Unreachable child regions are leaked on freeze about verona HOT 3 CLOSED

I may have hit this same bug in my own compiler as I work to make it target the verona runtime - I hit a leak in a similar situation recently and hadn't figured out how to fix it yet.

But of course I assumed that it was my codegen that was doing something wrong 😁 (which may still be the case)

from verona.

It is slightly worse than

From a quick look at rt/region/freeze.h, there should be a find_iso_fields call somewhere, probably in the while (!to_dealloc.empty()) loop, followed by a loop to release the found regions.

The find_iso_fields follows fields, so you can't have deallocated any objects at this point. It could be we find the iso fields at:

              p->finalise();
              to_dealloc.push(p);

and then deallocate the sub-isos later. That assumes finalisers have read-only access to the state. Otherwise, this will require two passes of the to_dealloc objects.

from verona.

Closed by #90

from verona.