inline variable in switch

[gavinking]

Previously, I had been avoiding adding support for the rather convenient form switch (val=expression()) because it is in principle ambiguous: val=expression() is a legal expression. However, I just noticed that we let the syntax try (resource=expression()) slip into the language, which features the same in-principle ambiguity.

I don't think either of these is truly ambiguous in any meaningful sense: the semantic distance between an assignment and an inline variable specification is not very large, and I anyway don't think anyone writes assignments in switch or try.

So I'm inclined to allow this syntax in both switch and try. If anyone disagrees with my analysis, then we should disallow the syntax in both places.

[gavinking]

Initial implementation of this is in branch 1119. Requires support in the backends, obviously.

Also still requires updates to the spec and tests.

[lucaswerkmeister]

I anyway don't think anyone writes assignments in switch or try.

I did it here… but it won’t hurt much if the switch introduces a new variable instead of assigning to an existing variable.

[gavinking]

but it won’t hurt much if the switch introduces a new variable instead of assigning to an existing variable

:-(

Whenever I think "surely no-one has ever done this", it always turns out that someone has done it, and I break their code.

[lucaswerkmeister]

Eh, the code isn’t even a week old, you don’t have to feel bad about breaking that :D

And I agree that very few people are likely to have done that.

[gavinking]

This still needs a spec. I'm on it!

[gavinking]

Spec is done. Closing!