Incompatible type of nothing in switch expression with object cases

[lucaswerkmeister]

shared void run() {
    value c = switch (0) case (0) smaller case (1) equal else nothing;
}

source/tmp/run.ceylon:2: error: incompatible types

    value c = switch (0) case (0) smaller case (1) equal else nothing;
                                                              ^

required: Comparison
found: Object

The error disappears if both cases evaluate to the same object (e. g. case (1) smaller); in that case, the compiler casts nothing to smaller_.

As always, generated code, @noanno’ed.

[tombentley]

The problem here was that the type of the switch is these days smaller|equal (as a result of ceylon/ceylon-spec#865) so I suppose I need to add a denoteableType() call somewhere.

My concern is that denoteableType() is going to see liberal use through the code generator as we see more of this sort of bug. I don't know whether there's a right or a wrong place to call denoteableType(), because aren't we supposed to reify types like smaller|equal ?

[gavinking]

We only reify them when they are type args, and according to the spec such types are never inferred.

Sent from my iPhone

On 19 Jan 2015, at 12:28, Tom Bentley notifications@github.com wrote:

The problem here was that the type of the switch is these days smaller|equal (as a result of ceylon/ceylon-spec#865) so I suppose I need to add a denoteableType() call somewhere.

My concern is that denoteableType() is going to see liberal use through the code generator as we see more of this sort of bug. I don't know whether there's a right or a wrong place to call denoteableType(), because aren't we supposed to reify types like smaller|equal ?

—
Reply to this email directly or view it on GitHub.

[tombentley]

We only reify them when they are type args

Right. So when constructing a reified type argument I need to ensure I'm not using a denotableType()-ed type. But otherwise I can use denotableType() freely.

according to the spec such types are never inferred.

So does that mean this bug is a typechecker bug?

[gavinking]

@tombentley read again what I just wrote :)

[gavinking]

Hint: there are no type args in the above code!

[tombentley]

I assume that means it's not a typechecker bug, and we can expect to find this foo|bar stuff everywhere except for stuff like value and implicit values like for variables.

[gavinking]

No, it's not a bug, and yes, the backend needs to be able to accommodate expressions with types like this.