Some indices generalizations for linalg/generic

[timholy]

This was also the occasion for #18031, but this version just keeps compatibility with what we currently have.

[andreasnoack]

Can you explain this a little more? Is length(AbstractArray) deprecated in favor _length(AbstractArray)? Should length and/or _length not be used with AbstractArrays anymore?

[timholy]

Just for julia 0.5. See http:https://docs.julialang.org/en/latest/devdocs/offset-arrays/.

[andreasnoack]

I see. In the documentation, your suggestion is to use length(linearindices(A)) instead of length(A) so would it make sense to use it here? Is the idea that in the future, when the code base has had some time to adjust to the new AbtractArray interface, length can again be used for extracting the number of elements in an array? If you stay with the _length, I think it makes sense to explain it in a comment.

[timholy]

It's equivalent, see how linearindices and _length are implemented. I guess the thing I like about _length is that it's search-and-replaceable and even @deprecateable (in case packages start calling it directly, though I'd strongly prefer if they just internally defined _length(a) = length(linearindices(a)).

I can add a comment if you like. Perhaps at the using Base: _length statement?

[tkelman]

@nanosoldier runbenchmarks(ALL, vs=":master")

[nanosoldier]

Your benchmark job has completed - possible performance regressions were detected. A full report can be found here. cc @jrevels

[andreasnoack]

I know but I think it would be useful with a comment such that, in the future, we'll know why we are using a non-exported and non-documented function to compute the number of elements in an array.

[timholy]

It's already there now, do you like it? Or should I move it somewhere else?

[andreasnoack]

I see. Thanks. I looked in the wrong file. It was something like that I had in mind.

[timholy]

Feel free to suggest further changes, or if none perhaps we can merge.