In scala, the following 2 functions serve exactly the same purpose:
#tailrec
final def fn(str: String): Option[String] = {
Option(str).filter(_.nonEmpty).flatMap { v =>
fn(v.drop(1))
}
}
#tailrec
final def fn2(str: String): Option[String] = {
Option(str).filter(_.nonEmpty) match {
case None => None
case Some(v) => fn2(v.drop(1))
}
}
However #tailrec only works in second case, in the first case it will generate the following error:
Error: could not optimize #tailrec annotated method fn: it contains a
recursive call not in tail position
Option(str).filter(_.nonEmpty).flatMap { v =>
Why this error was given? And why these 2 codes generate different kinds JVM bytecode
For fn to be tail-recursive, the recursive call must be the last action in the function. If you pass fn to another function such as flatMap then the other function is free to perform other actions after calling fn and therefore the compiler cannot be sure that it is tail recursive.
In some cases the compiler could detect that calling fn is the last action in the other function, but not in the general case. And this would rely on a specific implementation of that other function so the tailrec annotation might become invalid if that other function were changed, which is an undesirable dependency.
Specifically for the last question:
And why these 2 codes generate different kinds JVM bytecode
Because on JVM there's no guarantee that the JAR containing Option class at runtime is the same as was seen at compile-time. This is good, because otherwise even minor versions of libraries (including standard Java and Scala libraries) would be incompatible, and you'd need all dependencies to be using the same minor version of their common dependencies.
If that class doesn't have a suitable flatMap method, you'll get AbstractMethodError, but otherwise semantics of Scala require that its flatMap method must be called. So the compiler has to emit bytecode to actually call the method.
Kotlin works around this by using inline functions and Scala 3 will support them too, but I don't know if it'll use them for such cases.
Consider the following:
List('a', 'b').flatMap(List(_,'g')) //res0: List[Char] = List(a, g, b, g)
I seems pretty obvious that flatMap() is doing some internal post-processing in order to achieve that result. How else would List('a','g') get combined with List('b','g')?
Scala seems to behave like Java when it comes to the magic conversion of primitives:
val a: Int = 1
val b: Double = 2.3
println(a + b) // 3.3
println(Math.max(a, b)) // 2.3
More often than not, this has been a source of bugs in my code. Is there a way to disable these implicit conversions so that my example give a compilation warnning/error? I would really rather have to write
print(a.toDouble + b)
println(Math.max(a.toDouble, b))
every single time I need such conversions.
Use WartRemover. A wart like that isn't built-in, but could be written (see "Writing Wart Rules" in README). Though now that I think, it's probably more work than I thought initially.
scalac also has -Ywarn-numeric-widen option (together with -Xfatal-warnings to turn the warnings to errors), but I don't know if there are any cases not covered by it.
Closed. This question is opinion-based. It is not currently accepting answers.
Want to improve this question? Update the question so it can be answered with facts and citations by editing this post.
Closed 4 years ago.
Improve this question
So far implicit parameters in Scala do not look good for me -- it is too close to global variables, however since Scala seems like rather strict language I start doubting in my own opinion :-).
Question: could you show a real-life (or close) good example when implicit parameters really work. IOW: something more serious than showPrompt, that would justify such language design.
Or contrary -- could you show reliable language design (can be imaginary) that would make implicit not neccessary. I think that even no mechanism is better than implicits because code is clearer and there is no guessing.
Please note, I am asking about parameters, not implicit functions (conversions)!
Updates
Global variables
Thank you for all great answers. Maybe I clarify my "global variables" objection. Consider such function:
max(x : Int,y : Int) : Int
you call it
max(5,6);
you could (!) do it like this:
max(x:5,y:6);
but in my eyes implicits works like this:
x = 5;
y = 6;
max()
it is not very different from such construct (PHP-like)
max() : Int
{
global x : Int;
global y : Int;
...
}
Derek's answer
This is great example, however if you can think of as flexible usage of sending message not using implicit please post an counter-example. I am really curious about purity in language design ;-).
In a sense, yes, implicits represent global state. However, they are not mutable, which is the true problem with global variables -- you don't see people complaining about global constants, do you? In fact, coding standards usually dictate that you transform any constants in your code into constants or enums, which are usually global.
Note also that implicits are not in a flat namespace, which is also a common problem with globals. They are explicitly tied to types and, therefore, to the package hierarchy of those types.
So, take your globals, make them immutable and initialized at the declaration site, and put them on namespaces. Do they still look like globals? Do they still look problematic?
But let's not stop there. Implicits are tied to types, and they are just as much "global" as types are. Does the fact that types are global bother you?
