling involves about 3 classes, is very cheap. This
> will lower the initial integration barrier and let the community get
> used to- and give feedback on the new changes.
I think that's a good idea, there is no direct need to make those
protected methods part of the public API as the usefulness of those
methods will be limited and the major use case will basically be
provided by LazyObjectBinding already.
> This will leave only 1 change that we are committed to, and that's the
> new API on ObservableValue (which is the map method in this case). The
> method looks good; the only question, which has arisen in a few places,
> is how to handle null. As we discussed here, this method works like its
> ReactFX counterpart, ignoring null. My questions would be:
> 1. Is there a good reason to allow null? If so, do we add a new method
> for it, or do we pass some parameter to the current method to indicate that?
In JavaFX, null is something we have to deal in some fashion as
properties can easily be null. For the "primitive" properties, null (if
encountered) is translated to a default value. For StringProperty it
could be an empty string although JavaFX doesn't do this. For
ObjectProperty there is no sensible default possible.
In ReactFX, nulls are indeed skipped when mapping as it considers null
to be an empty value, and empty values are skipped according to the
documentation. The code below will not throw an NPE in the mapping
function and will simply result in null:
Var.newSimpleVar(null).map(x -> x + "2").getValue();
This is similar to the PoC implementation:
new SimpleStringProperty().map(x -> x + "2").getValue());
Having worked with ReactFX and also the PoC, I think it would be very
cumbersome to have to deal with nulls in mapping functions, as many
simple mappings expressed with a short lambda would need to deal with
the null case with a ternary or an if/else block.
In the PoC any mapping you could need that requires mapping null
explicitely can be expressed in another form:
.map(x -> x == null ? "empty!" : x + "2")
becomes:
.map(x -> x + "2").orElse("empty!)
which is not only more concise, but allows to delay dealing with null
until the very end:
.map(x -> x == null ? "empty!" : fetchDataWhichCouldBeNull(x))
.map(x -> x == null ? "empty!" : x + "2")
versus:
.map(Helper::fetchDataThatCouldBeNull)
.map(x -> x + "2")
.orElse("empty!")
You don't have to delay it though, if for some reason you would want to
map the 2nd null case differently, you could use "orElse" after each
mapping still.
Although for mapping this may seem somewhat contrived, for selecting (or
flatMapping) where you go through a chain of properties (like
Node->Scene->Window) being able to delay dealing with nulls leads to
more concise and IMHO more expressive code.
So in summary, and to answer your first question, I don't think there is
a good reason to allow null to be passed to mapping functions. We do
need to deal with nulls though, and that's what "orElse" is for. This
could also be done with an additional parameter to "map" (a
"mapOrDefault" similar to "getOrDefault" from the collections API) but I
think we'd be better served with multiple methods that take a single
argument as the resulting code is easier to understand especially when
one of the arguments is a lambda.
> 2. If we want to replace the Bindings.select (non-type safe) API, can we
> do it with our current way of treating null?
In the current Bindings.select API, null is skipped when encountered and
the resulting value of the chain will be null. This is exactly what
"flatMap" in the PoC does as well, in other words:
Bindings.select(nodeProperty, "scene", "window", "showing")
is exactly equivalent to:
nodeProperty.flatMap(Node::sceneProperty)
.flatMap(Scene::windowProperty)
.flatMap(Window::showingProperty)
No null checks are needed, and the binding will be null if any of the
selected properties contain null. Note that both versions return null
despite the fact that the last property selected is a primitive boolean
which normally cannot hold null.
However, Bindings also offers selectBoolean. In this case it does
indeed return false when any of the properties contains null, but it
also logs this uncomfortable warning then:
WARNING: Value of select-binding has wrong type, returning default value
(+ stack trace omitted)
The warning is somewhat deceptive as the binding does refer to a boolean
so it is not strictly of the wrong type, but an intermediate value
encountered was null and this cannot be cast to a primitive boolean. It
should probably just convert to the default value without any warning.
So, to answer your question, I think we can indeed replace
Bindings.select with the PoC's typesafe equivalent. There is almost no
need to create specific primitive versions of this mechanism as users
can use "orElse" to map to a suitable primitive value as the last step
if null is undesired.
> Do you think that this is a valid approach?
Yes, I think our messages may have crossed paths as I suggested limiting
the API (based on your earlier recommendations) in a post on the
"Enhancements for JavaFX 18" thread in a reply to Kevin Rushforth.
