1) singleton PMCs like Random or Env got destroyed during DOD runs
2) and a really nasty one:
During (overridden vtable) method calls, the caller doesn't know that a
method is called, so the return continuation in P1 wasn't preserved in
the context - the caller just assumes, that its preserved. Now when a
method calls another sub or method, a new return continuation is created
in P1, which unachored the previous one - it got destroyed during a DOD
run. A DOD run is easily triggered during tracing, so this matches the
symptoms of recently reported DOD bugs.
Jens Rieks had mailed me his EBNF parser some time ago, which happend to
expose that bug. After ++some hours, I could eventually nail it down to
what is now t/op/gc_10.pasm.
leo