As for use cases, they are many, but we can do a brief review based on their history. Originally, afaik, Scala did not have implicits. What Scala had were view types, a feature many other languages had. We can still see that today whenever you write something like T <% Ordered[T], which means the type T can be viewed as a type Ordered[T]. View types are a way of making automatic casts available on type parameters (generics).
Scala then generalized that feature with implicits. Automatic casts no longer exist, and, instead, you have implicit conversions -- which are just Function1 values and, therefore, can be passed as parameters. From then on, T <% Ordered[T] meant a value for an implicit conversion would be passed as parameter. Since the cast is automatic, the caller of the function is not required to explicitly pass the parameter -- so those parameters became implicit parameters.
Note that there are two concepts -- implicit conversions and implicit parameters -- that are very close, but do not completely overlap.
Anyway, view types became syntactic sugar for implicit conversions being passed implicitly. They would be rewritten like this:
def max[T <% Ordered[T]](a: T, b: T): T = if (a < b) b else a
def max[T](a: T, b: T)(implicit $ev1: Function1[T, Ordered[T]]): T = if ($ev1(a) < b) b else a
The implicit parameters are simply a generalization of that pattern, making it possible to pass any kind of implicit parameters, instead of just Function1. Actual use for them then followed, and syntactic sugar for those uses came latter.
One of them is Context Bounds, used to implement the type class pattern (pattern because it is not a built-in feature, just a way of using the language that provides similar functionality to Haskell's type class). A context bound is used to provide an adapter that implements functionality that is inherent in a class, but not declared by it. It offers the benefits of inheritance and interfaces without their drawbacks. For example:
def max[T](a: T, b: T)(implicit $ev1: Ordering[T]): T = if ($ev1.lt(a, b)) b else a
// latter followed by the syntactic sugar
def max[T: Ordering](a: T, b: T): T = if (implicitly[Ordering[T]].lt(a, b)) b else a
You have probably used that already -- there's one common use case that people usually don't notice. It is this:
new Array[Int](size)
That uses a context bound of a class manifests, to enable such array initialization. We can see that with this example:
def f[T](size: Int) = new Array[T](size) // won't compile!
You can write it like this:
def f[T: ClassManifest](size: Int) = new Array[T](size)
On the standard library, the context bounds most used are:
Manifest // Provides reflection on a type
ClassManifest // Provides reflection on a type after erasure
Ordering // Total ordering of elements
Numeric // Basic arithmetic of elements
CanBuildFrom // Collection creation
The latter three are mostly used with collections, with methods such as max, sum and map. One library that makes extensive use of context bounds is Scalaz.
Another common usage is to decrease boiler-plate on operations that must share a common parameter. For example, transactions:
def withTransaction(f: Transaction => Unit) = {
val txn = new Transaction
try { f(txn); txn.commit() }
catch { case ex => txn.rollback(); throw ex }
}
withTransaction { txn =>
op1(data)(txn)
op2(data)(txn)
op3(data)(txn)
}
Which is then simplified like this:
withTransaction { implicit txn =>
op1(data)
op2(data)
op3(data)
}
This pattern is used with transactional memory, and I think (but I'm not sure) that the Scala I/O library uses it as well.
The third common usage I can think of is making proofs about the types that are being passed, which makes it possible to detect at compile time things that would, otherwise, result in run time exceptions. For example, see this definition on Option:
def flatten[B](implicit ev: A <:< Option[B]): Option[B]
That makes this possible:
scala> Option(Option(2)).flatten // compiles
res0: Option[Int] = Some(2)
scala> Option(2).flatten // does not compile!
<console>:8: error: Cannot prove that Int <:< Option[B].
Option(2).flatten // does not compile!
^
One library that makes extensive use of that feature is Shapeless.
I don't think the example of the Akka library fits in any of these four categories, but that's the whole point of generic features: people can use it in all sorts of way, instead of ways prescribed by the language designer.
If you like being prescribed to (like, say, Python does), then Scala is just not for you.
Sure. Akka's got a great example of it with respect to its Actors. When you're inside an Actor's receive method, you might want to send a message to another Actor. When you do this, Akka will bundle (by default) the current Actor as the sender of the message, like this:
trait ScalaActorRef { this: ActorRef =>
...
def !(message: Any)(implicit sender: ActorRef = null): Unit
...
}
The sender is implicit. In the Actor there is a definition that looks like:
trait Actor {
...
implicit val self = context.self
...
