this produces a huge performance improvement for queues with large internal tables.
an internal table of large size may appear if the array had lots of elements inserted into it and later deleted. this resulted in major performance losses for the reader of the elements, as zend_hash_internal_pointer_reset_ex() had to scan through many IS_UNDEF offsets to find the actual first element.
there are two ways to attack this problem:
1) reallocate the internal table as elements are deleted to reduce the internal table size - this proved to be relatively ineffective 2) track the start and end of the hashtable to avoid repeated scans during every shift() call - this is the approach taken in this commit, and provides major performance benefits
the test case written in #42 now runs to completion substantially faster, without any performance degradation.
more tests are needed to ensure that this works fully as intended, but I chose to take the safe route with invalidating vs updating the offsets, so I think it should be good.
this produces a huge performance improvement for queues with large internal tables.
an internal table of large size may appear if the array had lots of elements inserted into it and later deleted. this resulted in major performance losses for the reader of the elements, as zend_hash_internal_pointer_reset_ex() had to scan through many IS_UNDEF offsets to find the actual first element.
there are two ways to attack this problem: 1) reallocate the internal table as elements are deleted to reduce the internal table size - this proved to be relatively ineffective 2) track the start and end of the hashtable to avoid repeated scans during every shift() call - this is the approach taken in this commit, and provides major performance benefits
the test case written in #42 now runs to completion substantially faster, without any performance degradation.
more tests are needed to ensure that this works fully as intended, but I chose to take the safe route with invalidating vs updating the offsets, so I think it should be good.