Your suggestion takes this a bit further by leaving the new methods in
ObjectBinding package protected and reducing the new API in
ObservableValue to its bare essentials. I think that's a fine approach
as it keeps the API that we're commiting to small and allows the highest
flexibility for future extensions.
If an agreement can be reached on the initial API, I can rework the PoC
and also add the unit tests (I'll need to convert the JUnit 5 tests I
have to JUnit 4).
--John
>
> - Nir
>
> On Wed, Apr 7, 2021 at 11:30 PM John Hendrikx <hjohn at xs4all.nl
> <mailto:hjohn at xs4all.nl>> wrote:
>
> On 07/04/2021 03:41, Nir Lisker wrote:
> > In the PoC I made I specifically also disallowed 'null' as an
> input
> >
> >
> > I like the way ReactFX does it where the property is empty. I
> think that
> > this is also what you mean by disallowing `null` (in other contexts,
> > "disallowing null" would mean throwing an exception).
>
> Yes, it is the same concept as ReactFX calling a property "empty",
> but I
> was hesitant to call this as `null` is a valid value for many JavaFX
> properties (a Scene can be null, a String can be null, etc.) which I
> don't think means the same as it being empty (in the Optional sense).
> But as long as the documentation is clear, I don't mind calling it
> either.
>
> >
> > Not entirely sure what you mean by this.
> >
> >
> > Basically, what you said. My point was that this is a different API
> > section. The first deals with expanding the observables/properties
> > methods. The second with listeners methods. Even if mapping a property
> > requires a new listening model, like subscriptions, this is done under
> > the hood. Exposing this API should be a separate step. At least that's
> > how I see it.
>
> Yes, I think it is good to limit new API as much as possible to reduce
> scope and increase the chances of its acceptance. The subscription
> parts
> can be designed separately and do not need to be public at this point.
> They can be moved to a helper, or the implementation can take the extra
> effort to use standard listeners.
>
> >
> > I'd be happy to spend more time and work on this. Perhaps it
> would be
> > possible to collaborate on this?
> >
> >
> > That would be good. I will need to re-review the ReactFX internals and
> > see how your proposal differs exactly.
>
> Yes, I think that would be good to do.
>
> I've done some comparisons myself and didn't find a difference in
> functionality with `Val` (so far). It is a new implementation though, I
> didn't really look at how `Val` was done internally as implementing it
> directly into JavaFX is quite different (I had to make a few minor
> changes in `ObjectBinding` to allow for the choice of lazy binding). I
> was also initially more focused on Streams only to realize at a later
> point that having a Stream implement ObservableValue was not going
> to be
> pretty (I suspect this also happened when ReactFX was created, which is
> why Val/Var were later introduced in 2.x).
>
> Both the PoC and Val do lazy binding and are null safe and provide
> methods to deal with null/empty.
>
> The main thing I didn't do yet is provide a `filter` method. Filtering
> properties that you want to use for bindings seems awkard as a binding
> should always have some kind of value. The `filter` method in ReactFX
> basically maps the value to `null` when it doesn't match the filter.
> I've left this out as you can easily achieve this with `map` and
> `filter` seems to be too easy to misunderstand.
>
> Aside from that, ReactFX's Val offers a lot of other methods that are I
> think a bit too specialized to consider at this point, like the
> `animate`, `pin`, `mapDynamic` and `suspendable` methods.
>
> Val also has all the other `Optional` methods (ifPresent, isPresent,
> isEmpty) but I think they may make the API a bit confusing (an
> observable value is not the same as an optional). I've also not had a
> need for these so far in practice and you can easily convert the
> current
> value to get this functionality with `Optional.ofNullable`.
>
> Finally Val offers a few methods to convert to ReactFX's streams. While
> convenient, I think static methods like `Values.of`, `Invalidations.of`
> or `Changes.of` would make for a less cluttered API to do stream
> conversions -- this would also make it possible to leave this part of
> the API up to a 3rd party.
>
> > By the way, do you make a distinction between ReactFX's Val and
> Var in
> > your proposal (one being read-only)?
>
> No, `ObservableValue` is basically the same as `Val`, and the
> equivalent
> to `Var` is `ObjectProperty`. Aside from it being a good companion to
> `Val` (and less typing), I don't see a reason to implement `Var`.
