Currently there is a fairly tight integration between the runtime and snmalloc. We should add an abstraction layer here that allows us to more easily replace the allocator.
This could be very useful for systematic testing. ABA problems require rapid recycling of memory, but this is unlikely with snmalloc and very unlikely with SCUDO inside Asan. This means that systematic testing will seldom discover ABA bugs that involve allocations being recycled.
We should provide a layer that can be overridden in different configurations to increase the likelihood of discovering ABA problems.
Currently there is a fairly tight integration between the runtime and snmalloc. We should add an abstraction layer here that allows us to more easily replace the allocator.
This could be very useful for systematic testing. ABA problems require rapid recycling of memory, but this is unlikely with snmalloc and very unlikely with SCUDO inside Asan. This means that systematic testing will seldom discover ABA bugs that involve allocations being recycled.
We should provide a layer that can be overridden in different configurations to increase the likelihood of discovering ABA problems.