DataFrames should be indexable by CartesianIndex{2}

[ExpandingMan]

It should be possible to do, for example

for idx ∈ CartesianIndices(size(df))
    println(df[idx])
end

[nalimilan]

I guess that would be consistent, but in what situations would it be useful?

[ExpandingMan]

It's just a standard way to iterate over things. I for one use it more and more. It's already column major so it's a good way to iterate. If you ever are iterating over a matrix and a dataframe at the same time, it might be useful.

I'd be happy to make the PR, when I get to it.

[nalimilan]

Iterating over a data frame and a matrix at the same time is unlikely to be a good idea given how slow it's going to be compared to iterating over columns...

[ExpandingMan]

Well, they are both column major so if you have to iterate over the dataframe anyway I don't really see what harm it would do (other than typing issues of course).

Regardless, is there some reason you are opposed to making dataframes indexable by CartesianIndex?

[nalimilan]

Well the type-instability issue is a big problem in terms of performance. I'm not really opposed to supporting cartesian indexing, I'm just not sure it's really useful.

[bkamins]

I think it is not a problem to have it for AbstractDataFrame and possibly for DataFrameRow also if we want a complete coverage.

[bkamins]

We now allow CartesianIndex{2} indexing so I am closing this.

[bkamins]

(we do not provide CartesianIndices though. If you feel it should be added please reopen and it can be discussed.