Vladimir Lipskiy <[EMAIL PROTECTED]> wrote: > I surmise that there is something which set up > interpreter->has_early_DOD_PMCs = 1 > before the test runs.
Yep, probably the stdio PMCs. These (or better all) are inititalized currently with: PObj_needs_early_DOD_SET(pmc); interpreter->has_early_DOD_PMCs = 1; which isn't quite correct. The stdio PMCs should better live somewhere in the interpreter globals and being marked (as long as the streams are open). According to Dan, Parrot shouldn't set IO PMCs to "needs_early_DOD", the HLL should emit a C<needs_destroy> opcode if needed. So that tests are probably ok. leo