Starting with 2.10, -Xlint complains about classes defined inside of package objects. But why? Defining a class inside a package object should be exactly equivalent to defining the classes inside of a separate package with the same name, except a lot more convenient.
In my opinion, one of the serious design flaws in Scala is the inability to put anything other than a class-like entity (e.g. variable declarations, function definitions) at top level of a file. Instead, you're forced to put them into a separate ''package object'' (often in package.scala), separate from the rest of the code that they belong with and violating a basic programming rule which is that conceptually related code should be physically related as well. I don't see any reason why Scala can't conceptually allow anything at top level that it allows at lower levels, and anything non-class-like automatically gets placed into the package object, so that users never have to worry about it.
For example, in my case I have a util package, and under it I have a number of subpackages (util.io, util.text, util.time, util.os, util.math, util.distances, etc.) that group heterogeneous collections of functions, classes and sometimes variables that are semantically related. I currently store all the various functions, classes, etc. in a package object sitting in a file called io.scala or text.scala or whatever, in the util directory. This works great and it's very convenient because of the way functions and classes can be mixed, e.g. I can do something like:
package object math {
// Coordinates on a sphere
case class SphereCoord(lat: Double, long: Double) { ... }
// great-circle distance between two points
def spheredist(a: SphereCoord, b: SphereCoord) = ...
// Area of rectangle running along latitude/longitude lines
def rectArea(topleft: SphereCoord, botright: SphereCoord) = ...
// ...
// ...
// Exact-decimal functions
class DecimalInexactError extends Exception
// Format floating point value in decimal, error if can't do exactly
formatDecimalExactly(val num: Double) = ...
// ...
// ...
}
Without this, I would have to split the code up inconveniently according to fun vs. class rather than by semantics. The alternative, I suppose, is to put them in a normal object -- kind of defeating the purpose of having package objects in the first place.
But why? Defining a class inside a package object should be exactly equivalent to defining the classes inside of a separate package with the same name,
Precisely. The semantics are (currently) the same, so if you favor defining a class inside a package object, there should be a good reason. But the reality is that there is at least one good reason no to (keep reading).
except a lot more convenient
How is that more convenient?
If you are doing this:
package object mypkg {
class MyClass
}
You can just as well do the following:
package mypkg {
class MyClass
}
You'll even save a few characters in the process :)
Now, a good and concrete reason not to go overboard with package objects is that while packages are open, package objects are not.
A common scenario would be to have your code dispatched among several projects, with each project defining classes in the same package. No problem here.
On the other hand, a package object is (like any object) closed (as the spec puts it "There can be only one package object per package"). In other words,
you will only be able to define a package object in one of your projects.
If you attempt to define a package object for the same package in two distinct projects, bad things will happen, as you will effectively end up with two
distinct versions of the same JVM class (n our case you would end up with two "mypkg.class" files).
Depending on the cases you might end up with the compiler complaining that it cannot find something that you defined in the first version of your package object,
or get a "bad symbolic reference" error, or potentially even a runtime error. This is a general limitation of package objects, so you have to be aware of it.
In the case of defining classes inside a package object, the solution is simple: don't do it (given that you won't gain anything substantial compared to just defining the class as a top level).
For type aliase, vals and vars, we don't have such a luxuary, so in this case it is a matter of weighing whether the syntactic convenience (compared to defining them in an object) is worth it, and then take care not to define duplicate package objects.
I have not found a good answer to why this semantically equivalent operation would generate a lint warning. Methinks this is a lint bug. The only thing that I have found that must not be placed inside a package object (vs inside a plain package) is an object that implements main (or extends App).
Note that -Xlint also complains about implicit classes declared inside package objects, even though they cannot be declared at package scope. (See http://docs.scala-lang.org/overviews/core/implicit-classes.html for the rules on implicit classes.)
I figured out a trick that allows for all the benefits of package objects without the complaints about deprecation. In place of
package object foo {
...
}
you can do
protected class FooPackage {
...
}
package object foo extends FooPackage { }
Works the same but no complaint. Clear sign that the complaint itself is bogus.
Related
I created a class extend scala.Immutable
class SomeThing(var string: String) extends Immutable {
override def toString: String = string
}
As I expected, scala compiler should help me prevent change state of class SomeThing. But when I run this test
"Test change state of immutable interface" should "not allow" in {
val someThing = new SomeThing("hello")
someThing.string = "hello 1"
println(someThing)
}
The result is hello 1 and scala compiler don't throw any warning or error.
