Daniel Karrasch

Thanks for the notice. In that thread, I read "organizational tokens". Does that mean we can solve this at the organizational (i.e., JuliaLinearAlgebra) level? I'll try to take a look...

I think it's rather the other way around: it's pure luck (and due to a specific implementation) that `pairwise!` works here: ```julia julia> SqEuclidean()(vec(X), vec(Y)) -31616 ``` But we may...

But what would that help? For your `Int16` example, you'd get `result_type(x, y) = Int16`, and there you get your overflow. To prevent it, you would need to know the...

To prevent the overflow, you would need to promote your input (as you can see from your OP), which you can do as the user/package author. I don't think we...

> Also at the moment `pairwise!()` and `colwise!()` behaviors are different. That's why I said the fact that `pairwise!` works is rather luck. It's due to this performance optimization, which...

> I don't understand why this is needed.. I think this is because some points may be far from the query point, represented by some point in the canonical image,...

I may not have time for this, but this should be merged once tests pass, and then bumped on the release-1.12 branch over at julialang.

It's an `@example` block, generated with Literate.jl. An example could be ```julia ```@example custom rand(2,2)'ones(3) ```

I changed things such that there are two methods of each signature type: first a legacy method, that has `MulAddMul`s in their final argument. Those get destructured and the call...

I don't think this can be reasonably resolved. Reading a diagonal element should just read `D.diag`. Reading an off-diagonal element may not be representable by the same type, as in...