Anvil is a small Java library for creating reactive user interfaces. Originally inspired by React, it suits well as a view layer for MVVM or Redux design patterns.
¹Reflection is still used to inflate views once (standard XML inflater does the same thing, so no performance loss here).
// build.gradle
repositories {
jcenter()
}
dependencies {
compile 'co.trikita:anvil-sdk15:0.5.0'
}
Anvil comes in multiple builds for different minimal SDK versions:
API levels 16, 17, 18, 22 or 23 are not added because they had very few UI-related methods added.
Normally you would write your layouts in XMLs, then get your views by their IDs and set their listeners, finally you would observe your data and modify view properties as your data changes.
Anvil simplifies most of this boring routine.
First, add a static import that makes it much easier to write your view:
import static trikita.anvil.DSL.*;
Next, declare your layout, assign event listeners and bind data:
public int ticktock = 0;
public void onCreate(Bundle b) {
super.onCreate(b);
setContentView(new RenderableView(this) {
@Override
public void view() {
linearLayout(() -> {
size(MATCH, MATCH);
padding(dip(8));
orientation(LinearLayout.VERTICAL);
textView(() -> {
size(MATCH, WRAP);
text("Tick-tock: " + ticktock);
});
button(() -> {
size(MATCH, WRAP);
text("Close");
// Finish current activity when the button is clicked
onClick(v -> finish());
});
});
}
});
}
Here we've created a very simple reactive view and added it to our Activity. We've declared our layout (a LinearLayout with a TextView inside). We've defined styles (width, height, orientation and padding). We've set OnClickListener to the button. We've also bound a variable "ticktock" to the text property inside a TextView.
Next, let's update your views as your data changes:
ticktock++;
Anvil.render();
At this point your layout will be updated and the TextView will contain text "Tick-Tock: 1" instead of "Tick-Tock: 0". However the only actual view method being called would be setText().
You should know that there is no need to call Anvil.render() inside your event listeners, it's already triggered after each UI listener call:
public void view() {
linearLayout(() -> {
textView(() -> {
text("Clicks: " + numClicks);
});
button(() -> {
text("Click me");
onClick(v -> {
numClicks++; // text view will be updated automatically
});
});
});
}
You may find more Anvil examples for Java 6, Java 8 and Kotlin at
No magic. When a renderable object is being constructed there are 3 types of operations: push view, modify some attribute of the current view, and pop view. If you're familiar with incremental DOM - Anvil follows the same approach.
Pushing a view adds it as a child to the parent view from the top of the stack. Attribute modification simply sets the given property to the current view on top of the stack. Pop unwinds the stack.
When you mount this layout (assuming the name
is "John"):
linearLayout(() -> {
textView(() -> {
text("Hello " + name);
});
});
It does the following sequence of actions:
Push LinearLayout (adding it to the root view)
Push TextView (adding it to the linear layout)
Attr text "Hello John" (calling setText of the TextView)
Pop
Pop
The only trick is that these actions are cached into a so called "virtual layout" - a tree-like structure matching the actual layout of views and their properties.
So when you call Anvil.render()
next time it compares the sequence of
actions with the cache and skips property values if they haven't change. Which means on the next
Anvil.render()
call the views will remain untouched. This caching technique
makes render a very quick operation (having a layout of 100 views, 10
attributes each you can do about 4000 render cycles per second!).
Now, if you modify the name
from "John" to "Jane" and call Anvil.render() it will do the following:
Push LinearLayout (noop)
Push TextView (noop)
Attr text "Hello Jane" (comparing with "Hello John" from the pervious render,
noticing the difference and calling setText("Hello Jane") to the TextView)
Pop
Pop
So if you modify one of the variables "bound" to some of the attributes - the cache will be missed and attribute will be updated.
For all event listeners a "proxy" is generated, which delegates its
method calls to your actual event listener, but calls Anvil.render()
after each
method. This is useful, because most of your data models are modified when
the user interacts with the UI, so you write less code without calling
Anvil.render()
from every listener. Remember, no-op renders are very fast.
