benchmarking

can replace method body with

this(null, lowerBound, upperBound, hiddenColumns);

I changed the VisibleColsCollection constructor, though it makes it inconsistent with VisibleSeqsCollection (which actually needs AlignmentI)

We are going around in circles here. I already more or less considered this - it is what I wanted to do when I was looking at RoaringBitmaps (which are a souped up Bitset implementation). It might be a lot easier to do now that HiddenColumns is substantially cleaned up but I think there are still issues to consider.

In terms of memory: Current Java implementation of Bitset uses relatively large amounts of memory with only a few entries, if those entries happen to be at the high end (see e.g. http://java-performance.info/bit-sets/ ) because it has to allocate an array to hold all the unset values up to the highest set entry. This seems to me to be quite a likely use case in Jalview. This is especially a consideration if we are going to plan to use multiple alignment-wide Bitsets (for hidden columns, column selections, gaps, ... more?)

In terms of performance: I didn't test Bitset, but RoaringBitmaps, however they are meant to be more performant so it would be surprising if Bitset were much better. Performance was largely worse or similar to the pre-existing implementation. One of the things which made the performance worse was poor performance of the cardinality calculation (relative to the implementation then, of calculating the full sum over the list of ranges) and inversion operations - both of which, unfortunately, are valuable for simplification of the code.

I do agree with you about re-use for ColumnSelection and/or gaps, probably with another iteration over HiddenColumns to get it into suitable shape. The re-use options for HiddenColumns have only really become apparent/possible with the recent clean up, or I would have treated this as a redesign rather than a refactor. I think now the plan for re-use ought to be done as a proper design task, considering the implications and possibilities, with documented decisions and discussed before implementation, rather than a post-hoc design change via code review.

Agree about not modifying the passed in BitSet parameter and will try to fix. But the hideColumns(BitSet toHide, int start, int end) is actually different to hideColumns(BitSet) as the former also clears all the columns between start and end first; the latter just merges in the new hidden columns from the BitSet. Clearly that needs to be made clear in the Javadoc so I'll fix that too.

Superseded by other changes