Require all tuples in eachindex to have the same length.

[navdeeprana]

Potential fix for #47898

[Moelf]

Suggested change

	all(==(lent), t2) || throw_eachindex_mismatch_indices((), t, t2...)
	all(t -> length(t)==lent, t2) || throw_eachindex_mismatch_indices((), t, t2...)

[Moelf]

yeah sorry I was stupid

[navdeeprana]

any updates on this pr?

[Moelf]

basically my first proposed change was bad, so you need to revert to I believe

all(t -> length(t)==lent, t2)

[navdeeprana]

basically my first proposed change was bad, so you need to revert to I believe

all(t -> length(t)==lent, t2)

I did not implement the changes you suggested, so I thought it was good to go.

[Moelf]

@oscardssmith did: 6e3e170

so you need to fix that.

You can always see the full change this PR current has by heading towards: https://github.com/JuliaLang/julia/pull/48125/files

[navdeeprana]

Okay done.

[adienes]

what is the purpose of the first ::Tuple{} parameter?

[timholy]

I think it resolves dispatch:

julia> methods(Base.throw_eachindex_mismatch_indices)
# 2 methods for generic function "throw_eachindex_mismatch_indices" from Base:
 [1] throw_eachindex_mismatch_indices(::IndexCartesian, inds...)
     @ abstractarray.jl:324
 [2] throw_eachindex_mismatch_indices(::IndexLinear, inds...)
     @ abstractarray.jl:321

It's not obvious to me that one couldn't use the IndexLinear dispatch but this seems fine.

[timholy]

I think it resolves dispatch:

julia> methods(Base.throw_eachindex_mismatch_indices)
# 2 methods for generic function "throw_eachindex_mismatch_indices" from Base:
 [1] throw_eachindex_mismatch_indices(::IndexCartesian, inds...)
     @ abstractarray.jl:324
 [2] throw_eachindex_mismatch_indices(::IndexLinear, inds...)
     @ abstractarray.jl:321

It's not obvious to me that one couldn't use the IndexLinear dispatch but this seems fine.

[timholy]

Let's rerun CI here (by closing and reopening) and then it looks good to merge.

[timholy]

This looks like it worked (I haven't checked very carefully but the failures seem unrelated). If you're re-purposing the LinearIndices variant I think you should delete the method you're not calling.

[navdeeprana]

Done.

.. So it's OK to define methods that call other methods that haven't yet been defined, as long as we don't actually trigger that call until later.

So that's why it doesn't fail during build with IndexLinear().

[timholy]

I'm tempted to merge this but a bit concerned about the failures on apple. They seem identical to one another, and in a quick survey of recent PRs I don't see the same error. You aren't by any chance running on apple, are you @navdeeprana? If so, can you run the same test locally and check?

[navdeeprana]

No, I am on linux. :(

[timholy]

Let's just close and reopen and see if it happens again.

[timholy]

Apple failed again, perhaps suggesting it's a consequence of this PR. If so, we can't merge this until that's fixed. Can anyone with a mac investigate why?