Anvil is written in Java 7, but its API is designed to use lambdas as well, so in modern times it's recommended to use Anvil with Java8/Retrolambda or Kotlin.
Syntax is a bit different for each language, but it's very intuitive anyway.
Java 6 without lambdas:
public void view() {
o (linearLayout(),
orientation(LinearLayout.VERTICAL),
o (textView(),
text("...")),
o (button(),
text("..."),
onClick(myListener)));
}
Java 8 + RetroLambda:
public void view() {
linearLayout(() -> {
orientation(LinearLayout.VERTICAL);
textView(() -> {
text("....");
});
button(() -> {
text(".....");
onClick(v -> {
....
});
});
});
}
Kotlin:
public override fun view() {
linearLayout {
orientation(LinearLayout.VERTICAL)
textView {
text("....")
}
button {
text("....")
onClick { v ->
...
}
}
}
}
Anvil library contains only a few classes to work with the virtual layout, but most
of the DSL (domain-specific language describing how to create views/layouts and set
their attributes) is generated from android.jar
.
Here's a list of classes and methods you need to know to work with Anvil like a pro:
Anvil.Renderable
- functional interface that one should implement to
describe layout structure, style and data bindings.
Anvil.mount(View, Anvil.Renderable)
- mounts renderable layout into a
View or a ViewGroup
Anvil.unmount(View)
- unmounts renderable layout from the View
removing all its child views if it's a ViewGroup
Anvil.render()
- starts a new render cycle updating all mounted views
Anvil.currentView(): View
- returns the view which is currently being
rendered. Useful in some very rare cases and only inside the Renderable's
method view()
to get access to the real view and modifying it manually.
RenderableView
- a most typical implementation of Anvil.Renderable.
Extending this class and overriding its method view()
allows to create
self-contained reactive components that are mounted and unmounted
automatically.
RenderableAdapter
- extending this class and overriding its getCount()
,
getItem(int)
and view(int)
allows to create lists where each item is a
standalone reactive renderable object.
RenderableAdapter.withItems(list, cb(index, item))
- a shortcut to create
simple adapters for the given list of items. cb
is a lambda that describes
the layout and bindings of a certain list item at the given index.
The bottom part of the iceberg is Anvil DSL.
DSL consists of a few handwritten property setters, but most of it is generated from java classes in the android SDK.
See a full list of the DSL methods for each API level here.
The setters are named as the view methods from Android SDK, but without the "set" prefix. E.g. "text(s)" instead of "setText(s)", "backgroundDrawable(d)" instead of "setBackgroundDrawable(d)" and so on.
Event listeners also have names from the Android SDK, but without the "set" prefix and the "Listener" suffix, e.g. "onClick" instead of "setOnClickListener".
For LayoutParams the bindings can't be generated easily, so it was faster to write them manually:
size(width, height)
- set width and height. Special constants like WRAP
, FILL
and MATCH
are available.dip(x)
- returns the value in pixels for the dimension in density-independent
pixels. Often used with size, padding or margin properties.margin(m)
, margin(h, v)
, margin(l, t, r, b)
- set view margin for all 4
sides, for horizontal/vertical dimensions or for each side individually.weight(w)
- modifies view layout weight.layoutGravity(g)
- modifies layout gravity of a view. Common constants like
START
, END
, CENTER
, CENTER_VERTICAL
, etc. are available.align(verb, anchor)
, align(verb)
- base functions for relative layout params.above(id)
, alignBaseline(id)
, alignBottom(id)
, alignEnd(id)
, alignLeft(id)
, alignParentBottom()
,
alignParentEnd()
, alignParentLeft()
, alignParentRight()
, alignParentStart()
, alignParentTop()
,
alignRight(id)
, alignTop(id)
, below(id)
, centerHorizontal()
, centerVertical()
, centerInParent()
,
toEndOf(id)
, toLeftOf(id)
, toRightOf(id)
, toStartOf(id)
- all possible settings for relative layout paramsA few bindings have been written for other use cases which we find useful:
init(Runnable)
- executes a runnable once, useful to initialize custom views (see also Anvil.currentView()
).R()
- returns a Resources
object associated with the current view. Useful for
