Announcing Scala.js 0.6.0-RC2

Jan 23, 2015.

We are excited to announce the second release candidate of Scala.js 0.6.0, aka 0.6.0-RC2! Unless there are blocking issues with this RC, or issues requiring binary incompatible changes, this version will become 0.6.0 final.

Compared to 0.6.0-RC1, this is strictly a bug fix release. The following bugs have been fixed in this second RC:

  • #1451 ScalaDoc run crashes with property @JSExports
  • #1452 PhantomJS sometimes doesn’t shut down properly on Linux
  • #1453 Critical incorrect constant folding of * (-1) in the optimizer
  • #1455 Runs for ScalaDoc complain about @JSExport(SomeFinalVal)
  • #1458 PhantomJS 2 expects a scheme name (file:///) for all urls, not just a path to local files

Scala.js 0.6.0-RC2 is backward and forward binary compatible with 0.6.0-RC1. Libraries already published for RC1 need therefore not be published again against RC2.

To upgrade from 0.6.0-Mx/RC1, simply change the version number. If you’re upgrading from M1/M2, you might need to address some compilation errors regarding the js package, but it should be straightforward. Should you encounter any trouble, do not hesitate to ask.

The rest of this announcement is cumulative with the changes introduced in the milestones and RC1, and is therefore written wrt. 0.5.6.


As the change in “major” version number witnesses, this release is not binary compatible with 0.5.x. Libraries need to be recompiled and republished using 0.6.0-RC1 or RC2 to be compatible. More importantly, this release is not source compatible with 0.5.x either.

Please report any issues on GitHub.

The following libraries and testing frameworks have already been upgraded and published for 0.6.0-RC1/RC2:

  • DOM types: "org.scala-js" %%% "scalajs-dom" % "0.7.0"
  • jQuery types: "be.doeraene" %%% "scalajs-jquery" % "0.7.0"
  • Scala.rx: "com.lihaoyi" %%% "scalarx" % "0.2.7-RC1"
  • ScalaTags: "com.lihaoyi" %%% "scalatags" % "0.4.3-RC1"
  • Autowire: "com.lihaoyi" %%% "autowire" % "0.2.4-RC1"
  • µPickle: "com.lihaoyi" %%% "upickle" % "0.2.6-RC1"
  • µTest: "com.lihaoyi" %%% "utest" % "0.2.5-RC1" % "test"
  • MiniTest: "org.monifu" %%% "minitest" % "0.10" % "test"

Preparations before upgrading from 0.5.x

Upgrade to 0.5.6 if not already done

Before upgrading to 0.6.0-RC2, we strongly recommend that you upgrade to Scala.js 0.5.6, and address all deprecation warnings. Scala.js 0.5.6 contains warnings for the most vicious breaking changes of 0.6.x.

Migrate away from the Scala.js Jasmine test framework

If you use the Jasmine test framework wrapper for Scala.js (scalajs-jasmine-test-framework), migrate away from it to one of the other testing frameworks for Scala.js. The Jasmine test framework wrapper is not a good testing framework for Scala.js code, and is being removed in 0.6.x.

Possible replacements:

Note that these testing frameworks also need to upgrade to 0.6.0-RC1/RC2 before you can use them.

Upgrade to 0.6.0-RC2 from 0.5.6

Basically, you need to apply the same kind of changes to your build files as in this commit, which mostly consists in:

  • Upgrade to sbt >= 0.13.6 (the current version is 0.13.7).
  • Adaptations to new groupId and artifact names for Scala.js packages.
  • Adaptation to the new AutoPlugin infrastructure of the sbt plugin.
  • Drop the prefix ScalaJSKeys. for Scala.js-specific sbt keys, as they are not needed anymore.
  • Upgrade to 0.6.0-RC2-enabled versions of your dependencies.

