Dot syntax involving CartesianIndex and Tuple

[juliohm]

I would like to understand how this operation should be done instead?

start = CartesianIndex(1,2,3)
offset = (1,2,3)
@. start + offset

ERROR: iteration is deliberately unsupported for CartesianIndex. Use I rather than I..., or use Tuple(I)...

It is often the case that we need to offset from a CartesianIndex with a Dims or Tuple object. I could convert the Tuple to a CartesianIndex and avoid the dot syntax, or I could do @. start.I + offset to get access to the underlying Tuple. It is not that elegant though to have to think about Tuple vs. CartesianIndex.

Should the syntax above be allowed?

[KristofferC]

or I could do @. start.I + offset to get access to the underlying Tuple.

The error message says to use Tuple.

[juliohm]

Thanks, I am fixing the code locally. I guess my question is more on the lines of how useful this restriction with CartesianIndex + Tuple is. I know it educates users, but it also blocks use cases like the one above which seem reasonable to me?

[nalimilan]

AFAIK the error is there because it would be very slow anyway? Better throw than trap people.

Cc: @timholy

[timholy]

Not quite sure what to do here. #26227 has the most nuanced thinking I know of on this topic.

[timholy]

Looking at this again, I should have pointed out the obvious: why do you need to offset with a tuple? Just convert it to a CartesianIndex:

julia> i2 = CartesianIndex(3,5)
CartesianIndex(3, 5)

julia> i1 = CartesianIndex(1,1)
CartesianIndex(1, 1)

julia> i1+i2
CartesianIndex(4, 6)

The case of what we should do about scalar increments is harder to figure out.

[juliohm]

Hi Tim, In many cases we offset with Dims objects. I could convert them to CartesianIndex but I was wondering why we dont have methods to do this conversion for the user? Is there any penalty in converting tuples to CartesianIndex or in using idx.I?

…

On Fri, Oct 12, 2018, 06:40 Tim Holy ***@***.***> wrote: Looking at this again, I should have pointed out the obvious: why do you need to offset with a tuple? Just convert it to a CartesianIndex: julia> i2 = CartesianIndex(3,5)CartesianIndex(3, 5) julia> i1 = CartesianIndex(1,1)CartesianIndex(1, 1) julia> i1+i2CartesianIndex(4, 6) The case of what we should do about scalar increments is harder to figure out. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#29597 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ADMLbYgO4EU16z6v3M0BfEvwXCI7vaZIks5ukGOdgaJpZM4XXaXK> .

[timholy]

If you're worried about penalties, instead of asking others just try benchmarking yourself.

As far as methods to do conversions, CartesianIndex is conceptually a scalar and tuple is a container. That makes it unclear what to do: should you generate 3 CartesianIndexes, each with a broadcasted scalar-add? Or add them elementwise?