This was a great talk that touched on a lot of details with very cool proposed approaches, but I’m resolving the thread because I’m now more convinced about equal support for offset indices. What really tipped the scale was this comment pointing out that a DimensionMismatch error can be thrown for unmatched axes in general, and I figured it’s not much on top of DimensionMismatch errors being thrown for 1-based axes of different lengths.
I still think it would be easier to write (n÷2+1):n than ((end-begin+1)÷2+begin):end (maybe end-begin+1 could be given its own keyword to correspond to length?), but features have tradeoffs and require_one_based_indexing is always an option.