Add WHEN construct, like switch, but better!

[siy29181]

A traditional switch is basically just a statement that can substitute a series of simple if/else that make basic checks. However it cannot replace all sort of if/else sequences but just those which compare a value with some constant. So, you can only use a switch to perform an action when one specific variable has a certain precise value.
This is quite limited and while is better than having nothing1, it is useful only in few circumstances. Instead the Kotlin’s when can offer much more than that:

1. It can be used as an expression or a statement (i.e., it can return a value or not).
2. It has a better and safer design.
3. It can have arbitrary condition expressions.
4. It can be used without an argument.

[svick]

Are you aware of the C# 7.0 when clause for the switch statement, the C# 8.0 switch expression, and the change to allow the switch expression to be used as a statement, which might be included in C# 9.0 (#2632)? They don't address all your requirements, but I think they go a long way to solving many of them.

In other words, this is an area that has been actively worked on in the past few releases of C# and is likely going to be further improved in the next few releases to come.

[HaloFour]

Without more details and some examples it's a little difficult to ascertain what the poster is asking for, but it sounds kind of like VB.NET's Select Case True where you can have a switch statement/expression where each case arm is an arbitrary bool expression:

Select Case True
    Case Sky = Blue
        DoSomething()
    Case Water = Wet
        DoSomethingElse()
    Case 2 + 2 = 4
        DoOtherSomething()
End Select

[huoyaoyuan]

These should be all about feature gap between statements and expressions. I'd really want rust style expressions here:

result = 
if (Sky == Blue) { DoSomething(); 1 } 
else if (Water == Wet) { DoSomethingElse(); 2 }
else { 3 };

Another reason for not using long ?: sequence is lack of ability to step in expressions.

[CyrusNajmabadi]

Another reason for not using long ?: sequence is lack of ability to step in expressions.

That's a tooling problem, not a language problem.

[ronnygunawan]

Select Case True
    Case Sky = Blue
        DoSomething()
    Case Water = Wet
        DoSomethingElse()
    Case 2 + 2 = 4
        DoOtherSomething()
End Select

How I wish we could mix .vb and .cs in a same project.

[DavidArno]

Such syntax already exists in C#. It's just ugly:

// Statement form:
switch (true)
{
    case true when sky == Blue:
        DoSomething();
        break;
    case true when water == Wet:
        DoSomethingElse();
        break;
    case true when 2 + 2 == 4:
        DoOtherSomething();
}

// expression form:
var x = true switch {
    _ when sky == Blue => Something(),
    _ when water == Wet => SomethingElse(),
    _ when 2 + 2 == 4 => OtherSomething()
};

Both feel as much of a misuse of a switch as the VB example. I don't see the fact that VB supports such vulgar syntax as justification for improving the noise in the C# version.

However, the idea of if/else as an expression and expression blocks with a value (no return statement) as the final statement, { DoSomething(); 1 }, is growing on me.

[sharwell]

@DavidArno You can use case var _ when as a slightly more readable form of case true when (the keyword true is easy to misread as part of the when, but var _ is not)

[gulshan]

According to my #156 (very generic) if-else expression proposal, it will look like-

var x =
    if(sky == Blue) Something()
    else if(water == Wet) SomethingElse()
    else if(2 + 2 == 4) OtherSomething()
    else throw new InvalidOperationException();

Edit: I forgot the expression must have an else/default path, even according to my proposal.

[sharwell]

@gulshan that seems very similar to this:

var x =
    (sky == Blue) ? Something() :
    (water == Wet) ? SomethingElse() :
    (2 + 2 == 4) ? OtherSomething() :
    throw new InvalidOperationException();

[DavidArno]

@sharwell, it's not so much "similar" as nigh on identical. I personally find those `if's way easier to read than nested ternary expressions though. So they are identical, save for one being readable 😉

It can be hard to justify adding new syntax that is just another way of doing something that is already supported. But in my view, ternary expressions aren't fit for purpose these days. C# continues to add expression-based features but those nested ternaries render such expressions unreadable all too easily. So I'd really like to see an expression-based if/else offered up as an alternative.

