Bug in Scalajs when compiling pattern matching? - compilation

I currently cannot reproduce this in a minimum working example, but I'm working on it.
Adding a line in a dead code causes a crash at run-time.
Here are the relevant code snippets, first from the scala file:
// We are inside an object. Expr is a trait with many case classes.
case class Problem(in: Expr, out: Expr)
type RepairOutput = Either[String, Stream[Expr]]
private sealed trait RepairFinalTransform
...
private sealed trait TransformExtend extends RepairFinalTransform {
def continuation(repair: Problem => RepairOutput)
: Expr => RepairOutput
}
private case class TransformSolutionAndContinue(f: Expr => (Problem, Expr => Expr)) extends TransformExtend {
def continuation(repair: Problem => RepairOutput): Expr => RepairOutput = (x: Expr) => f(x) match {
case (p, rewriter) => repair(p) andMap rewriter
}
}
private case class TransformAndContinue(f: Expr => RepairOutput) extends TransformExtend {
def continuation(repair: Problem => RepairOutput): Expr => RepairOutput = f
}
The following code snippet
case c: TransformExtend =>
r = r andThen c.continuation(_repair)
, when working, is compiled as:
if ($is_Lexample_LambdaCalculus$TransformExtend(x1)) {
var x5 = $as_Lexample_LambdaCalculus$TransformExtend(x1);
var e$1 = r;
r = new $c_Lexample_LambdaCalculus$Retransformer().init___s_util_Either(e$1).andThen__F1__s_util_Either(x5.continuation__F1__F1(new $c_sjsr_AnonFunction1().init___sjs_js_Function1((function($this) {
return (function(original_problem$2) {
var original_problem = $as_Lexample_LambdaCalculus$Problem(original_problem$2);
return $m_Lexample_LambdaCalculus$().$$undrepair__Lexample_LambdaCalculus$Problem__s_util_Either(original_problem)
})
})(this))))
}
However, as soon as I add one occurrence of TransformAndContinue(???) (I did not have any before) in a place that is currently never executed, the above code becomes:
if ($is_Lexample_LambdaCalculus$TransformExtend(x1)) {
var x5 = $as_Lexample_LambdaCalculus$TransformExtend(x1);
var e$1 = r;
r = new $c_Lexample_LambdaCalculus$Retransformer().init___s_util_Either(e$1).andThen__F1__s_util_Either(x5.f$1)
}
Note that x5 is supposed to be a type cast to TransformExtend, which does not have the field f, only the continue method. Since f exists in both classes, its return type differs. Besides, the case class TransformSolutionAndContinue was used in many other places, so there is no reason why it calls f instead of continue.
So the translation to Javascript seems wrong (and actually, the errors fail on this point). I'm using scalajs 0.6.19 and compile using fastOptJS and scalatest.
Can you tell me what to change in my code so that the compilation is correct? Is it a known bug?

Related

Prae:Wrapper: need to use iter_mut of interior Vec but Prae:Wrapper only provides immutable access

I'm using the prae crate for validation and the following function gives me errors:
fn advance_rotors(&mut self) {
self.rotors.get()[0].rotate();
let mut iterhandle = self.rotors.iter_mut().peekable(); // Error at iter_mut() #0599
while let Some(el) = iterhandle.next() {
match iterhandle.peek_mut() {
Some(next_rotor) => match el.should_advance_next() {
true => {
next_rotor.rotate(); // This line requires mutable access to next_rotor
}
false => (),
},
None => (),
}
}
}
and the definition of my struct here:
pub struct Enigma {
reflector: Reflector,
rotors: RotorConfig, // Only mutable via getter and setter functions
}
the struct of interest here is RotorConfig which is generated using the define! macro from prae. Here's the code:
prae::define! {
#[derive(Debug)]
RotorConfig: Vec<Rotor>; // I need to be able to call the rotate method of each rotor in this vec. This requires mutability
validate(RotorConfigError) |config| {
match config.len(){
3..=4 => (),
_ => return Err(RotorConfigError::Size)
}
match config.iter().unique().count(){
3..=4 =>(),
_ => return Err(RotorConfigError::Duplicate)
}
Ok(())
};
}
the issue stems from the fact that prae only allows for immutable access to the internal representation via getter and setter functions so as to ensure the validity of the values inside. As you can see in my advance_rotors function I wrote before implementing validation I'm getting an error because I need to call rotor.rotate mutably. I'm at a loss as to how to accomplish this
After posting this I realized that I can simply provide interior mutability by using the following impl block
impl RotorConfig{
fn advance_rotors(&mut self)
{
self.0[0].rotate();
let mut iterhandle = self.0.iter_mut().peekable();
while let Some(el) = iterhandle.next() {
match iterhandle.peek_mut() {
Some(next_rotor) => match el.should_advance_next() {
true => {
next_rotor.rotate();
}
false => (),
},
None => (),
}
}
}
}
As you can see the function largely remains unchanged except that we replace self.rotors with self.0

