Introduce case-when syntax.

[Kray-G]

I try to start discussion about some future features.

Introduction

Thankfully according to increasing stars and visited by a lot of people, I am often asked about break in switch-case.
This is a design decision by following meanings, but I also sometimes think that I would like to solve this issue.

• Here are reasons.
  • This Kinx is to be similar with C and JavaScript.
  • The looking about no break in switch-case feels me the line will be connected to the next case. I imagine almost all C programmers will feel like that.

switch (x) {
case 1:
    a = 1;
case 2:
    b = 1;
}

How do you think about it?

Consideration

That is why, I considered the following 2 solutions.

• break should be necessary, but fallthrough keyword will be introduced and it is also necessary when it is fall through.
• Introducing another syntax like case-when instead of switch-case. switch-case will be stayed the same.

Conclusion

I am now planning to introduce case-when syntax for this.
Although switch-case is a statement, I thnk to make case-when be as an expression.
Therefore, if you write multiple statements in when clause, I assume that you will have to write a callback function.
Maybe the arguments of that function will be values specified of case and when.

var x = case y
    when 1:
        1
    when 2:
        10
    when variable: { &(y, variable)
        return y + 1;
    }
;

How about this?

Whatever you accept this or not, I am glad if giving me a feedback.
Of course I will be welcome if you have any other ideas.

[htsign]

I want the feature of case-in in ruby rather than case-when.
If adopted this, we can write a code like below:

var x = case [y, z]
    when [1, _]: 1
    when [2, _]: 10
    when [v, 0]: {
        return v + 42;
    }
    default: // or `else` keyword
        0
;

[Kray-G]

Thanks for examples, I understood what you meant. But I have a question about one thing. If I have misunderstanding, please let me know.

At the last case, I think the variable v is not an lvalue, don't you think so?
It is because it means "when y equals to v". Therefore it doesn't mean it matches to any cases, and I think it means, if v has some value and y equals to that value, it will return v + 42.

What do you think?

Thanks in advance.

[htsign]

Therefore it doesn't mean it matches to any cases, and I think it means, if v has some value and y equals to that value, it will return v + 42.

Oh, that's true.
During the discussion, I equated it with the pattern matching - it used in the functional paradigm.
By the way, other languages solve the problem by adding expressions:

(* OCaml *)
let x =
  match y with
  | 1 -> 1
  | 2 -> 10
  (* `=` is equality operator / it means "y is captured in v and v equals z then ..." *)
  | v when v = z -> v + 42
  (* same above *)
  | _ when y = z -> y + 42
  (* default case *)
  | _ -> -1

// Scala
val x = y match {
  case 1 => 1
  case 2 => 10
  case v if v == z => v + 42
  case _ => -1
}

[Kray-G]

I see, you mean it is possible to use if modifier for the special condition, right?
Surely, it looks the case of a pattern matching syntax will include a normal case-when in Ruby...

Here are some examples. I think the range will be also available.
How about these?

var x = case y
    when 1:
        1
    when 2..10:
        10
    when v if v == m:
        v + 42
    when v if v == n:
        v + 58
    else:
        -1
;

For array, we can use a spread/rest operator.

var x = case [y, z, i, j, k]
    when [1, ...a]:
        a[2]
    when [2..10, a, b]:
        a + b + 10
    when [v, m, ...n] if 10 < v && v <= 20:
        v + m + n[0] + 42
    when [v, 50, ...a] if v == n:
        v + 58
    else:
        -1
;

For object, Kinx don't have a spread/rest operator.

var obj = { x: 10, y: 20, sub: { a: 1, b: 2, c: 3 } };
var x = case obj
    when { x: 5, y: y, sub: z } if y == m:
        { => y }
    when { x: 6, y: y, sub: { a: 1, b: b, c: c} }:
        y + b + c;
    when { x: 10, y: y, sub: z }:
        10
;

Thanks and best regards.

[htsign]

Great examples. I love that codes!! 😄
But it's hard to implement them, don't you?

[Kray-G]

Okay I will make it!
I'm going to define a specification clearly first based on this discussion.
As you say the implementation may be hard, so I will try some ideas.

Many thanks a lot.

[Kray-G]

Conclusion

Overview

I have implemented this feature as suited with a current supported feature,
but there are some differences from the examples on this discussion.

1. If-modifier ... condition needs parentheses like if (v == m)
2. Colon ... you don't need to put a colon. It is possible to put a colon to separate a condition and a body of when clause, but it is just for an appearance.
3. Block & function ... by putting a block to when clause it is an automatically called function object. But this is a special feature for a block. A function like { => y } which is not a block is never called automatically. A function object itself is just returned as a result of when clause. When you want to call it, use { => y}() instead.
4. Semicolon ... case-when is an expression, so a body of when clause except in a block can not use a semicolon.

Related Updates

Before I implemented this feature, I tried that a pattern matching mechanism is applied to declaration statement first.
It is because I thought that this feature can be realized by the same mechanism.
That is correct and it was successful to implement it.
Therefore the pattern matching algorithm now can be used also in the following syntaxes.

• declaration.
• assignment.
• function argument.

See documents described in References section for details.

References

Finally, I wrote a specification for this feature at a document below.

• case-when
  • A specification of case-when.
• declaration
  • It supported a pattern matching mechanism also in declaration.
• assignment
  • It supported a pattern matching mechanism also in assignment.
• function
  • It supported a pattern matching mechanism also in function arguments.

If you have any problem, please let me know.

Thank you very much.

[Kray-G]

I updated case-when and some related features in the current master trunk.
It will be released in the next release.
See switch-when and Case-When Expression for details.
The summary of those are as follows.

• Updates of case-when
  • Pin operator such as ^a. Just a means an lvalue to match all, but pin operator of ^a means the value is a rvalue and to compare values.
  • Alternative pattern by |. 1 | 2 | 3 means to match 1, to match 2, or to match 3.
  • Error when it is array and the length is different.
  • Putting a function object in a when clause is available. It means to check if a returned value by the function is true.
  • _ in the when clause means to ignore binding a value at when condition.
• Introduced switch-when statement.
  • The benefit of switch-case is to compile to a jump by table in some cases. This is good for the performance, but it never do so in case-when.
  • It is possible to use when instead of case in switch statement. When when is used, break will be automatically done after the last statement.
  • Instead, fallthrough keyword is prepared and it will make it disable to do auto break.
  • else label is also prepared. It means a default label with auto break.
  • Of course, switch-when statement can be also used in native.

If you have any problem, feel free to let me know.
Best regards.

[htsign]

Great update!

Putting a function object in a when clause is available. It means to check if a returned value by the function is true.

What is meaning of "true", True or true that is an alias of 1 ?
As far as I read "Function Call in When Clause" section of the reference, I think it's the latter.

It is possible to use when...

It is wrong the markup.

[Kray-G]

Thanks for the feedback!

Putting a function object in a when clause is available. It means to check if a returned value by the function is true.

What is meaning of "true", True or true that is an alias of 1 ?
As far as I read "Function Call in When Clause" section of the reference, I think it's the latter.

Now in the current code, the true means just a non-zero value, but it will be changed a little in the release,
because I feel it had better be same as a condition of if. This means the a of if (a).
To say it a little more detailed, it will meet the following conditions.

• undefined/null ... always false.
• Integer ... true means a non-zero value.
• BigInteger ... always true.
• Double ... true if the distance between 2 values are bigger than DBL_EPSILON.
• String ... true if it is not empty string.
• Function ... always false.
• Object ... basically true.

It is possible to use when...

It is wrong the markup.

Thanks for a report. I fixed it.

best regards.