Unknown failure in CI on Risc-V

[mjp41]

For #617 the CI failed a couple of times on RISC-V in the following way.

24/40 Test #25: func-first_operation-check .........Subprocess aborted***Exception:  11.82 sec
......
assert fail: (back != nullptr) || (address_cast(front.load()) == address_cast(&stub)) in /home/runner/work/snmalloc/snmalloc/src/snmalloc/backend_helpers/../mem/remoteallocator.h on 69 

/home/runner/work/snmalloc/snmalloc/build/func-first_operation-check(_ZN8snmalloc8PALPOSIXINS_8PALLinuxEXadL_Z6writevEEXadL_Z5fsyncEEE17print_stack_traceEv+0x3c)[0x4000036904]

https://github.com/microsoft/snmalloc/actions/runs/5221697554/jobs/9455200536

The PR did not affect the code in anyway. The failure has not been seen outside of this PR, and a separate branch to exercise this more has not resulted in any failures. I'm raising this issue in case we see this again.

[nwf-msr]

Not to ask a stupid question, but... is that assert actually valid in the presence of interleaving? It looks like it could trip if, say, the call to invariant in dequeue at

snmalloc/src/snmalloc/mem/remoteallocator.h

Line 147 in d9f5bd0

invariant();

evaluates back prior to the first enqueue taking it away from nullptr at

snmalloc/src/snmalloc/mem/remoteallocator.h

Lines 117 to 118 in d9f5bd0

	freelist::QueuePtr prev =
	back.exchange(capptr_rewild(last), std::memory_order_acq_rel);

but then evaluates front after enqueue set it at

snmalloc/src/snmalloc/mem/remoteallocator.h

Line 127 in d9f5bd0

front.store(capptr_rewild(first));

[mjp41]

I think you are correct. Would switching the disjuncts work? I.e. check back second.

[mjp41]

Closed by #619