Implement zero! for arrays

[fredrikekre]

Implements zero! to set all elements of an array to 0.

I was really surprised this wasn't a thing already, and since it is a very common thing to do I think it might be nice to have a special version of fill! for this purpose. I think it makes the code very readable and it is obvious what the function does.

julia> A = rand(2,2)
2×2 Array{Float64,2}:
 0.885381  0.234   
 0.943264  0.777748

julia> zero!(A)
2×2 Array{Float64,2}:
 0.0  0.0
 0.0  0.0

(I also really like in-place ! functions 😄 )

Comments?

[ararslan]

Typically imperative grammar is used when describing how functions act. So in this case it would be "Set all elements..." rather than "Sets all elements..." Should also probably be "the array" rather than just "array".

[StefanKarpinski]

This reads pretty clearly to me, tbh.

[stevengj]

I'm not sure we need a separate function for this; it's pretty easy to type fill!(A, 0)

[ararslan]

@stevengj That's a fair point, though for complete generality wouldn't it need to be something like fill!(A, zero(eltype(A)))? That's more annoying (though not entirely unreasonable).

[stevengj]

In most practical applications for this I suspect you will know that you have numbers compatible with 0 and can do x .= 0

[stevengj]

Or you will have some other scalar variable y and can do zero(y)

It would be useful to survey the packages to see where filling with zero actually shows up

[fredrikekre]

I'm not saying its difficult to use fill!(x, 0), but rather that it feels like zero is missing its eqivalent in-place version. If people don't think this is worth implementing then feel free to close this PR.

[stevengj]

The need for a lot of in-place operations is greatly reduced by the .= assignment syntax (new in 0.5). In fact, we might consider getting rid of fill!(x, v) (in favor of x .= v).

[johnmyleswhite]

I'm not saying its difficult to use fill!(x, 0), but rather that it feels like zero is missing its eqivalent in-place version.

You mean zeros, not zero, right? zero would be hard to remove, but I wish we would remove zeros at some point (along with trues, falses and ones). Don't think many others agree, though.

[fredrikekre]

You mean zeros, not zero, right?

Well, either one. Here I implemented zero! but could as well have been zeros! I guess. But as pointed out in comments above maybe this is not a good idea after all.

[fredrikekre]

1. Would this make more sense as an implementation of zeros!(in-place version of zeros) instead of zero! as this PR currently implements? I start to think that would make more sense, since AFAICT zeros is only implemented for arrays. (While zero is used for many types.) eg:

julia> A = rand(2,2)
2×2 Array{Float64,2}:
ze r0.56361   0.904141
 0.723398  0.87581 

julia> zeros!(A)
2×2 Array{Float64,2}:
 0.0  0.0
 0.0  0.0

(It should be noted that both zero(A) and zeros(A) return a zero'd copy of A, so in that perspective both zero! and zeros! would make sense as in-place implementations.)

2. Do people feel like this specialization of fill! is useful to implement in Base? If not, this PR can be closed.

Thanks for the input!

[JeffBezanson]

I don't think we need to add this function. I agree with @johnmyleswhite that we should rather go in the direction of minimizing this kind of redundancy. zeros and ones are sort of grandfathered in, but that's a debate for another day.

[fredrikekre]

I guess that settles this then. Thanks for the input!