There are severe inefficiencies in `toArray` (or, to be more precise, `rangeToArray`, which toArray calls). For small ranges, the inefficiencies hardly matter, and, for large ranges, the code was likely to exhaust memory before the execution speed mattered. But, with PR #3834 now merged into the codebase, the memory problem is much lessened, and there is now merit in looking into performance.
The code currently accesses each cell in the range. This PR changes it to access only cells which have been initialized - fewer cells processed, and many fewer function calls for those that are. For ranges which are densely filled, there will be little difference. However, for ranges which are lightly filled (a very common occurrence in issues which are raised for this project), the effect will be staggering. The sample file which accompanied 3834 saw an enormous reduction in memory consumption from 20GB to 32MB when toArray was used on it. The same file still took well over an hour to process even after the merge; after this PR, it is processed in approximately a second.
While testing this change, it became apparent that Collection/Cells does not handle column XFD (highest allowed) correctly. That is corrected and new tests added.
oleibman