> A fix wouldn't be unwelcome but it seems it would have a good chance to cause performance regression - a lot more allocated values maybe on a lot of inner loops. I guess escape analysis might help avoid the allovations in the general case. ?
It seems unlikely unless the code is already incorrect (aka you're closing over or otherwise leaking the iteration variable).
But regardless of how I dislike the current loop's scoping this is a significant semantics change so it would obviously have to be opt-in (and it would hopefully come alongside making range loops less crummy e.g. with an actual iterator interface).
