GCRoots are relatively expensive: they are scanned twice on every GC and they are - for the time being - scanned conservatively even if most heap storage can be, or will be, scanned precisely. We suspect that there is some use of GCRoots where none is needed. The VM shell is pretty clean at this point but the Flash Player needs vetting. This bug exists to make that work visible, even if details can't always appear here.

Just evidence of false pointers in roots and the problems they bring: in a different bug, Tommy makes this remark: "For instance, one bug we've had is where a GCRoot doesn't initialize itself and it contains values from a previous allocation that prevent the GCRoot from ever being deleted (sometimes GCRoot deletion happens in a GCFinalizedObject dtor and if the root contains a pointer to that object then presto ... memory leak)."

It is known that the Player creates roots to hold onto event-related storage that makes a trip through native code where it's not visible to the GC. The lock/unlock API introduced in bug #589172 should be used to reengineer that code so that it no longer uses roots.

This bug pertains both to the VM and to the Player.

No assignee, updating the status.

No assignee, updating the status.

No assignee, updating the status.

Tamarin is a dead project now. Mass WONTFIX.

Tamarin isn't maintained anymore. WONTFIX remaining bugs.