>
> --John
>
> >
> > On Sun, Apr 4, 2021 at 12:43 PM John Hendrikx <hjohn at xs4all.nl
> <mailto:hjohn at xs4all.nl>
> > <mailto:hjohn at xs4all.nl <mailto:hjohn at xs4all.nl>>> wrote:
> >
> >
> >
> > On 02/04/2021 08:47, Nir Lisker wrote:
> > > Hi John,
> > >
> > > I've had my eyes set on ReactFX enhancements for a while too,
> > especially as
> > > a replacement for the unsafe "select" mechanism. One of the
> things
> > that
> > > kept me from going forward with this is seeing what Valhalla
> will
> > bring.
> > > Generic specialization might save a lot of duplication work on
> > something
> > > like this, and Tomas touched another related issue [1], but
> since
> > it could
> > > be a long time before that happens, it's worth planning what we
> > can extract
> > > from ReactFX currently.
> >
> > Agreed, Valhalla is certainly a highly anticipated feature but I
> > fear it
> > is still a couple of years away.
> >
> > Even without any initial support for dealing with "? extends
> Number"
> > from the various ObservableValue specializations I think
> looking into
> > this can already be tremendous help.
> >
> > The proof of concept mainly requires you convert the Number to a
> > suitable type when reading the property but has no problems in the
> > other
> > direction:
> >
> > label.widthProperty().map(Number::doubleValue).map(x -> x
> + 1);
> >
> > Not pretty, but certainly workable. Specific methods could be
> > introduced
> > (even at a later time) to make this more streamlined, similar
> to what
> > the Stream API offers with 'mapToDouble' etc.
> >
> > > I think that we should break the enhancements into parts.
> > > The first that I would advise to look at are the additions to
> > > properties/observables. Tomas had to create Val and Var
> because he
> > couldn't
> > > change the core interfaces, but we can. Fitting them with
> the Optional
> > > methods like `isPresent`, `isEmpty`, `ifPresent`, `map`.
> `flatMap`
> > etc.;
> > > and `select` and friends, is already a good start that will
> > address many
> > > common requirements.
> >
> > Yes, Val/Var had to be created for that reason, and also because
> > properties don't quite behave the same as streams -- streams
> with a
> > "toBinding" method results in things people didn't quite expect.
> >
> > As far as the Optional methods go, I'm not entirely sure
> properties
> > would benefit from all of them. Properties are not immutable like
> > Optional and it may make less sense to fit them with 'isPresent',
> > 'isEmpty' and 'ifPresent' ('ifPresent' would I think need to
> behave
> > similar to 'addListener' or 'subscribe').
> >
> > In the PoC I made I specifically also disallowed 'null' as an
> input for
> > functions like 'map' and 'flatMap' (opting to use 'orElse'
> semantics
> > for
> > 'null'), as this for allows much cleaner mapping (and
> especially flat
> > mapping when selecting nested properties). If 'null' were to be
> > allowed,
> > I think at a minimum we'd need to add another method to allow
> for easy
> > selecting of nested properties to avoid:
> >
> > obs.flatMap(x -> x == null ? null : x.otherProperty())
> >
> > > The second part is related to listeners. The subscription model
> > and event
> > > streams try to solve the memory issues with hard and weak
> > references, and
> > > allow better composition.
> >
> > Not entirely sure what you mean by this. JavaFX's current
> model uses
> > weak references which was I think an unfortunate decision as
> it can
> > result in huge confusion. For example, a direct binding will
> work, but
> > with an indirection step a binding stops working:
> >
> > button.textProperty()
> > .concat("World") // weak binding used here
> > .addListener((obs, old, cur) -> System.out.println(cur));
> >
> > The above stops working, but without the 'concat' it keeps
> working.
> >
> > I think the use of weak listeners should be avoided and
> instead other
> > mechanisms should be provided to make cleaning up easier. This
> is the
> > main reason for 'conditionOn' and why ReactFX even had a
> specialized
> > version of it: 'conditionOnShowing(Node)'.
> >
> > > The third part is for collections - things like
> transformation lists
> > > (LiveList) and for other collections.
> >
> > This is indeed best saved for last. The problems there I think
> are less
> > of an issue for now.
> >
> > > Since these share behavior under the hood, we need to look
> ahead,
> > but in
> > > terms of functionality, I think we should take smaller steps. It
> > will also
> > > be easier to propose these then.