Enum as Parameter in TypeScript

Isn't it possible to set the type of a parameter to an Enum? Like this:
private getRandomElementOfEnum(e : enum):string{
var length:number = Object.keys(e).length;
return e[Math.floor((Math.random() * length)+1)];
}
Following error is thrown:
Argument expression expected.(1135)
With any obviously everything is alright:
private getRandomElementOfEnum(e : any):string{
var length:number = Object.keys(e).length;
return e[Math.floor((Math.random() * length)+1)];
}
Is there a possibility or a little workaround to define an enum as a parameter?
It's not possible to ensure the parameter is an enum, because enumerations in TS don't inherit from a common ancestor or interface.
TypeScript brings static analysis. Your code uses dynamic programming with Object.keys and e[dynamicKey]. For dynamic codes, the type any is convenient.
Your code is buggy: length() doesn't exists, e[Math.floor((Math.random() * length)+1)] returns a string or an integer, and the enumeration values can be manually set…
Here is a suggestion:
function getRandomElementOfEnum<E>(e: any): E {
var keys = Object.keys(e),
index = Math.floor(Math.random() * keys.length),
k = keys[index];
if (typeof e[k] === 'number')
return <any>e[k];
return <any>parseInt(k, 10);
}
function display(a: Color) {
console.log(a);
}
enum Color { Blue, Green };
display(getRandomElementOfEnum<Color>(Color));
Ideally, the parameter type any should be replaced by typeof E but the compiler (TS 1.5) can't understand this syntax.
You can do better than any:
enum E1 {
A, B, C
}
enum E2 {
X, Y, Z
}
function getRandomElementOfEnum(e: { [s: number]: string }): number {
/* insert working implementation here */
return undefined;
}
// OK
var x: E1 = getRandomElementOfEnum(E1);
// Error
var y: E2 = getRandomElementOfEnum(window);
// Error
var z: string = getRandomElementOfEnum(E2);
I agree with #Tarh. Enums in TypeScript are just Javascript objects without a common interface or prototype (and if they are const enum, then they are not even objects), so you cannot restrict types to "any enum".
The closest I could get is something like the following:
enum E1 {
A, B, C
}
enum E2 {
X, Y, Z
}
// make up your own interface to match TypeScript enums
// as closely as possible (not perfect, though)
interface Enum {
[id: number]: string
}
function getRandomElementOfEnum(e: Enum): string {
let length = Object.keys(e).length / 2;
return e[Math.floor((Math.random() * length))];
}
This works for all enums (without custom initializers), but it would also accept other arrays as input (and then fail because the method body relies on the very specific key structure that TypeScript enums have).
So unless you have a real need for such a "generic" function, make typesafe functions for the individual enum types (or a union type like E1|E2|E3) that you actually need.
And if you do have this need (and this might very well be an X-Y-problem that can be solved in a better, completely different way given more context), use any, because you have left typesafe territory anyway.
Summing up the previous answers with some new syntax - a generic typesafe function, which works with numeric enums as well as string enums:
function getRandomElementOfEnum<T extends {[key: number]: string | number}>(e: T): T[keyof T] {
const keys = Object.keys(e);
const randomKeyIndex = Math.floor(Math.random() * keys.length);
const randomKey = keys[randomKeyIndex];
// Numeric enums members also get a reverse mapping from enum values to enum names.
// So, if a key is a number, actually it's a value of a numeric enum.
// see https://www.typescriptlang.org/docs/handbook/enums.html#reverse-mappings
const randomKeyNumber = Number(randomKey);
return isNaN(randomKeyNumber)
? e[randomKey as keyof T]
: randomKeyNumber as unknown as T[keyof T];
}
Another possible option not mentioned above is using the actual values. This is however possible only when you know all the options. This, in my opinion is definitely better than any.
doSomething(a: string, b: 'this'|'can'|'work'): void {
//do something
}
Tested on TypeScript 3.9.7
Solution
type EnumTypeString<TEnum extends string> =
{ [key in string]: TEnum | string; }
type EnumTypeNumber<TEnum extends number> =
{ [key in string]: TEnum | number; }
| { [key in number]: string; }
type EnumType<TEnum extends string | number> =
(TEnum extends string ? EnumTypeString<TEnum> : never)
