Initially closed as a "possible misunderstanding", then acknowledged as broken on 1.9 and fixed on 2.0, then broken again on 2.0, and the issue was left closed. Seems pretty clear to me that it's not the wanted behavior.
Thanks for info, Dan.
Glad to see that at least someone is already working on it :), and it probably will be fixed in ver2.
Well, unfortunately that’s not the case — my point was that, even though this was acknowledged as a bug in 1.9 that had been fixed in 2.0, the GitHub issue was left closed after I reported that it had regressed again in 2.0. And given that none of the simple test cases I’ve provided for this and other variable precedence bugs have been incorporated into the test suite, there’s no reason to think that, even if it is fixed, it will stay that way.