Index multi-dimensional arrays with a tuple?

[wsliang]

Would it make sense to allow tuples to index multi-dimensional arrays? For example, if A is a 2D matrix, then A[(2, 1)] would do the same thing as A[2,1].

The rationale for this is that sometimes it is more convenient to work with, and pass around, a single variable which contains all the indexing information, instead of d separate coordinates. For example, if loc is an (Int, Int) tuple representing the current location on a grid, it would be nice to be able to write:

loc = next_location(loc)
sum += A[loc]

instead of sum += A[loc...].

One potential cause for confusion is that this would mean A[[1:2]] and A[(1, 2)] would do different things. However, seeing as A[(1, 2)] is currently undefined, it seems like it should be ok.

[simonster]

This proposal would seem to conflict with the plans to use tuples as fixed-size immutable arrays (#5857), which would suggest that A[(2, 1)] should be equivalent to A[[2, 1]].

[JeffBezanson]

It might be possible in the future to wrap a tuple in an immutable type with no overhead.

[toivoh]

Still, you can already to A[(1,2)...] to get the same effect, and i suspect
it would equally efficient as well. (In both cases, as long as the size of
the tuple is known, you're good.)

[cbecker]

My opinion is that, regardless of what is done, it has to be clear what such operation means.
In that sense, A[myTuple...] is very intuitive.

[mbauman]

This can now be done by wrapping the tuple in a CartesianIndex. In fact, you can mix-and-match CartesianIndexes, integers, arrays, and arrays of CartesianIndex together seamlessly. And in most cases the wrapper can be optimized away. Between that and splatting, I think this is well covered.

[timholy]

It definitely is. One still might wonder about defining all the operations we allow on CartesianIndex for tuples, and then we wouldn't even need CartesianIndex. But that's a little scary too.

Semi-related: recent case in which we discuss indexing with tuples-of-ClosedIntervals, and had to define some slightly unconventional methods.