| (TEnum extends number ? EnumTypeNumber<TEnum> : never)
type EnumOf<TEnumType> = TEnumType extends EnumType<infer U>
? U
: never
Usage
EnumType:
function forEachEnum<TEnum extends string | number>(
enumType: EnumType<TEnum>,
callback: (value: TEnum, key: string) => boolean|void,
) {
for (let key in enumType) {
if (Object.prototype.hasOwnProperty.call(enumType, key) && isNaN(Number(key))) {
const value = enumType[key] as any
if (callback(value, key)) {
return
}
}
}
}
EnumOf:
function forEachEnum2<TEnumType>(
enumType: TEnumType,
callback: (value: EnumOf<TEnumType>, key: string) => boolean|void,
) {
for (let key in enumType) {
if (Object.prototype.hasOwnProperty.call(enumType, key) && isNaN(Number(key))) {
const value = enumType[key] as any
if (callback(value, key)) {
return
}
}
}
}
Tests
enum EnumAsString {
Value1 = 'value 1',
Value2 = 'value 2',
}
enum EnumAsNumber {
Value1 = 1,
Value2 = 2,
}
// Error
let sn: EnumType<string> = EnumAsNumber
// Correct
let ns: EnumType<number> = EnumAsString // I have not found a solution for the error here
let nn: EnumType<number> = EnumAsNumber
let Nn: EnumType<EnumAsNumber> = EnumAsNumber
let ss: EnumType<string> = EnumAsString
let Ss: EnumType<EnumAsString> = EnumAsString
forEachEnum(EnumAsString, value => {
let e: EnumAsString = value
let s: string = value
let n: number = value // Error
})
forEachEnum(EnumAsNumber, value => {
let e: EnumAsNumber = value
let s: string = value // Error
let n: number = value
})
forEachEnum2(EnumAsString, value => {
let e: EnumAsString = value
let s: string = value
let n: number = value // Error
})
forEachEnum2(EnumAsNumber, value => {
let e: EnumAsNumber = value
let s: string = value // Error
let n: number = value
})
May be this trick will fit:
enum AbstractEnum { // put somewhere in hidden scope
}
private getRandomElementOfEnum(e: typeof AbstractEnum) {
...
}
#selinathat's solution is great only if you have few types. but what if we have more ? for example :
doSomething(a: string, b: 'this'|'can'|'work'|'test1'|'test2'|'test3'): void {
//do something
}
its pretty ugly hah !?
i prefer to use keyof :
interface Items {
'this',
'can',
'work',
'test1',
'test2',
'test3',
}
doSomething(a: string, b: keyof Items): void {
//do something
}
Here is an example that allows passing an enum with a typechecked value of that enum using a generic. It's really a response to a slightly different question here that was marked as a duplicate: Typescript how to pass enum as Parameter
enum Color {
blue,
};
enum Car {
cadillac,
};
enum Shape {
square,
}
type SupportedEnums = typeof Color | typeof Car;
type InvertTypeOf<T> = T extends typeof Color ? Color :
T extends typeof Car ? Car : never;
function getText<T extends SupportedEnums>(enumValue: InvertTypeOf<T>, typeEnum: T) string | undefined {
if (typeEnum[enumValue]) {
return `${enumValue}(${typeEnum[enumValue]})`;
}
}
console.log(getText(Car.cadillac, Car)); // 0(cadillac)
console.log(getText(0, Color)); // 0(red)
console.log(getText(4, Color)); // undefined
// #ts-expect-error Color is not Car
console.log(getText(Color.blue, Car));
// #ts-expect-error Car is not a Color
console.log(getText(Car.toyota, Color));
// #ts-expect-error Shape is not in SupportedEnums
console.log(getText(5, Shape));
// #ts-expect-error Shape is not in SupportedEnums
console.log(getText(Shape.square, Shape));
I had the same kind of problem, and i did this
private getOptionsFromEnum(OptionEnum: Record<string, string>): Array<SelectOption> {
return Object.keys(OptionEnum).map((value) => {
return {
name: OptionEnum[value],
value,
} as SelectOption;
});
}
I made a helper type to accept any enum as a paramaeter, then you can handle whatever you need next with Object or by calling an index of the Enum.
type Enum = Record<string | number, string | number>
Now use it to accept any enum as parameter:
function enumValues<T extends Enum>(enum: T, filter?: "string"): string[];
function enumValues<T extends Enum>(enum: T, filter?: "number"): number[];
function enumValues<T extends Enum>(enum: T, filter?: undefined): (string|number)[];
function enumValues<T extends Enum>(enum: T, filter?: "string" | "number") {
return Object.values(enum).filter(x => !filter || typeof x === filter);
}
enum color {
red,
green,
blue
}
console.log(enumValues(color,"string"));
// output ['red','green','blue']
console.log(enumValues(color,"number"));
// output [0,1,2]

