Version 1.0-RC15

Improvements

The most important change for the project is the upgrade to the latest Kotlin version, that is 1.9.22 at the time of this release and gradle 8.5.

• PR #789: Kotlin 1.9.20 & refreshVersions plugin
• PR #841: Refresh Versions for upcoming RC13 Release (Kotlin 1.9.22 for example)
• PR #845: Update Gradle to v8.5
• PR #811: Move docs for new serialization module to docs sites
• PR #817: Update the headless documentation of portalling
• PR #819: Make RootStore.runWithJob public
• PR #822: Improves PopUpPanel Rendering
• PR #823: Improves and enhances the reactive styling
• PR #824: Improve event handling in dataCollection
• PR #826: Improves the validation message filtering and resets the default behaviour
• PR #829: Add Scope handling for Portal Containers
• PR #830, #834: Avoid events in flatMapLatest
• PR #835: Improves PopUpPanel
• PR #836: Improves TicTacToe example
• PR #838: Add UI-Tests to GitHub Build-Process
• PR #839: Add documentation for Stores current property

Fixed Bugs

• PR #847: Fix implicit dependency error in Gradle for publishing
• PR #843: Fix javadoc jar publication for publishing to MavenCentral
• PR #842: Fix Selection Behaviour of ListBox
• PR #812: Fixes WebComponent example
• PR #833: Fix nested Portals
• PR #837: Fix Mountpoint in Portals
• PR #838: Fix some regression in DataCollection

Credits

Special thanks to @jillesvangurp for his amazing work on #789!

Version 1.0-RC12

Improvements

• PR #790: Query parameters for GET
• PR #808: Make ComponentValidationMessage a data class
• PR #804: Removes portal stylings

Fixed Bugs

PR #806 - Proper management of Store Lifecycle (Jobs, Handlers and Flows)

Motivation

Currently there is no real lifecycle-management for Jobs of Handlers and Stores. The initial idea was to hide those details from the user in order to make the API and overall user-experience as easy and pleasant as possible.

Sadly this was a bad idea - not only does it lead to memory leaks, but also to unexpected behaviour due "handledBy" invocations running endlessly. Imagine a delayed change of a store's data initiated by a handler which is no longer "visible", because the Mountpoint in which the handler was called was already destroyed. This is definitely a problem, as it contradicts the reliable reactive behaviour of fritz2.

That being said, especially the latter problem arises very seldomly, which is the reason why we encountered and subsequently adressed this problem so late on the final route to 1.0 version of fritz2.

Nevertheless we have to fix it, so here we go.

Conclusion

There is a need for an explicit lifecycle-management for all fritz2 elements which are related to reactive behaviour such as:

• Stores
• Handlers
• (History - more or less for streamlining reasons. Appears rarely outside of a Store oder WithJob, so there are no behaviour changes)
• (Router - just for streamlining reasons. Appears only once inside an application, so there are no behaviour changes)

Within a RenderContext, those elements should rely on the managed Job of their context to be finished, as the RenderContext gets dropped by any render*-function. This solves the memory leaks as well as the unexpected behaviour of a neverending local store or handledBy call.

Outside of any RenderContext or other receivers that offer some Job, the job-handling should be made explicit. Typically this applies for global stores that hold application states. Those elements are discrete and rather small in number and are intended to run for the whole lifecycle of an application, so there is no harm in creating new jobs for those cases.

Solution

RootStores

Previously, RootStores created their own jobs which were never cancelled. With this PR, every store initialization needs a job: Job parameter.

class RootStore<D>(initialData: D, **job: Job**,  override val id: String = Id.next())
fun <D> storeOf(initialData: D, **job: Job**, id: String = Id.next())

Because the jobs within a RenderContext are already managed, we also added convenience functions which directly use the RenderContext-job if a store is created with the storeOf-factory:

// `WithJob` is a base interface of `RenderContext`
fun <D> WithJob.storeOf(initialData: D, job: Job = this.job, id: String = Id.next())
//                                                 ^^^^^^^^
//                                                 the job is taken from its receiver context

Each store which is created outside of a WithJob-Scope or directly using the constructor needs a user managed Job. The user himself is responsible to take care of the job lifecycle and cancelling the job when it is not needed anymore. We suggest using the job of the surrounding RenderContext, because that one is always properly managed.

In real world applications, it is normal to have some globally defined Stores. Don't be afraid to simply create new Jobs for those without any "management" by yourself. Those stores are intended to run for as long as the application runs. So there is no need to stop or cancel them.
Forever running jobs inside such global stores were a normal occurance before this PR and will remain so after this PR.