Why they have to add Immutable trait without help us prevent object mutable?
There are several aspects to this question.
1. A simple one is that Scala compiler can't really ensure immutability for many various reasons. For example, the main target platform JVM allows modifying even final fields using reflection. Another reason this is not enforceable is code like this
/////////////////////////////////////////
//// library v1
package library
class LibraryData(val value:Int)
/////////////////////////////////////////
//// code that uses the library
package app
class UserData(val data:LibraryData) extends Immutable
/////////////////////////////////////////
//// library v2
package library
class LibraryData(var value:Int) //now change it to var!
Since the "library" is compiled independently of the "app" and doesn't even know about existence of the "app" there is no point in time where compiler can catch the broken contract.
2. More fundamental misunderstanding you seem to have is what trait does. In this context trait (or "interface" in some other languages) represents a contract between the implementation and the user-code about how the implementation can and should behave. However not every kind of a contract can be represented as a trait (at least without making the code super-complicated). For example, for a mutable collection there is a contract that size should return the number of times add (or +=) has been called but there is no way to represent such a contract as a trait besides declaring that there are methods size and += with corresponding signatures. On the other hand, for most of the contracts there is no way to enforce implementation to follow the contract . For example, an implementation of size that always returns 0 technically matches all the types but is clearly breaking the contract.
Similarly Immutable doc says:
A marker trait for all immutable data structures such as immutable collections.
So it is just a marker trait which is one of the ways to work around contracts that can't be really represented as types. And it says that whoever implements that trait claims to be an immutable object. Your code claims that but clearly breaks the contract. So technically it is your fault for not respecting the contract.
I'm a relatively new Scala user and I wanted to get an opinion on the current design of my code.
I have a few classes that are all represented as fixed length Vector[Byte] (ultimately they are used in a learning algorithm that requires a byte string), say A, B and C.
I would like these classes to be referred to as A, B and C elsewhere in the package for readability sake and I don't need to add any extra class methods to Vector for these methods. Hence, I don't think the extend-my-library pattern is useful here.
However, I would like to include all the useful functional methods that come with Vector without having to 'drill' into a wrapper object each time. As efficiency is important here, I also didn't want the added weight of a wrapper.
Therefore I decided to define type aliases in the package object:
package object abc {
type A: Vector[Byte]
type B: Vector[Byte]
type C: Vector[Byte]
}
However, each has it's own fixed length and I would like to include factory methods for their creation. It seems like this is what companion objects are for. This is how my final design looks:
package object abc {
type A: Vector[Byte]
object A {
val LENGTH: Int = ...
def apply(...): A = {
Vector.tabulate...
}
}
...
}
Everything compiles and it allows me to do stuff like this:
val a: A = A(...)
a map {...} mkString(...)
I can't find anything specifically warning against writing companion objects for type aliases, but it seems it goes against how type aliases should be used. It also means that all three of these classes are defined in the same file, when ideally they should be separated.
Are there any hidden problems with this approach?
Is there a better design for this problem?
Thanks.
I guess it is totally ok, because you are not really implementing a companion object.
If you were, you would have access to private fields of immutable.Vector from inside object A (like e.g. private var dirty), which you do not have.
Thus, although it somewhat feels like A is a companion object, it really isn't.
If it were possible to create a companion object for any type by using type alias would make member visibility constraints moot (except maybe for private|protected[this]).
Furthermore, naming the object like the type alias clarifies context and purpose of the object, which is a plus in my book.
Having them all in one file is something that is pretty common in scala as I know it (e.g. when using the type class pattern).
Thus:
No pitfalls, I know of.
And, imho, no need for a different approach.
A very strange thing indeed. I have the following project structure:
myproject/one/two
Inside package myproject I have a class:
abstract class A (two: Buffer[Int])
and then, inside package one I have:
object B extends A (Buffer[Int](1, 2, 3)) {
val с = two.map(_ + 1) // ERROR
}
However, the erros says:
object map is not a member of package
myproject/one/two
which is obviously erroneous because it should be perfectly clear that I don't refer to the packages here, but to the local variable... And two also is not shown in context-assist after this. in B, but is shown in A (Scala-IDE). Is this an intended behavior and I am doing something wrong or is it a bug?
UPDATE:
(simultaneously suggested by Nicolas :D ) Been able to resolve the name collision by specifying two as val (making it public). I did not notice at first, but it was private and unavailable in the successor class. Nevertheless I am still wondering, why and how did Scala pick up a package instead of saying that the variable does not exist or is not accessible?