}
This creates the implicit value within the scope of your own code, and it allows you to do easy things like this:
someOtherActor ! SomeMessage
Now, you can do this as well, if you like:
someOtherActor.!(SomeMessage)(self)
or
someOtherActor.!(SomeMessage)(null)
or
someOtherActor.!(SomeMessage)(anotherActorAltogether)
But normally you don't. You just keep the natural usage that's made possible by the implicit value definition in the Actor trait. There are about a million other examples. The collection classes are a huge one. Try wandering around any non-trivial Scala library and you'll find a truckload.
One example would be the comparison operations on Traversable[A]. E.g. max or sort:
def max[B >: A](implicit cmp: Ordering[B]) : A
These can only be sensibly defined when there is an operation < on A. So, without implicits we’d have to supply the context Ordering[B] every time we’d like to use this function. (Or give up type static checking inside max and risk a runtime cast error.)
If however, an implicit comparison type class is in scope, e.g. some Ordering[Int], we can just use it right away or simply change the comparison method by supplying some other value for the implicit parameter.
Of course, implicits may be shadowed and thus there may be situations in which the actual implicit which is in scope is not clear enough. For simple uses of max or sort it might indeed be sufficient to have a fixed ordering trait on Int and use some syntax to check whether this trait is available. But this would mean that there could be no add-on traits and every piece of code would have to use the traits which were originally defined.
Addition:
Response to the global variable comparison.
I think you’re correct that in a code snipped like
implicit val num = 2
implicit val item = "Orange"
def shopping(implicit num: Int, item: String) = {
"I’m buying "+num+" "+item+(if(num==1) "." else "s.")
}
scala> shopping
res: java.lang.String = I’m buying 2 Oranges.
it may smell of rotten and evil global variables. The crucial point, however, is that there may be only one implicit variable per type in scope. Your example with two Ints is not going to work.
Also, this means that practically, implicit variables are employed only when there is a not necessarily unique yet distinct primary instance for a type. The self reference of an actor is a good example for such a thing. The type class example is another example. There may be dozens of algebraic comparisons for any type but there is one which is special.
(On another level, the actual line number in the code itself might also make for a good implicit variable as long as it uses a very distinctive type.)
You normally don’t use implicits for everyday types. And with specialised types (like Ordering[Int]) there is not too much risk in shadowing them.
Based on my experience there is no real good example for use of implicits parameters or implicits conversion.
The small benefit of using implicits (not needing to explicitly write a parameter or a type) is redundant in compare to the problems they create.
I am a developer for 15 years, and have been working with scala for the last 1.5 years.
I have seen many times bugs that were caused by the developer not aware of the fact that implicits are used, and that a specific function actually return a different type that the one specified. Due to implicit conversion.
I also heard statements saying that if you don't like implicits, don't use them.
This is not practical in the real world since many times external libraries are used, and a lot of them are using implicits, so your code using implicits, and you might not be aware of that.
You can write a code that has either:
import org.some.common.library.{TypeA, TypeB}
or:
import org.some.common.library._
Both codes will compile and run.
But they will not always produce the same results since the second version imports implicits conversion that will make the code behave differently.
The 'bug' that is caused by this can occur a very long time after the code was written, in case some values that are affected by this conversion were not used originally.
Once you encounter the bug, its not an easy task finding the cause.
You have to do some deep investigation.
Even though you feel like an expert in scala once you have found the bug, and fixed it by changing an import statement, you actually wasted a lot of precious time.
Additional reasons why I generally against implicits are:
They make the code hard to understand (there is less code, but you don't know what he is doing)
Compilation time. scala code compiles much slower when implicits are used.
In practice, it changes the language from statically typed, to dynamically typed. Its true that once following very strict coding guidelines you can avoid such situations, but in real world, its not always the case. Even using the IDE 'remove unused imports', can cause your code to still compile and run, but not the same as before you removed 'unused' imports.
There is no option to compile scala without implicits (if there is please correct me), and if there was an option, none of the common community scala libraries would have compile.
For all the above reasons, I think that implicits are one of the worst practices that scala language is using.
Scala has many great features, and many not so great.
When choosing a language for a new project, implicits are one of the reasons against scala, not in favour of it. In my opinion.
Another good general usage of implicit parameters is to make the return type of a method depend on the type of some of the parameters passed to it. A good example, mentioned by Jens, is the collections framework, and methods like map, whose full signature usually is:
def map[B, That](f: (A) ⇒ B)(implicit bf: CanBuildFrom[GenSeq[A], B, That]): That
Note that the return type That is determined by the best fitting CanBuildFrom that the compiler can find.
