Change in generic_vecnorms

[blas-ko]

By default, generic_vecnorm of 1,2,Inf,-Inf,p, takes norm(v)^p (taking p as the general case which includes 1,2,Inf,-Inf). Normally, v is a Number so norm(v) just applies the

@inline vecnorm(x::Number, p::Real=2) = p == 0 ? (x==0 ? zero(abs(x)) : oneunit(abs(x))) : abs(x)

method.
This change is made because in some strange cases, v is not a Number but an Array and it's more convenient to take norm(v,p) instead of norm(v).

[andreasnoack]

I support this but see the previous discussion in #22945.

[blas-ko]

I believe, from what I read from #22945, that it's a matter of taste wheather to define the inner norm as just norm(v) or norm(v,p). I still think it's more convenient to define the inner norm as norm(v,p) becuase in any vector space where you would like to have a p-norm of an element, you should expect your element to have a p-norm defined too.

Nevertheless, I could do a workaround of this problem by defining

norm(x,p::Real=2) = norm( norm.(x[:],p), p)

in my specific case (which has to do with TaylorSeries package).

Hope this helps at all in the discussion.

[andreasnoack]

@blas-ko Could you give a more specific example of your use case?