On the sbt command line, not much changes, except the way you use the fastOpt and fullOpt stages. In Scala 0.5.x, you could run in the fastOpt stage with:

> fastOptStage::run

In 0.6.x, the stage is regulated by the setting scalaJSStage, which is one of:

  • PreLinkStage (default): uses Rhino
  • FastOptStage: fastOpt mode, uses Node.js or PhantomJS
  • FullOptStage: fullOpt mode, uses Node.js or PhantomJS

You can change it from the command line with

> set scalaJSStage := FastOptStage
> run # runs in fastOpt mode

In a multi-project build, you’ll want to change it for all projects, which can be done with in Global:

> set scalaJSStage in Global := FastOptStage

Major changes

This section discusses major changes affecting source compatibility, which may or may not apply to your project.

ClassCastException becomes an undefined behavior

The JVM, in its incommensurable magnanimity, throws nicely specified exceptions when you do something bad with your code. For example, it will nicely throw a ClassCastException if you perform an invalid .asInstanceOf, or an ArithmeticException if you divide an integer by 0.

Since the beginning of time, Scala.js has handled most of these things as undefined behavior, i.e., anything can happen if these cases happen. Until 0.5.x, ClassCastExceptions were properly reported, though. We have found, however, that checking these buggy cases costs up to 100% overhead to the overall execution time of a Scala.js program.

In Scala.js 0.6.x, therefore, invalid casts become an undefined behavior as well. However, the compiler will still be nice with you in the PreLink and FastOpt stages, by throwing an UndefinedBehaviorError if you perform an invalid cast (instead of a ClassCastException). UndefinedBehaviorError is a fatal error, meaning it won’t be caught by case NonFatal(e) handlers. In fullOpt mode, the checks are removed for maximum efficiency.

You must not catch UndefinedBehaviorError, since that would cause your program to behave differently in the fullOpt stage than in the other stages. The idea of UndefinedBehaviorError is that you can enjoy strict checks and stack traces while developing.

If you really want ClassCastExceptions to be thrown reliably (in all stages), you can enable them in your application, at the expense of runtime performance, with the following sbt setting:

scalaJSSemantics ~= { _.withAsInstanceOfs(org.scalajs.core.tools.sem.CheckedBehavior.Compliant) }

This applies to the entire application, including dependencies. There is no way to select parts of the application where this applies, because there is no way to make that sound.

The scala.scalajs.js package has been simplified

We have removed a lot of historical warts from the scala.scalajs.js package, mostly types and APIs with equivalents among normal Scala types and libraries:

  • js.String, js.Boolean, js.Number and js.Undefined have been removed, as well as their js.prim.* equivalent. String, Boolean, Double and Unit should be used instead, respectively.
  • js.parseInt(s) and js.parseFloat(s) should be replaced by s.toInt and s.toDouble, respectively.
  • js.NaN, js.Infinity should be replaced by Double.NaN and Double.PositiveInfinity, respectively.
  • js.isNaN(x) should be replaced by x.isNaN.
  • js.isFinite(x) should be replaced by !x.isNaN && !x.isInfinite.

Methods provided by ECMAScript 5.1 on primitive strings and numbers can be enabled by importing the following implicit conversions:

import js.JSStringOps._
import js.JSNumberOps._

js.native in facade types

When writing facade types, it was previously recommended to use ??? as a fake body for fields and methods. You should now use js.native instead, as in:

trait Foo extends js.Object {
  var bar: Int = js.native
  def foobar(x: Int): String = js.native
}

The compiler will emit a warning if you use any other body. The warning will become an error in 1.0.0.

@JSExport exports to fully qualified names by default

As announced by deprecation warnings in the 0.5.6 compiler, putting @JSExport without an explicit name on an object or class changes meaning between 0.5.x and 0.6.x. Consider this code:

package babar

@JSExport
class Foo

In 0.5.x, Foo is exported as Foo. In 0.6.x, it is exported as babar.Foo instead.

Testing frameworks adaptations

If you are not a testing framework implementor, this section does not apply to you. Please follow the migration guidelines of any testing framework you may use.

Until 0.5.x, Scala.js had a custom, ad-hoc substitute for the sbt testing interface, which allows testing frameworks to integrate with sbt. Although quite good in its own right, it suffered from several limitations, including the inability for one project to use more than one testing framework at the same time, and severe discrepencies with the JVM sbt testing interface. Scala.js 0.6.x now supports its JS version of the original sbt testing interface, with all its power, API, and usability features. We also offer tools to make your testing framework fully source-compatible with the JVM and JS variants of the testing interface, without a single line of platform-specific source code.