For another example of this, see that answer. There, the return type of the method Arithmetic.apply is determined according to a certain implicit parameter type (BiConverter).
It's easy, just remember:
to declare the variable to be passed in as implicit too
to declare all the implicit params after the non-implicit params in a separate ()
e.g.
def myFunction(): Int = {
implicit val y: Int = 33
implicit val z: Double = 3.3
functionWithImplicit("foo") // calls functionWithImplicit("foo")(y, z)
}
def functionWithImplicit(foo: String)(implicit x: Int, d: Double) = // blar blar
Implicit parameters are heavily used in the collection API. Many functions get an implicit CanBuildFrom, which ensures that you get the 'best' result collection implementation.
Without implicits you would either pass such a thing all the time, which would make normal usage cumbersome. Or use less specialized collections which would be annoying because it would mean you loose performance/power.
I am commenting on this post a bit late, but I have started learning scala lately.
Daniel and others have given nice background about implicit keyword.
I would provide me two cents on implicit variable from practical usage perspective.
Scala is best suited if used for writing Apache Spark codes. In Spark, we do have spark context and most likely the configuration class that may fetch the configuration keys/values from a configuration file.
Now, If I have an abstract class and if I declare an object of configuration and spark context as follows :-
abstract class myImplicitClass {
implicit val config = new myConfigClass()
val conf = new SparkConf().setMaster().setAppName()
implicit val sc = new SparkContext(conf)
def overrideThisMethod(implicit sc: SparkContext, config: Config) : Unit
}
class MyClass extends myImplicitClass {
override def overrideThisMethod(implicit sc: SparkContext, config: Config){
/*I can provide here n number of methods where I can pass the sc and config
objects, what are implicit*/
def firstFn(firstParam: Int) (implicit sc: SparkContext, config: Config){
/*I can use "sc" and "config" as I wish: making rdd or getting data from cassandra, for e.g.*/
val myRdd = sc.parallelize(List("abc","123"))
}
def secondFn(firstParam: Int) (implicit sc: SparkContext, config: Config){
/*following are the ways we can use "sc" and "config" */
val keyspace = config.getString("keyspace")
val tableName = config.getString("table")
val hostName = config.getString("host")
val userName = config.getString("username")
val pswd = config.getString("password")
implicit val cassandraConnectorObj = CassandraConnector(....)
val cassandraRdd = sc.cassandraTable(keyspace, tableName)
}
}
}
As we can see the code above, I have two implicit objects in my abstract class, and I have passed those two implicit variables as function/method/definition implicit parameters.
I think this is the best use case that we can depict in terms of usage of implicit variables.
I want to come out a way to define a new method in some existing class in scala.
For example, I think the asInstanceOf[T] method has too long a name, I want to replace it with as[T].
A straight forward approach can be:
class WrappedAny(val a: Any) {
def as[T] = a.asInstanceOf[T]
}
implicit def wrappingAny(a: Any): WrappedAny = new WrappedAny(a)
Is there a more natural way with less code?
Also, a strange thing happens when I try this:
scala> class A
defined class A
scala> implicit def toA(x: Any): A = x
toA: (x: Any)A
scala> toA(1)
And the console hang. It seems that toA(Any) should not pass the type checking phase, and it can't when it's not implicit. And putting all the code into a external source code can produce the same problem. How did this happen? Is it a bug of the compiler(version 2.8.0)?
There's nothing technically wrong with your approach to pimping Any, although I think it's generally ill-advised. Likewise, there's a reason asInstanceOf and isInstanceOf are so verbosely named; it's to discourage you from using them! There's almost certainly a better, statically type-safe way to do whatever you're trying to do.
Regarding the example which causes your console to hang: the declared type of toA is Any => A, yet you've defined its result as x, which has type Any, not A. How can this possibly compile? Well, remember that when an apparent type error occurs, the compiler looks around for any available implicit conversions to resolve the problem. In this case, it needs an implicit conversion Any => A... and finds one: toA! So the reason toA type checks is because the compiler is implicitly redefining it as:
implicit def toA(x: Any): A = toA(x)
... which of course results in infinite recursion when you try to use it.
In your second example you are passing Any to a function that must return A. However it never returns A but the same Any you passed in. The compiler then tries to apply the implicit conversion which in turn does not return an A but Any, and so on.
If you define toA as not being implicit you get:
scala> def toA(x: Any): A = x
<console>:6: error: type mismatch;
found : Any
required: A
def toA(x: Any): A = x
^
As it happens, this has been discussed on Scala lists before. The pimp my class pattern is indeed a bit verbose for what it does, and, perhaps, there might be a way to clean the syntax without introducing new keywords.