No more global handledBy

Previously, handledBy could be called everywhere. In a WithJob-context (e.g. a RenderContext), it used that job, otherwise it created a new job which was endlessly running within the application-lifecycle, which is what might have caused some unexpected side effects.

Now you can only run handledBy within

1. WithJob-scope (a RenderContext implements WithJob)
2. Within a RootStore-Instance

This ensures that the Job used inside of handledBy is always properly managed, which includes both cases:

• it gets cancelled by reactive rendering changes
• it runs forever because it is located inside a global Store which is intended to run forever

The 'handledBy'-functions within a RootStore are protected and can therefore only be used within the RootStore itself or in any derived custom store-implementation. A RootStore as receiver - e.g. using extension functions or apply/run - is not sufficient!

If you explicitely want to use the store-job outside the RootStore, you have to create an extension function with the WithJob receiver and call that function within the RootStore wrapped with the new runWithJob-function.

Example:

object MyStore : RootStore<String>("", Job()){
    init {
        runWithJob{ myFunction() }
    }
}

fun WithJob.myFunction() {
    flowOf("ABC") handledBy MyStore.update
}

Alongside with this change, a handledBy-call will also be interrupted if:

1. The store-job has been cancelled
2. The consumer-job (job of the scope where handledBy was called) has been cancelled

Also, the store's data-Flow will be completed when the store-job has been cancelled to prevent further side effects of the store.

Improve Rendering

When using reactive rendering, e.g. using Flow<T>.render or Flow<T>.renderEach, the render-operation was previously never interrupted if a new flow-value was emitted. The render-operations where queued after each other. This behaviour has changed with this PR. When the Flow emits a new value, the current render-task will be interrupted and it will directly re-render the content with the latest value. This will improve performance.

In rare circumstances, this might also cause different behaviour. In our opinion this could only happen when the reactive rendering is abused for state handling outside of Stores, for example with mutating some var foo outside of the rendering block. Since we consider such implementations bad practise, we recommend to change these constructs.

Migration Guide

Stores

For all global Stores of an application, just add some newly created Job:

// previously
object MyApplicationStore : RootStore<AppState>(AppState(...)) {
}

// now
object MyApplicationStore : RootStore<AppState>(AppState(...), job = Job()) {
}

// previously
val storedAppState = storeOf(AppState())

// now
val storedAppState = storeOf(AppState(), job = Job())

If you encounter a compiler error due to missing job-parameter inside a RenderContext, please have a look at the last section about "Chasing Memory Leaks" where the dangers and solutions as explained in depth.

Global handledBy Calls

You simply have to move those calls inside some Store or some RenderContext - there is no standard solution which fits all situations. You have to decide for yourself where the data handling fits best.

If you have two stores and changes to one should also change the other, consider handleAndEmit as alternate approach or move the call to the dependent store (the one that holds the handler passed to handledBy).

History

The history-function without receiver has been removed. So you either have to move the history code inside some Store (which is the common and recommended approach) or you have to invoke the constructor manually:

// previously
val myHistory = history()

// now
object MyStore : RootStore<MyData>(MyData(), job = Job()) {
    val history = history()
}

// or if really needed outside a store:
val myHistory = History(0, emptyList(), Job())

Router

If you really need to call the Router-constructors manually, you have to provide a Job now:

// previously
val router = Router(StringRoute(""))

// now
val router = Router(StringRoute(""), job = Job())

Consider using the routerOf-factories, they will create a Job automatically.

Chasing Memory Leaks

Besides the previously shown dedicated proposals for fixing compiler errors, we encourage you to scan your code for potential memory-issues we simply cannot prevent by our framework:

Imagine the creation of a Store inside a RenderContext where a new Job() is created:

// previously
fun RenderContext.renderSomething() = div {
    object MyApplicationStore : RootStore<AppState>(AppState(...)) {
    }
}

// now - satisfies compiler but is wrong!
fun RenderContext.renderSomething() = div {
    object MyApplicationStore : RootStore<AppState>(AppState(...), job = Job()) {
    //                                                             ^^^^^^^^^^^
    //                                                             This fixes the compiler error, but in most cases
    //                                                             this is *wrong*. The job-object is never cancelled
    //                                                             -> memory leak!
    }
}

// now - recommended approach
fun RenderContext.renderSomething() = div {
    object MyApplicationStore : RootStore<AppState>(AppState(...), job = this@renderSomething.job) {
    //                                                             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
    //                                                             Reuse job of the surrounding RenderContext
    //                                                             and you will be fine.
    }
}

Last but not least: If you...

