it = map.erase(it) potentially passes object twice about robin-hood-hashing HOT 6 CLOSED

I've finally fixed the issue. As explained above, I've added a buffer for the overflow, so there's no wrap around any more. The added benefit is that a lot less & mMask are necessary. The disadvantage is that for the worst case of bucket size 256, I need an additional 256 buckets. That's not too bad though. For larger maps I still only need 256 extra buckets, and for small maps the buffer is also smaller.

from robin-hood-hashing.

The iterator pos must be valid and dereferenceable. Thus the end() iterator (which is valid, but is not dereferenceable) cannot be used as a value for pos.

from robin-hood-hashing.

@ktprime, this is not about the end() iterator, this is about the backward-shift-deletion which shuffles elements around when erasing an element. See Tessil/robin-map#20

from robin-hood-hashing.

Any progress on this or any workaround? If not, is there any alternative unordered_map that you recommend? Thanks.

from robin-hood-hashing.

Unfortunately, not yet. This is an edge case that's very difficult to solve in an efficient way. If you need the standard behavior, switch to std::unordered_map.

from robin-hood-hashing.

I think of all the possible solutions to this problem, the least bad is to ditch the wrap around and add a buffer to the end of the data array.

Since mInfo is limited, The size of that buffer can also be very limited: std::min(mMaxNumElementsAllowed, 128).

from robin-hood-hashing.