Thanks, I rephrased "returns" to "received". I meant the argument received.

In your class B above, you are implicitly assuming that individual indices cannot be tuple:

• B()[(1,2) (3,4)] will detect it is multi-index (it receives a len 2 nested tuple),

• but B()[(1,2)] will believe it received 2 indices, while it actually was passed only 1 index of size 2.

I am comparing to function, foo(**args) vs. foo[**args], because they are mathematically equivalent (indexing is a function from index space), and share a similar "argument (un)packing" functional feature (and because in the end both are written as methods)

I understand that 1-length tuples have been identified with their scalar value (its single entry), but I find it weird and can't understand why

• it makes packing quite different than for usual functions

• You can not tell if arguments are "n indices" versus "a single n-tuple index", but python finds that a[1] and a[1,] are different. So if I cannot distinguish and base a decision on, why would others ?

• I do not see the problem if getitem was always receiving a tuple regardless of the indexer dimension (keep scalar index wrapped)

So I see what you lose and don't guess what you gain.