Can I store methods in an array?

Consider the following code:
class Test {
func func1(arg1: Int) -> Void {
println(arg1)
}
var funcArr: Array< (Int) -> Void > = [func1] // (!) 'Int' is not a subtype of 'Test'
}
I'm trying to store the method func1in an array, but as you can see, this doesn't seem to work because func1supposedly only takes an argument of type Test. I assume this has something to do with methods needing to be associated with an object.
For some more clarification, have a look at the following code where I let swift infer the type of the array:
class Test {
func func1(arg1: Int) -> Void {
println(arg1)
}
var funcArr = [func1]
}
Test().funcArr[0](Test()) // Compiles, but nothing gets printed.
Test().funcArr[0](1) // (!) Type 'Test' does not conform to protocol 'IntegerLiteralConvertible'
Test().func1(1) // Prints '1'
A possible workaround for this problem is moving func1outside of the class like so:
func func1(arg1: Int) -> Void {
println(arg1)
}
class Test {
var funcArr = [func1]
}
Test().funcArr[0](1) // Prints '1'
This works fine for this simple example, but is less than ideal when I actually need to operate on an Object of type Test in the function. I can of course add another parameter to the function to pass an instance of Testto the function, but this seems clunky.
Is there any way I can store methods in an array?
Ultimately, what I want to be able to do is testObject.funcArr[someInt](someParam) and have that function work as a method belonging to testObject. Any clever workarounds are of course also welcome.
Instance methods in swift are just curried functions, and the first argument is implicitly an instance of the class (i.e. self). And that's why these two are equivalent:
Test().func1(0)
Test.func1(Test())(0)
So when you try to put that function in the array, you're reveling its real nature: the method func1 on Test is actually this class method:
class func1(self_: Test)(arg1: Int)
So when you refer to simply func1 (without an "instance context") it has type Test -> Int -> Void, instead of the expected Int -> Void, and that's why you get the error
Int is not a subtype of Test
So the real issue is that when you store the methods in funcArr the instance is not known yet (or if you will, you're referring the function at a class level). You can work around this using a computed property:
var funcArr: [Int -> Void] { return [func1] } // use a computed property
Or another valuable option could be simply to acknowledge the real type of func1 and explicitly passing the instance. E.g.
...
var funcArr = [func1]
...
let test = Test()
let func1 = test.funcArr[0]
func1(test)(0) // prints 0
update
Freely inspired by this other Q/A (Make self weak in methods in Swift) I came up with a similar solution that stores the method references.
func weakRef<T: AnyObject, A, B>(weakSelf: T, method: T -> A -> B) -> A -> B {
return { [unowned weakSelf] in { a in method(weakSelf)(a) } }()
}
class Test {
var methodRefs: [Int -> Void] = []
init() {
methodRefs.append(weakRef(self, Test.func1))
}
func func1(arg1: Int) {
println(arg1)
}
}
In order to store a method, you should remember that the method is invoked on a class instance. What's actually stored in the array is a curried function:
(Test) -> (Int) -> Void
The first function takes a class instance and returns another function, the actual (Int) -> Void method, which is then invoked on that instance.
To make it more explicit, the array type is:
var funcArr: [(Test) -> (Int) -> Void] = [Test.func1]
Then you can invoke the method with code like this:
var test = Test()
var method = test.funcArr[0]
method(test)(1)
Suggested reading: Curried Functions