Version 1.0-RC11

Improvements

• PR #794: Documentation - Fix and improve some mapping example

Fixed Bugs

• PR #795: Fix beforeUnmount Crashes due to Job Cancellations
• PR #797: Fix renderEach fast deleting Predecessors

Version 1.0-RC10

Fixed Bugs

• PR #793: Fix reactively nested beforeUnmount Calls

Version 1.0-RC9

Fixed Bugs

• PR #792: Semove setFocus from PopUpPanel

Version 1.0-RC8

Improvements

PR #783 / PR #788 - New utility render-functions for RenderContext

PR #783 and PR #788 add new utility render-functions inside RenderContext. @serras initially brought up the idea and implementations to reduce boilerplate code by adding some convenience render-functions for common UI-patterns.

Motivation

Often only some specific state of the data model should be rendered, while for other states the UI artefact should disappear. Typical cases are nullable types, where only for the none null part something should be rendered. The following new functions are added to the fritz2's core to adress those cases.

Please regard: If you ever need to render some UI also for the other state-variants, then just refer to the standard render-function and use explicit case analysis like if-else or when expressions inside the render-block.

New Functions

The renderIf(predicate: (V) -> Boolean) function only renders the given content, when the given predicate returns true for the value inside the Flow<V>.

val store = storeOf<String>("foo")
render {
    store.data.renderIf(String::isNotBlank) {
        p { +it }
    }
}

The renderNotNull() function only renders the given content, when the value inside the Flow<V?> is not null.

val store = storeOf<String?>(null)
render {
    store.data.renderNotNull {
        p { +it }
    }
}

The renderIs(klass: KClass<W>) function only renders the given content, when the value inside the Flow<V> is a type of klass.

interface I
class A: I
class B: I

val store = storeOf<I>(A())
render {
    store.data.renderIs(A::class) {
        p { +"A" }
    }
}

PR #781: Migrate popper.js to Floating-UI and add Portalling

This PR adresses two closely related aspects within our headless components:

• migrate outdated popper.js to the newer Floating-UI (see https://floating-ui.com/docs/migration)
• introduce a portalling mechanism to prevent overlays beeing clipped (see https://floating-ui.com/docs/misc#clipping)

The first one is quite simple to explain: popper.js is outdated and therefor it makes sense to upgrade to the successor project. Floating-UI fits so much better than popper.js for fritz2 and our headless-components, as it is in fact by itself only a headless component. It offers functionality to calculate the position of some floating element, but it does no more render some styling by itself to the DOM. This obviously is great news for our implementations of popup-elements, as the integration removes some pain points and works much smoother together with fritz2's core mechanics!

For the user there are not much changes; the core approach to implement PopUpPanel as base for some floating brick, does not change. It is only about some properties that have different names and sometimes slightly different values. But we gain much more flexibility like the new concept of midlewares, that can be easily added by custom bricks. So in the end the user has more power to control and modify the floating element for his custom needs.

The other aspect of this PR adresses fundamental problems with overlayed elements in general (this includes the headless modals and toast on top of the PopupPanel based components): There will always be a chance for clipping errors, if the overlay is rendered close to the trigger, which often is kinda deep down inside the DOM hierarchy. Please read about in the description of the Floating-UI project.

That said, nothing will change on API level for this aspect. It is a pure implementation aspect, that has changed now: All headless-overlay components will now render their overlay-bricks inside the original target-node of the outer render-function, which acts as a portal-root. The trigger will create only some placeholder node into the DOM, whereas the real overlay content will be rendered inside the portal-root quite upside in the DOM. The placeholder acts as a portal to the final rendering place.

This is a common pattern and so we stick to the wording inside our code and name everything about it with portal prefix.

API-Breaking Remarks and Migration-Guide

Be aware that this PR might be API breaking for the headless package and the components, that implements the PopupPanel-class for some of their bricks. Also your own Implementation that relies on PopupPanel might break.

We do not consider this API breaking for the whole project, so we prefer to present this under the improvement aspect. We have no knowledge about users, using our headless-component yet, so obviously this will affect primarely ourselves.

There is one action needed to keep the headless components working: To use portalling you have to render the portalRoot manually in our your render {} Block like this:

fun main() {
    //...

    render {
        // custom content
        // ...
        
        portalRoot() // should be the last rendered element
    }
}

After this patch, the Headless-Components listBox, menu, modal, popOver, toast and tooltip should be working again.