> >
> > I've for this reason kept the PoC small with only the most basic
> > functionality. I did however add some work for a different
> > subscription
> > model, mainly because the internals of this code benefits
> greatly from
> > it. It is however kept to a minimum.
> >
> > I'd be happy to spend more time and work on this. Perhaps it
> would be
> > possible to collaborate on this?
> >
> > --John
> >
> > >
> > > - Nir
> > >
> > > [1]
> > >
> >
> https://github.com/TomasMikula/ReactFX/wiki/Creating-a-Val-or-Var-Instance#the-javafx-propertynumber-implementation-issue
> > >
> > > On Wed, Mar 24, 2021 at 11:49 PM John Hendrikx
> <hjohn at xs4all.nl <mailto:hjohn at xs4all.nl>
> > <mailto:hjohn at xs4all.nl <mailto:hjohn at xs4all.nl>>> wrote:
> > >
> > >> I just wanted to draw some attention to a recent proof of
> concept
> > I made
> > >> in this pull request: https://github.com/openjdk/jfx/pull/434
> > >>
> > >> It is based on the work I did in
> > >> https://github.com/hjohn/hs.jfx.eventstream which is in
> part based on
> > >> work done in ReactFX by Tomas Mikula. The PR itself however
> shares no
> > >> code with ReactFX and is
> > >> completely written by me.
> > >>
> > >> If there is interest, I'm willing to invest more time in
> > smoothing out
> > >> the API and documentation, investigating further how this would
> > interact
> > >> with the primitive types and adding unit test coverage (I have
> > extensive
> > >> tests, but thesea are written in JUnit 5, so they would require
> > >> conversion or JavaFX could move to support JUnit 5).
> > >>
> > >> What follows below is the text of the PR for easy reading.
> > Feedback is
> > >> appreciated.
> > >>
> > >> ================
> > >>
> > >> This is a proof of concept of how fluent bindings could be
> > introduced to
> > >> JavaFX. The main benefit of fluent bindings are ease of
> use, type
> > safety
> > >> and less surprises. Features:
> > >>
> > >> Flexible Mappings
> > >> Map the contents of a property any way you like with map,
> or map
> > nested
> > >> properties with flatMap.
> > >>
> > >> Lazy
> > >> The bindings created are lazy, which means they are always
> > invalid when
> > >> not themselves observed. This allows for easier garbage
> > collection (once
> > >> the last observer is removed, a chain of bindings will stop
> observing
> > >> their parents) and less listener management when dealing
> with nested
> > >> properties. Furthermore, this allows inclusion of such
> bindings in
> > >> classes such as Node without listeners being created when
> the binding
> > >> itself is not used (this would allow for the inclusion of a
> > >> treeShowingProperty in Node without creating excessive
> listeners, see
> > >> this fix I did in an earlier PR: #185)
> > >>
> > >> Null Safe
> > >> The map and flatMap methods are skipped, similar to
> > java.util.Optional
> > >> when the value they would be mapping is null. This makes
> mapping
> > nested
> > >> properties with flatMap trivial as the null case does not
> need to be
> > >> taken into account in a chain like this:
> > >>
> >
> node.sceneProperty().flatMap(Scene::windowProperty).flatMap(Window::showingProperty).
> > >>
> > >> Instead a default can be provided with orElse or orElseGet.
> > >>
> > >> Conditional Bindings
> > >> Bindings can be made conditional using the conditionOn
> method. A
> > >> conditional binding retains its last value when its
> condition is
> > false.
> > >> Conditional bindings donot observe their source when the
> condition is
> > >> false, allowing developers to automatically stop listening to
> > properties
> > >> when a certain condition is met. A major use of this feature is
> > to have
> > >> UI components that need to keep models updated which may
> outlive
> > the UI
> > >> conditionally update the long lived model only when the UI is
> > showing.