The bit about new keywords is that one of Scala goals is to make the language scalable through libraries, instead of turning the language into a giant quilt of ideas that passed someone's criteria for "useful enough to add to the language" and, at the same time, making other ideas impossible because they weren't deemed useful and/or common enough.
Anyway, nothing so far has come up, and I haven't heard that there is any work in progress towards that goal. You are welcome to join the community through its mailing lists and contribute to its development.
Using Scala's command line REPL:
def foo(x: Int): Unit = {}
def foo(x: String): Unit = {println(foo(2))}
gives
error: type mismatch;
found: Int(2)
required: String
It seems that you can't define overloaded recursive methods in the REPL. I thought this was a bug in the Scala REPL and filed it, but it was almost instantly closed with "wontfix: I don't see any way this could be supported given the semantics of the interpreter, because these two methods must to be compiled together." He recommended putting the methods in an enclosing object.
Is there a JVM language implementation or Scala expert who could explain why? I can see it would be a problem if the methods called each other for instance, but in this case?
Or if this is too large a question and you think I need more prerequisite knowledge, does someone have any good links to books or sites about language implementations, especially on the JVM? (I know about John Rose's blog, and the book Programming Language Pragmatics... but that's about it. :)
The issue is due to the fact that the interpreter most often has to replace existing elements with a given name, rather than overload them. For example, I will often be running through experimenting with something, often creating a method called test:
def test(x: Int) = x + x
A little later on, let's say that I'm running a different experiment and I create another method named test, unrelated to the first:
def test(ls: List[Int]) = (0 /: ls) { _ + _ }
This isn't an entirely unrealistic scenario. In fact, it's precisely how most people use the interpreter, often without even realizing it. If the interpreter arbitrarily decided to keep both versions of test in scope, that could lead to confusing semantic differences in using test. For example, we might make a call to test, accidentally passing an Int rather than List[Int] (not the most unlikely accident in the world):
test(1 :: Nil) // => 1
test(2) // => 4 (expecting 2)
Over time, the root scope of the interpreter would get incredibly cluttered with various versions of methods, fields, etc. I tend to leave my interpreter open for days at a time, but if overloading like this were allowed, we would be forced to "flush" the interpreter every so often as things got to be too confusing.
It's not a limitation of the JVM or the Scala compiler, it's a deliberate design decision. As mentioned in the bug, you can still overload if you're within something other than the root scope. Enclosing your test methods within a class seems like the best solution to me.
% scala28
Welcome to Scala version 2.8.0.final (Java HotSpot(TM) 64-Bit Server VM, Java 1.6.0_20).
Type in expressions to have them evaluated.
Type :help for more information.
scala> def foo(x: Int): Unit = () ; def foo(x: String): Unit = { println(foo(2)) }
foo: (x: String)Unit <and> (x: Int)Unit
foo: (x: String)Unit <and> (x: Int)Unit
scala> foo(5)
scala> foo("abc")
()
REPL will accept if you copy both lines and paste both at same time.
As shown by extempore's answer, it is possible to overload. Daniel's comment about design decision is correct, but, I think, incomplete and a bit misleading. There's no outlawing of overloads (since they are possible), but they are not easily achieved.
The design decisions that lead to this are:
All previous definitions must be available.
Only newly entered code is compiled, instead of recompiling everything ever entered every time.
It must be possible to redefine definitions (as Daniel mentioned).
It must be possible to define members such as vals and defs, not only classes and objects.
The problem is... how to achieve all these goals? How do we process your example?
def foo(x: Int): Unit = {}
def foo(x: String): Unit = {println(foo(2))}
Starting with the 4th item, A val or def can only be defined inside a class, trait, object or package object. So, REPL puts the definitions inside objects, like this (not actual representation!)
package $line1 { // input line
object $read { // what was read
object $iw { // definitions
def foo(x: Int): Unit = {}
}
// val res1 would be here somewhere if this was an expression
}
}
Now, due to how JVM works, once you defined one of them, you can't extend them. You could, of course, recompile everything, but we discarded that. So you need to place it in a different place:
package $line1 { // input line
object $read { // what was read
object $iw { // definitions
def foo(x: String): Unit = { println(foo(2)) }
}
}
}
And this explains why your examples are not overloads: they are defined in two different places. If you put them in the same line, they'd all be defined together, which would make them overloads, as shown in extempore's example.
As for the other design decisions, each new package import definitions and "res" from previous packages, and the imports can shadow each other, which makes it possible to "redefine" stuff.