If you should against all expectations encounter a break, please simply refer to the updated list of properties within the headless documentation. This list and the following explanations should enable you to adapt to the new properties.

Further improvements

• PR #785: Add binary compatibility validation

Fixed Bugs

• PR #778: Fix link to Arrow docs

Credits

Special thanks to @serras for his effords and contributions to this release!

Version 1.0-RC7

Fixed Bugs

• PR #775: Fix a problem where a PopUpPanel's arrow would initially be visible regardless of the popup being hidden.

Version 0.14.6

Improvements

• PR #776: Update everything for latest Kotlin 1.9.0 version

Version 1.0-RC6

Breaking Changes

PR #772: Add Inspector.mapNull() method and fix Store.mapNull() path

This PR changes the behavior of Store.mapNull() so that the path of the derived Store is the same as in the parent Store. This is the correct behavior, as a Store created via mapNull() does not technically map to another hierarchical level of the data model. As a result, mapNull() works the same on both Stores and Inspectors, making the validation process more straight-forward. Previously, the Store's id has been appended to the path upon derivation.

This might potentially be API breaking as the behavior of Store.mapNull() regarding the resulting path changes.

Additionally, it adds an Inspector.mapNull() extension function that works like Store.mapNull().
Just like on a Store, a default value is passed to the method that is used by the resulting store when the parent's value is null.

Improvements

PR #765: Adds convenience execution functions for Unit metadata in validators

For convenience reasons it is now possible to call a validator with metadata type Unitwithout the metadata parameter.

Imagine a Validatorof type Validator<SomeDomain, Unit, SomeMessage>:

data class SomeDomain(...) {
    companion object {
        val validator: Validator<SomeDomain, Unit, SomeMessage> = validation { inspector -> ... }
        //                                   ^^^^
        //                                   no "real" metadata needed
    }
}
val myData: SomeDomain = ...

// now invoke the execution process...

// old
SomeDomain.validator(myData, Unit) // need to pass `Unit`!

// new
SomeDomain.validator(myData) // can omit `Unit`

Fixed Bugs

• PR #767: Fixes an issue with overlapping validation messages of mapped stores
• PR #766: Fix popups in scroll containers/modals
• PR #771: Improves docs

Version 1.0-RC5

Breaking Changes

PR #763: Validation: Remove explicit nullability from metdata type

This PR changes the Validation's metadata type to not be explicitly be nullable. Nullable types are still allowed, however.
The ValidatingStore API has been changed accordingly.

Validation

Before

@JvmInline
value class Validation<D, T, M>(private inline val validate: (Inspector<D>, T?) -> List<M>) {
    operator fun invoke(inspector: Inspector<D>, metadata: T? = null): List<M> = this.validate(inspector, metadata)
    operator fun invoke(data: D, metadata: T? = null): List<M> = this.validate(inspectorOf(data), metadata)
}

Now

@JvmInline
value class Validation<D, T, M>(private inline val validate: (Inspector<D>, T) -> List<M>) {
//                                                                         ^^^
//                                                  Metadata type is no longer explicitly nullable.
//                                                  Thus, it must always be specified.
//
    operator fun invoke(inspector: Inspector<D>, metadata: T): List<M> = this.validate(inspector, metadata)
    operator fun invoke(data: D, metadata: T): List<M> = this.validate(inspectorOf(data), metadata)
}

ValidatingStore

Before

open class ValidatingStore<D, T, M>(
    initialData: D,
    private val validation: Validation<D, T, M>,
    val validateAfterUpdate: Boolean = true,
    override val id: String = Id.next()
) : RootStore<D>(initialData, id) {
    // ...
   
   protected fun validate(data: D, metadata: T? = null): List<M> = /* ... */
}

Now

open class ValidatingStore<D, T, M>(
    initialData: D,
    private val validation: Validation<D, T, M>,
    private val metadataDefault: T,
//  ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
//  New parameter `metadataDefault`: Since the metadata must now be present at
//  all times, a default value needs to bespecified for the automatic validation
//  to work. During manual validation (via `validate(..)`), the metadata can still
//  be passed in as before.
    private val validateAfterUpdate: Boolean = true,
    override val id: String = Id.next()
) : RootStore<D>(initialData, id) {
    // ...
   
   protected fun validate(data: D, metadata: T): List<M> = /* ... */
//                                 ^^^^^^^^^^^
//                                 Metadata now has to be specified at all times
}

Additionally, the convenience factory storeOf(...) has been overloaded so ValidatingStore<D, Unit, M>s can be created without the need to specify Unit as the default metadata value manually.

