Why is ImmixAllocator reset twice?

[wks]

In src/plan/immix/mutator.rs:

pub fn immix_mutator_prepare<VM: VMBinding>(mutator: &mut Mutator<VM>, _tls: VMWorkerThread) {
    let immix_allocator = unsafe {
        mutator
            .allocators
            .get_allocator_mut(mutator.config.allocator_mapping[AllocationSemantics::Default])
    }
    .downcast_mut::<ImmixAllocator<VM>>()
    .unwrap();
    immix_allocator.reset();
}

pub fn immix_mutator_release<VM: VMBinding>(mutator: &mut Mutator<VM>, _tls: VMWorkerThread) {
    let immix_allocator = unsafe {
        mutator
            .allocators
            .get_allocator_mut(mutator.config.allocator_mapping[AllocationSemantics::Default])
    }
    .downcast_mut::<ImmixAllocator<VM>>()
    .unwrap();
    immix_allocator.reset();
}

The immix_allocator.reset() invocation does the following:

impl<VM: VMBinding> ImmixAllocator<VM> {
    pub fn reset(&mut self) {
        self.bump_pointer.reset(Address::ZERO, Address::ZERO);
        self.large_bump_pointer.reset(Address::ZERO, Address::ZERO);
        self.request_for_large = false;
        self.line = None;
    }
}

This method is idempotent, and we don't touch the allocators during GC. Why do we need to reset the ImmixAllocator in both prepare_func and release_func?

p.s. This behaviour exists in JikesRVM, and it has behaved like this ever since Immix was introduced into JikesRVM MMTk.

[wks]

One possible explanation is that it intended to reset ImmixCollector.copy (the counterpart of ImmixCopyContext::copy_allocator in Rust MMTk) during the PREPARE phase because it is used for evacuating objects. But it mistakenly reset the ImmixAllocator of the mutator, instead.