It's not as clear as you might think. Without a modifier, two is private to abstract class A class A. Thus your declaration of a is equivalent to abstract class A (private[this] A). It means that field two can't be seen from object B. A direct consequence is that the compiler look into the only defiition of two visible from B: the package two.
Imagine this:
val myObject = if(someCondition) {
new Whatever with Trait1
} else if(otherCondition) {
new Whatever with Trait2 with Trait3 with Trait4
} else {
new Whatever with Trait5
}
Is the myObject object "composed" at runtime, or is the scala compiler smart enough to generate the appropriate code at compile time? What kind of performance impact will it have on the code if you have multiple places that are applying traits like in the above code?
It's composed at compile-time
The traits will be added as interfaces to the resulting type, and any concrete methods from those traits will (usually) be copied to the class in their entirety.
Occasionally, the compiler may have to provide concrete implementations via forwarders to static methods, but this isn't usually the case.
Scala will create three anonymous classes, (except the last condition is a syntax error).
Note: These classes will be named using the order in which they are defined in the scope they are defined. So... OuterClass$anon$1 -> 3. Please avoid using these anonymous classes in any long-term Java-serialization as this locks down the order of anonymous classes in your code.
Other than that, it's an awesome convenience feature!
Ok, I'll explain why I ask this question. I begin to read Lift 2.2 source code these days.
It's good if you happened to read lift source code before.
In Lift, I found that, define inner class and inner trait are very heavily used.
object Menu has 2 inner traits and 4 inner classes. object Loc has 18 inner classes, 5 inner traits, 7 inner objects.
There're tons of codes write like this. I wanna to know why the author write like this.
Is it because it's the author's
personal taste or a powerful use of
language feature?
Is there any trade-off for this kind
of usage?
Before 2.8, you had to choose between packages and objects. The problem with packages is that they cannot contain methods or vals on their own. So you have to put all those inside another object, which can get awkward. Observe:
object Encrypt {
private val magicConstant = 0x12345678
def encryptInt(i: Int) = i ^ magicConstant
class EncryptIterator(ii: Iterator[Int]) extends Iterator[Int] {
def hasNext = ii.hasNext
def next = encryptInt(ii.next)
}
}
Now you can import Encrypt._ and gain access to the method encryptInt as well as the class EncryptIterator. Handy!
In contrast,
package encrypt {
object Encrypt {
private[encrypt] val magicConstant = 0x12345678
def encryptInt(i: Int) = i ^ magicConstant
}
class EncryptIterator(ii: Iterator[Int]) extends Iterator[Int] {
def hasNext = ii.hasNext
def next = Encrypt.encryptInt(ii.next)
}
}
It's not a huge difference, but it makes the user import both encrypt._ and encrypt.Encrypt._ or have to keep writing Encrypt.encryptInt over and over. Why not just use an object instead, as in the first pattern? (There's really no performance penalty, since nested classes aren't actually Java inner classes under the hood; they're just regular classes as far as the JVM knows, but with fancy names that tell you that they're nested.)
In 2.8, you can have your cake and eat it too: call the thing a package object, and the compiler will rewrite the code for you so it actually looks like the second example under the hood (except the object Encrypt is actually called package internally), but behaves like the first example in terms of namespace--the vals and defs are right there without needing an extra import.
Thus, projects that were started pre-2.8 often use objects to enclose lots of stuff as if they were a package. Post-2.8, one of the main motivations has been removed. (But just to be clear, using an object still doesn't hurt; it's more that it's conceptually misleading than that it has a negative impact on performance or whatnot.)
(P.S. Please, please don't try to actually encrypt anything that way except as an example or a joke!)
Putting classes, traits and objects in an object is sometimes required when you want to use abstract type variables, see e.g. http://programming-scala.labs.oreilly.com/ch12.html#_parameterized_types_vs_abstract_types
It can be both. Among other things, an instance of an inner class/trait has access to the variables of its parent. Inner classes have to be created with a parent instance, which is an instance of the outer type.
In other cases, it's probably just a way of grouping closely related things, as in your object example. Note that the trait LocParam is sealed, which means that all subclasses have to be in the same compile unit/file.
sblundy has a decent answer. One thing to add is that only with Scala 2.8 do you have package objects which let you group similar things in a package namespace without making a completely separate object. For that reason I will be updating my Lift Modules proposal to use a package object instead of a simple object.