Migration

• Validation<D, T, M>.invoke(...) now always requires metadata to be present. Add the missing metadata if necessary.
• The Validation<D, T, M> constructor now requires the paramer metadataDefault to be present. Add the missing metadats default if necessary.

PR #761: Inspector based Validation

Motivation

Real world domain objects are in most cases forms a deep object hierarchy by composing dedicated types in a sensefull way, for example some Person consists of fields like name or birthday but also complex fields like some Address, which itself represents some domain aspect with basic fields.

As validation is most of the time tied to its corresponding domain type, the validators mirror the same hierarchy as the domain classes. Thus they must be composable in the same way, the domain types are composed.

This was prior to this PR not supported by the 1.0-RC releases!

Solution

The API of the Validation type changes just a little: The validate lambda expression now provides no longer a (domain) type D, but an Inspector<D>! There are now two invoke functions that take as first parameter an Inspector<D> and as before just a D, which then constructs the inspector itself. So one can use the latter for the (external) call of a validation with the domain object itself and the former for calling a validator from inside another validator!

Remark: If you have used the recommended validation-factory functions, there will be no API breaking at all, as those already have used the Inspector<D>. So it is very unlikely that this change will break existing code!

Example

The following composition now works:

// if you use the `headless` components, prefer to use the `ComponentValidationMessage` instead of handcrafting your own!
data class Message(override val path: String, val text: String) : ValidationMessage {
    override val isError: Boolean = true
}

@Lenses
data class Person(
    val name: String,
    val birthday: LocalDate,
    val address: Address // integrate complex sub-model, with its own business rules
) {

    data class ValidationMetaData(val today: LocalDate, val knownCities: Set<String>)

    companion object {
        val validate: Validation<Person, ValidationMetaData, Message> = validation { inspector, meta ->
            inspector.map(Person.name()).let { nameInspector ->
                if (nameInspector.data.isBlank()) add(Message(nameInspector.path, "Name must not be blank!"))
            }
            inspector.map(Person.birthday()).let { birthdayInspector ->
                if (birthdayInspector.data > meta.today)
                    add(Message(birthdayInspector.path, "Birthday must not be in the future!"))
            }
            // call validator of `Address`-sub-model and pass mapped inspector into it as data source and for
            // creating correct paths!
            // Voilà: Validator Composition achieved!
            addAll(Address.validate(inspector.map(Person.address()), meta.knownCities))
        }
    }
}

@Lenses
data class Address(
    val street: String,
    val city: String
) {
    companion object {
        // enforce business rules for the `Address` domain
        val validate: Validation<Address, Set<String>, Message> = validation { inspector, cities ->
            inspector.map(Address.street()).let { streetInspector ->
                if (streetInspector.data.isBlank()) add(Message(streetInspector.path, "Street must not be blank!"))
            }
            inspector.map(Address.city()).let { cityInspector ->
                if (!cities.contains(cityInspector.data)) add(Message(cityInspector.path, "City does not exist!"))
            }
        }
    }
}

// and then use those:
val fritz = Person(
    "", // must not be empty!
    LocalDate(1712, 1, 24),
    Address("Am Schloss", "Potsdam") // city not in known cities list, see below
)

val errors = Person.validate(
    fritz,
    Person.ValidationMetaData(
        LocalDate(1700, 1, 1), // set "today" into the past
        setOf("Berlin", "Hamburg", "Braunschweig") // remember: no Potsdam inside
    )
)

// three errors would appear:
// Message(".name", "Name must not be blank!")
// Message(".birthday", "Birthday must not be in the future!")
// Message(".address.city", "City does not exist!")

Migration Guide

If you have used the Validation.invoke method directly, then prefer to switch to the dedicated validation-factories!
Inside your validation code just remove the inspectorOf(data) line, that you hopefully will find and change the name of the parameter to inspector.

If you have not used any inspector based validation code, you simple must change the field access in such way:

// inside validation code:
// old
data.someField

// new
inspector.data.someField

Also try to replace handcrafted path parameters of the Message objects by relying on the inspector object:

// old
add(SomeMessage(".someField", ...))

// new
add(SomeMessage(inspector.path, ...))

Improvements

• PR #762: Generate extension functions for Lens-Chaining to enable some fluent-style-API

Fixed Bugs

• PR #764: Fix navigation issue in Router
• PR #755: Fixes List Index related Bug in Headless DataCollection
• PR #752: Substitute tailwindcss class with vanilla CSS