Scala Macro: get param default value

I have the next code, and i would like to extract the default parametr from value.
//
def extractor[T] = macro extractorImpl[T]
def extractorImpl[T: c.WeakTypeTag](c: Context) = {
//first i got a type contructor
???
}
i try with attachments but attachments.all return a Set[Any] with (for example) SymbolSourceAttachment(val name: String = "new name")
SymbolSourceAttachment contain ValDef but i do not know how to extract from SymbolSourceAttachment ValDef.
By the way i should to get a Map[String, String]("name" -> "new name")
Example:
case class Person(name: String = "new name")
object Macro {
def extractor[T] = macro extractorImpl[T]
def extractorImpl[T: c.WeakTypeTag](c: Context) = {
import c.universe._
c.weakTypeOf[T].declarations.collect {
case a: MethodSymbol if a.isConstructor =>
a.paramss.collect {
case b => b.collect {
case c =>
c.attachments.all {
case d => println(showRaw(d)) // => SymbolSourceAttachment(val name: String = "new name")
}
}
}
}
}
}
And macro should return Map("name" -> "new name")
Since you're seeing SymbolSourceAttachment, I assume you're using macro paradise (because it's an internal attachment used only in paradise), so I'll feel free to use quasiquotes :)
There's no easy way to get values of default parameters in Scala reflection API. Your best shot would be reverse-engineering the names of methods that are created to calculate default values and then referring to those.
SymbolSourceAttachment would sort of work if your macro is expanding in the same compilation run that compiles the case class, but it would break under separate compilation (attachments aren't saved in class files), and it wouldn't work in vanilla Scala (because this attachment is exclusive to paradise).
=== Macros.scala ===
import scala.reflect.macros.Context
import scala.language.experimental.macros
object Macros {
def impl[T](c: Context)(T: c.WeakTypeTag[T]): c.Expr[Map[String, Any]] = {
import c.universe._
val classSym = T.tpe.typeSymbol
val moduleSym = classSym.companionSymbol
val apply = moduleSym.typeSignature.declaration(newTermName("apply")).asMethod
// can handle only default parameters from the first parameter list
// because subsequent parameter lists might depend on previous parameters
val kvps = apply.paramss.head.map(_.asTerm).zipWithIndex.flatMap{ case (p, i) =>
if (!p.isParamWithDefault) None
else {
val getterName = newTermName("apply$default$" + (i + 1))
Some(q"${p.name.toString} -> $moduleSym.$getterName")
}
}
c.Expr[Map[String, Any]](q"Map[String, Any](..$kvps)")
}
def extractor[T]: Map[String, Any] = macro impl[T]
}
=== Test.scala ===
case class C(x: Int = 2, y: String, z: Boolean = true)(t: String = "hello")
object Test extends App {
println(Macros.extractor[C])
}
17:10 ~/Projects/Paradise2103/sandbox/src/main/scala (2.10.3)$ scalac Macros.scala && scalac Test.scala && scala Test
Map(x -> 2, z -> true)