It's possible though that it would be too hard to add such syntax without it being a breaking change, so I'm ready to resign myself to it never happening.

[CyrusNajmabadi]

but those nested ternaries render such expressions unreadable all too easily.

I find that odd given that these consturcts have existed for 40+ years across a multitude of languages and it's been fine from a readability perspective. Indeed, this is the scenario they were created to address. I'd posit that you're simply coming from a lack of familiarity with them, and that's leading to you to find them unreadable.

[gulshan]

Personally I learnt and knew about the ternaries for several more years than the if-else expression. Still I always tried to avoid ternaries but had no such problem with if-else expressions. It's not that I cannot read through the nested ternary expressions. But it puts a cognitive load on me. When there are complex expressions nested in a ternary operator, it's very hard to parse mentally without parentheses and proper manual formatting. Which is not the case with if-else expressions. And I'm sure I'm not alone. Some examples contrasting them-

sky == Blue ? Something() : water == Wet ? SomethingElse();
if(sky == Blue) Something() else if(water == Wet) SomethingElse();

x != 0.0 ? Math.Sin(x) / x : 1.0;
if(x != 0.0) Math.Sin(x) / x else 1.0;

age <= 12 ? 10 : age > 70 ? 15 : 0;
if(age <= 12) 10 else if(age > 70) 15 else 0;

[ronnygunawan]

It's not that I cannot read through the nested ternary expressions. But it puts a cognitive load on me.

It depends on your formatting. I usually either put the ? and : at the beginning of the line:

var result =
    (sky == blue)
        ? Something()
        : (water == Wet)
            ? SomethingElse()
            : (2 + 2 == 4)
                ? OtherSomething()
                : throw new InvalidOperationException();

or format it like an if statement:

var result = (sky == blue) ? (
    Something()
) : (water == wet) ? (
    SomethingElse()
) : (2 + 2 == 4) ? (
    OtherSomething()
) : (
    throw new InvalidOperationException()
);

or the combination of both:

var result = (sky == blue) ? (
    (water == wet)
        ? Something()
        : SomethingElse()
) : (
    throw new InvalidOperationException()
);

The only cognitive load left is to remember not to put a semicolon after each expression.

[gulshan]

My point was, I have to put parentheses and format the code manually to keep the ternary expression code readable. The default formatter is not helping me. if-else expressions (like Kotlin) would be different.

[DavidArno]

I'd posit that you're simply coming from a lack of familiarity with them, and that's leading to you to find them unreadable.

That's of course a possibility, @CyrusNajmabadi. However, I probably wrote my first ternary expression in C some 30 years ago. So if I still lack unfamiliarity with them after that many decades, either I have a problem with grasping simple concepts, or nested ternary expressions are inherently unreadable. I know which option my money is on 😆

[sharwell]

Given the language already contains the exact structure requested by this issue (i.e. the issue only requests to replace an operator character with a keyword in the same location) and the current supported form is more succinct than the current issue, I would lean against making a change to have a more verbose way to do the same thing. If density is a concern for readability of the ternary operator form, the full if statement provides a clean alternative.

[gulshan]

I am also waiting for upcoming relational patterns in switch cases #2850 . Then if-else expression proposal can be kept aside.

[DenisValeev]

@DavidArno @sharwell
Hi there,

Just gonna throw it in here for your consideration.

Consider this example:

switch (intCode) {
 case 404:
 case 429:
 case var _ when intCode >= 500:
  throw new TransientManagedIdentityException($ "encountered transient managed identity service error with status code {intCode}");
 case 400:
  throw new BadRequestManagedIdentityException();
}

What if we could omit the var _ part, that would be awesome:

switch (intCode) {
 case 404:
 case 429:
 case when intCode >= 500:
  throw new TransientManagedIdentityException($ "encountered transient managed identity service error with status code {intCode}");
 case 400:
  throw new BadRequestManagedIdentityException();
}

[RikkiGibson]

following up on a bit older comment, but @DenisValeev's scenario looks like it would be nicely addressed by #2850 (see the Relational Patterns section)