multiple screen support (sizes, dpi, orientation etc).isPortrait()
- returns true if a screen is portrait-oriented. Useful for
tweaking layouts for different orientations.typeface(font)
- loads font from assets by its file name and sets the typeface to
a TextViewpadding(p)
, padding(h, v)
, padding(l, t, r, b)
- set view padding for all 4
sides, for horizontal/vertical dimensions or for each side individually.visibility(flag)
- sets the visibility property to View.VISIBLE
or View.GONE
depending on
the flag
boolean valueshadowLayer(radius, dx, dy, color)
- sets shadow layer of a TextViewonTextChanged(textWatcher)
- binds a text watcher to an EditText
. No
Anvil.render()
is called in this case, because you're likely to get an infinite
recursion.text(StringBuilder)
- binds a string builder to the edit text, so when you
change its contents - the edit text is changed, and if you type something
manually - the string builder gets modified. So far it's the only two-way
data binding, becayse TextWatcher is a complicated beast.onItemSelected(lambda)
- accepts a functional interface to handle a Spinner
events. onNothingSelected()
method is omitted, because it's rarely used anyway.If a binding you need is not in the list - please, check issue #27 and report it there.
A special case for animations is added:
anim(trigger, Animator)
- starts animation when trigger
is true, cancels it
when the trigger
becomes false.Finally, a few low-level DSL functions are there, which you would no need unless you want to write your own property setters or custom view builders:
v(class, attrs...)
- pushes view, applies attributes, doesn't pop the view.o()
, x()
- names that look like bullets, actually pop the view. These are used in Java 6 syntax.v(class, renderable)
- pushes the view, applies the renderable to fulfil
attributes and child views, pops the view. This is used in Java 8 and Kotlin
syntax.attr(func, value)
- checks the cache for the given value of the given
property setter function. Often used to create your own property setter
binding.If you're migrating an existing project to Anvil or if you prefer to keep your view layouts declared in XML - you can do so:
public override fun view() {
// A designer gave us an XML with some fancy layout:
// a viewgroup with a button and a progress bar in it
xml(R.layout.my_layout) {
backgroundColor(Settings.bgColor) // will modify root layout view color
withId(R.id.my_button) {
// button state may depend on some variable
enabled(isMyButtonEnabled)
// button click listener can be attached
onClick { v ->
...
}
}
withId(R.id.my_progress_bar) {
visible(isMyProgressBarShown)
progress(someProgressValue)
}
}
}
Here xml()
creates a view node, much like frameLayout()
or textView()
,
except for it uses an XML file to inflate the views, not the direct view class
constructor. Much like view "builders", xml()
takes a renderable lambda as a
parameter and uses that to modify the created view.
Any attribute setters will affect the root view from the XML layout.
If you want to modify attributes of the child views from the XML - you should
use withId()
to assign a renderable to a view with the given ID. You may not
follow the hierarchy of the child views, e.g. all your withId()
calls may be
nested as the views are nested in the XML, or may be direct children of the
xml()
renderable. xml()
calls can be nested, too.
xml()
also allows you to create views that can not be inflated from Java code
but only from XML (such as hostozintal progress bars) or any other views where
AttributeSet must be given or the views that rely on the onFinishInflate
method to be called.
withId()
call is not limited to XML views, it can be used for views declared
in code that have IDs. So if you have a custom viewgroup that creates its child
views automatically and assigns some IDs to them - you can still modify their
properties using withId()
. Also, if any views are created inside the
withId()
or xml()
- they will be appeneded to the view group:
v(MyComponent.java, () -> {
withId(R.id.child_view, () -> {
// R.id.child_view was implicitely created in MyComponent's constructor
// but we still can modify it here
});
textView(() -> {
// This textView will be appeneded to MyComponent layout
});
});
Code is distributed under MIT license, feel free to use it in your proprietary projects as well.