At least mostly-fix internal issue #1898095 / public issue #2868.
The problem is that Scala code may have two variables with the same
name that are simultaneously active, and dx didn't expect that ever to
be the case. The fix is simply to catch another case where a local
variable might be considered to become superfluous due to having an
empty extent. However, this is really more of a workaround, as the
result will be that such code when attached to a debugger will only
show one of the two (or more) same-named variables as being active at
any given time.
The thing that prevents a more satisfactory solution is that the
optimization code can end up "splitting up" a single local into
multiple registers, and the only way we have to track them is by
name/type. So, even without Scala's rules, it was already possible to
find a same-named local in multiple registers, but with the status
quo, finding that really meant that a local was moving from one
register to another. The local variable table code handles that case
by automatically "ending" variables that appear to move, and the Scala
case looks just like that.
In the long run, we probably want to have a way to tag local variables
that isn't just name/type, but that will have to come later.
(As a bonus, I did some minor renaming for clarity while I was in the
territory.)
2 files changed