Threading `Try`s through for-comprehension

Triggered by another question (which has been subsequently edited away though), I wanted to try out how easy it would be to chain calls to Scala 2.10's Try construct (cf. this presentation), using for-comprehensions.
The idea is to have a list of tokens and match them against a sequence of patterns, then return the first error or the successfully matched pattern. I arrived at the following pretty awkward version, and I wonder if this can be made simpler and nicer:
import util.Try
trait Token
case class Ident (s: String) extends Token
case class Keyword(s: String) extends Token
case class Punct (s: String) extends Token
case object NoToken extends Token
case class FunctionDef(id: Ident)
case class Expect[A](expectation: String)(pattern: PartialFunction[Token, A]) {
def unapply(tup: (Try[_], Token)) = Some(tup._1.map { _ =>
pattern.lift(tup._2).getOrElse(throw new Exception(expectation))
})
}
Now construct the expectations for Keyword("void") :: Ident(id) :: Punct("(") :: Punct(")") :: tail
val hasVoid = Expect("function def starts with void") { case Keyword("void") => }
val hasIdent = Expect("expected name of the function") { case id: Ident => id }
val hasOpen = Expect("expected opening parenthesis" ) { case Punct("(") => }
val hasClosed = Expect("expected closing parenthesis" ) { case Punct(")") => }
Construct a full test case:
def test(tokens: List[Token]) = {
val iter = tokens.iterator
def next(p: Try[_]) = Some(p -> (if (iter.hasNext) iter.next else NoToken))
def first() = next(Try())
val sq = for {
hasVoid (vd) <- first()
hasIdent (id) <- next(vd)
hasOpen (op) <- next(id)
hasClosed(cl) <- next(op)
} yield cl.flatMap(_ => id).map(FunctionDef(_))
sq.head
}
The following verifies the test mehod:
// the following fail with successive errors
test(Nil)
test(Keyword("hallo") :: Nil)
test(Keyword("void" ) :: Nil)
test(Keyword("void" ) :: Ident("name") :: Nil)
test(Keyword("void" ) :: Ident("name") :: Punct("(") :: Nil)
// this completes
test(Keyword("void" ) :: Ident("name") :: Punct("(") :: Punct(")") :: Nil)
Now especially the additional flatMap and map in yield seems horrible, as well as the need to call head on the result of the for comprehension.
Any ideas? Is Try very badly suited for for comprehensions? Shouldn't either Either or Try be "fixed" to allow for this type of threading (e.g. allow Try as a direct result type of unapply)?
The trick seems to be to not create Try instances in the inner structure, but instead let that throw exceptions and construct one outer Try.
First, let's get rid of the Try[Unit]'s:
case class Expect(expectation: String)(pattern: PartialFunction[Token, Unit]) {
def unapply(token: Token) =
pattern.isDefinedAt(token) || (throw new Exception(expectation))
}
case class Extract[A](expectation: String)(pattern: PartialFunction[Token, A]) {
def unapply(token: Token) = Some(
pattern.lift(token).getOrElse(throw new Exception(expectation))
)
}
Then the checks become:
val hasVoid = Expect ("function def starts with void") { case Keyword("void") => }
val getIdent = Extract("expected name of the function") { case id: Ident => id }
val hasOpen = Expect ("expected opening parenthesis" ) { case Punct("(") => }
val hasClosed = Expect ("expected closing parenthesis" ) { case Punct(")") => }
And the test method:
def test(tokens: List[Token]) = Try {
val iter = tokens.iterator
def next() = Some(if (iter.hasNext) iter.next else NoToken)
(for {
hasVoid() <- next()
getIdent(id) <- next()
hasOpen() <- next()
hasClosed() <- next()
} yield FunctionDef(id)).head // can we get rid of the `head`?
}

Resources