bucketCount = 0;

If other.Count() == 0 (because count == freeCount but bucketCount != 0), this could result in bad modFunctionIndex.

Refers to: lib/Common/DataStructures/BaseDictionary.h:153 in 1339070. [](commit_id = 1339070, deletion_comment = False)

        this->freeCount = 0;

Here modFunctionIndex missing update.

Refers to: lib/Common/DataStructures/BaseDictionary.h:302 in 1339070. [](commit_id = 1339070, deletion_comment = False)

            bucketCount = 0;

Missing modFunctionIndex update.

Refers to: lib/Common/DataStructures/BaseDictionary.h:311 in 1339070. [](commit_id = 1339070, deletion_comment = False)

        uint newBucketCount = SizePolicy::GetBucketSize(newSize, &modFunctionIndex);

In theory this is not safe. this->modFunctionIndex is changed while other variables are not (newSize, newBucketCount are using temporaries). If anything below failed in the middle we'd end up with bad modFunctionIndex. To be consistent and safer, consider also use a temporary "newModFunctionIndex" and be sure use it during rehash.

Refers to: lib/Common/DataStructures/BaseDictionary.h:1021 in 1339070. [](commit_id = 1339070, deletion_comment = False)

Please also fix WeaklyReferencedKeyDictionary::Resize() and Initialize(), both have the same unsafe pattern of changing modFunctionIndex before everything succeeds.

Any other dictionaries using modFunctionIndex having the same problem?

@jianchun please consider that the mod index set by previous dictionary was working for that dictionary.

@jianchun please consider that the mod index set by previous dictionary was working for that dictionary.

Yes it is a bug. But that's not the only problem. Several from my comments are apparent bugs too. The observation is that we have similar amount of failures in WeaklyReferencedKeyDictionary, which was not covered by current change. Why not do a thorough review and fix?

@jianchun Thank you for the review and pointers. Is it good to go?