> > >>
> > >> Some examples:
> > >>
> > >> void mapProperty() {
> > >> // Standard JavaFX:
> > >> label.textProperty().bind(Bindings.createStringBinding(() ->
> > >> text.getValueSafe().toUpperCase(), text));
> > >>
> > >> // Fluent: much more compact, no need to handle null
> > >> label.textProperty().bind(text.map(String::toUpperCase));
> > >> }
> > >>
> > >> void calculateCharactersLeft() {
> > >> // Standard JavaFX:
> > >>
> > >>
> >
> label.textProperty().bind(text.length().negate().add(100).asString().concat("
> > >>
> > >> characters left"));
> > >>
> > >> // Fluent: slightly more compact and more clear (no
> negate needed)
> > >> label.textProperty().bind(text.orElse("").map(v -> 100 -
> > v.length() +
> > >> " characters left"));
> > >> }
> > >>
> > >> void mapNestedValue() {
> > >> // Standard JavaFX:
> > >> label.textProperty().bind(Bindings.createStringBinding(
> > >> () -> employee.get() == null ? ""
> > >> : employee.get().getCompany() == null ? ""
> > >> : employee.get().getCompany().getName(),
> > >> employee
> > >> ));
> > >>
> > >> // Fluent: no need to handle nulls everywhere
> > >> label.textProperty().bind(
> > >> employee.map(Employee::getCompany)
> > >> .map(Company::getName)
> > >> .orElse("")
> > >> );
> > >> }
> > >>
> > >> void mapNestedProperty() {
> > >> // Standard JavaFX:
> > >> label.textProperty().bind(
> > >>
> Bindings.when(Bindings.selectBoolean(label.sceneProperty(),
> > >> "window", "showing"))
> > >> .then("Visible")
> > >> .otherwise("Not Visible")
> > >> );
> > >>
> > >> // Fluent: type safe
> > >> label.textProperty().bind(label.sceneProperty()
> > >> .flatMap(Scene::windowProperty)
> > >> .flatMap(Window::showingProperty)
> > >> .orElse(false)
> > >> .map(showing -> showing ? "Visible" : "Not Visible")
> > >> );
> > >> }
> > >>
> > >> void updateLongLivedModelWhileAvoidingMemoryLeaks() {
> > >> // Standard JavaFX: naive, memory leak; UI won't get garbage
> > collected
> > >>
> listView.getSelectionModel().selectedItemProperty().addListener(
> > >> (obs, old, current) ->
> > >> longLivedModel.lastSelectedProperty().set(current)
> > >> );
> > >>
> > >> // Standard JavaFX: no leak, but stops updating after a
> while
> > >>
> listView.getSelectionModel().selectedItemProperty().addListener(
> > >> new WeakChangeListener<>(
> > >> (obs, old, current) ->
> > >> longLivedModel.lastSelectedProperty().set(current)
> > >> )
> > >> );
> > >>
> > >> // Standard JavaFX: fixed version
> > >> listenerReference = (obs, old, current) ->
> > >> longLivedModel.lastSelectedProperty().set(current);
> > >>
> > >>
> listView.getSelectionModel().selectedItemProperty().addListener(
> > >> new WeakChangeListener<>(listenerReference)
> > >> );
> > >>
> > >> // Fluent: naive, memory leak... fluent won't solve this...
> > >> listView.getSelectionModel().selectedItemProperty()
> > >> .subscribe(longLivedModel.lastSelectedProperty()::set);
> > >>
> > >> // Fluent: conditional update when control visible
> > >>
> > >> // Create a property which is only true when the UI is
> visible:
> > >> ObservableValue<Boolean> showing = listView.sceneProperty()
> > >> .flatMap(Scene::windowProperty)
> > >> .flatMap(Window::showingProperty)
> > >> .orElse(false);
> > >>
> > >> // Use showing property to automatically disconnect long
> lived
> > model
> > >> // allowing garbage collection of the UI:
> > >> listView.getSelectionModel().selectedItemProperty()
> > >> .conditionOn(showing)
> > >> .subscribe(longLivedModel.lastSelectedProperty()::set);
> > >>
> > >> // Note that the 'showing' property can be provided in
> > multiple ways:
> > >> // - create manually (can be re-used for multiple
> bindings though)
> > >> // - create with a helper: Nodes.showing(Node node) ->
> > >> ObservableValue<Boolean>
> > >> // - make it part of the Node class; as the fluent bindings
> > only bind
> > >> themselves
> > >> // to their source when needed (lazy binding), this
> won't create
> > >> overhead
> > >> // for each node in the scene
> > >> }
> > >> Note that this is based on ideas in ReactFX and my own
> experiments in
> > >> https://github.com/hjohn/hs.jfx.eventstream. I've come to the
> > conclusion
> > >> that this is much better directly integrated into JavaFX,
> and I'm
> > hoping
> > >> this proof of concept will be able to move such an effort
> forward.
> > >>
> > >> --John
> > >>
> > >
> >
>
More information about the openjfx-dev
mailing list