An existing barebone cross-compiling testing framework can be found in our tests. Some highlights:

Adapting your testing framework to follow this structure is likely to be the easiest path of migration. You may also want to take a look at the PR we made to uTest to migrate to Scala.js 0.6.x.

Should you run into trouble, don’t hesitate to ask on the mailing list!

Enhancements

Defining cross-compiling projects with crossProject

When writing cross-compiling code, we need to have two separate projects in sbt for the JVM target and the JS target. The new CrossProject type, and its crossProject builder, helps in defining these pairs of projects in a DRY way.

See the documentation of CrossProject for more information and examples.

Faster!

Scala.js 0.6.x benefits from many performance improvements, most notably:

  • asInstanceOfs are unchecked (see above), giving fullOpt code up to twice as fast as before
  • Range.foreach, aka the for (i <- 0 until n) kind of loops, is inlined away, giving the same performance as an explicit while loop.
  • Higher-order operations on js.Arrays and js.Dictionarys (such as foreach, map, etc.) are inlined away as while loops.
  • Various improvements to the optimizer.

Scala collection API for js.Array[A] and js.Dictionary[A]

The title says it all: js.Array[A] and js.Dictionary[A] receive the entire Scala collection API, respectively of mutable.Buffer[A] and mutable.Map[String, A].

js.Array becomes the default implementation of mutable.Buffer, i.e., mutable.Buffer.empty returns a js.Array wrapped in a js.WrappedArray.

Implicits to make “writing JavaScript” easier

Sometimes, for example when porting existing JavaScript code, we want to just “write JavaScript” inside our Scala.js code. A new object js.DynamicImplicits (API) provides implicit conversions that allow to write dynamically typed JavaScriptish code directly in Scala.js with a mimimal amount of boilerplate. Needless to say, these implicits should be handled with care, but they can come in handy.

On-demand strict floats

Scala.js under-specifies Float operations by default, saying that they can sometimes behave as if they were Doubles. In 0.6.x, you can configure your application to use strict-float semantics, guaranteeing that all Float operations behave as on the JVM, with the appropriate truncation of precision (with the notable exception of .toString()). The following sbt setting enables this:

scalaJSSemantics ~= { _.withStrictFloats(true) }

Beware that this can have a major impact on performance on VMs that do not support the Math.fround function.

Obfuscation of class names

The scalaJSSemantics option also allows to obfuscate or otherwise rename the class names in the emitted .js file, as was requested in #1113. For example, this sbt setting empties out all class names in the package my.company:

scalaJSSemantics ~= (_.withRuntimeClassName { linkedClass =>
  val fullName = linkedClass.fullName
  if (fullName.startsWith("my.company.")) ""
  else fullName
})

This changes the value returned by x.getClass.getName or classOf[C].getName.

We publish to Maven Central

This should probably not affect sbt users, but it now becomes possible to imagine a Maven plugin for Scala.js. To this effect, the sbt plugin codebase has also been refactored, and all parts that are not strictly bound to sbt as a build tool have been extracted in Mavenized artifacts. An enthusiast Maven user could therefore build a Maven plugin with relatively few lines of code. As a measurable figure, the code specific to sbt contains only 1,686 lines of code.

Bugfixes

Amongst others, the following bugs have been fixed since 0.5.6:

  • #1430 ClassTag.unapply method (for deconstruction) fails for raw JS classes
  • #1423 String.getBytes returns trailing zeroes
  • #1324 Date.parse should return a Double, not an Int
  • #1349 Auto-completion in runMain task does not work
  • #1192 hashCode for floating points has a very bad distribution
  • #1402 Traversers does not handle the case of Debugger
  • #1451 ScalaDoc run crashes with property @JSExports
  • #1455 Runs for ScalaDoc complain about @JSExport(SomeFinalVal)
  • #1458 PhantomJS 2 expects a scheme name (file:///) for all urls, not just a path to local files