cmd/compile: stack overflow instead of initialization loop error

[griesemer]

https://play.golang.org/p/7e5OO_qt0m correctly reports an initialization loop error. But cmd/compile (1.10 Beta) crashes with a stack overflow.

Superficially looks like a new error in the inlining code.

[mdempsky]

The regression was introduced in c4b65fa.

I think the issue is the new code in hairyVisitor doesn't check if the named closure being called is a global or local variable. In inlnode, there's extra code in reassigned to make sure we don't try inlining a call to a global closure. That code was missed when the logic was copied into hairyVisitor.

[gopherbot]

Change https://golang.org/cl/83538 mentions this issue: cmd/compile: remove broken inlining accounting code