Groovy 6, switch re-classification results in expression switches being classified as statements. What is the intended behaviour of yield inside an arrow style switch statement in groovy 6? Should yield be rejected at compile time or supported with early exit semantics? The current 6.0 behaviour seems buggy to me: https://groovyconsole.dev/?g=groovy_6_0&codez=eNpLSU1TKMhMztZIys_PSU3MU0jLSUzXVKjmUgCC4vLMkuQMBQ1DmAAIJCcWpyoYKujaIYmBQGaaggZEc2Vmak6KgnpaZlFxiTqKGqhMcWpyfl4KQqqWC0LWcnElFhenFpVAnFRSVJqqqWBrCzMKAGONLXI#4pwF8p6YI5

Hm, also the error message seems buggy to me. If you replace

yield

by

return

it says

switch expression does not support

return

While it should be something like

arrow-style switch does not support

return

Now that not every arrow-style switch is an expression.

Supporting early exit with yield in arrow switch statements would keep them backwards compatible with groovy5 switch expressions, at least for this case

Very much looks like a bug unless i'm missing something here. The same switch behaves differently if it is part of a function gwc-experiment.appspot.com/?g=groovy_6_0&codez=…

As I said, yes, i think it is a bug. This is not about "inside a function" but about "used as an expression". Even your usage in the function does not count as expression, but as statement and the usual last-statement of the function defines return value kicks in. If you add a

return

in front of the

switch

, it is an expression and behaves like expected. I still think the bug is, that the switch-statement in arrow-shape is considered an expression when it actually is a statement. It should complain that there is

yield

and instead allow and require

return

and would then also behave as expected.

Good catch — that's a regression, not intended. In 5.1.2 and 6.0.0-RC-1 your snippet gives 'first'; from RC-3 (incl. 6.0.0) it gives 'second'. When a trailing switch became a statement (GROOVY-12408), the parser started lowering

yield x

to a plain expression, so a non-final yield is silently a no-op and execution falls through. The same lowering also affects value-discarded switches, and

yield

inside a loop within an arm now fails to compile. Intended behaviour is that it's supported with early-exit semantics: the reclassification was only meant to change the encoding, and 4.x/5.x code had to use

yield

in block arms, so rejecting it would break working code. In return position yield should behave like

return

, otherwise like a

break

out of the switch. We'll raise a JIRA and target 6.0.1. Workaround meanwhile: write

return switch (...) { ... }

(keeps it a switch expression), or restructure the arm so the only yield is the last statement (if/else with a yield in each branch).

Nevertheless, the error message is also wrong, isn't it? I mean if you place

return

instead of yield and it says that in switch expression

return

is not supported. I learned that now with 6, it is not a switch expression just because it has arrow-shape. So I wonder, why does it tell about switch expression when it is a switch statement and also why is

return

not supported if it is now a switch statement. 😦

The GEP-19 PR (Groovy 7) has the return/break/continue fix 42d5425cb7. I'll backport that to Groovy 6. The error dates back to when all bodies were closures, so return would have been return from the closure.