question about outer constructors

[githubtomtom]

StaticArrays.jl/src/SArray.jl

Line 32 in 3a9291d

@generated function (::Type{SArray{S, T, N}})(x::Tuple) where {S <: Tuple, T, N}

in above I see the use of generated function for outer constructors.

what is the difference (and disadvantage) using the following instead?

SArray{S, T, N}(x::Tuple) where {S<:Tuple, T, N} = SArray{S, T, N, tuple_prod(S) }(x)

also, in

StaticArrays.jl/src/SDiagonal.jl

Line 11 in 3a9291d

@inline (::Type{SDiagonal{N,T}})(a::Tuple) where {N,T} = Diagonal(SVector{N,T}(a))

why don't we define the following?

@inline SDiagonal{N,T}(a::Tuple) where {N,T} = Diagonal(SVector{N,T}(a))

thanks.

[c42f]

what is the difference (and disadvantage) using the following instead?

The fact this is @generated is likely a historical artifact from a time when the Base.@pure annotation didn't exist (keep in mind that StaticArrays dates from julia 0.5 or so). You'd need to read the history to know for sure. At this stage we could very likely replace this with a normal function.

why don't we define the following? @inline SDiagonal [...]

Again, likely historical (due to the syntax changes in the julia 0.6 -> 1.0 transition).

[githubtomtom]

thanks. for the first case, maybe using generated function could push the computation of tuple_prod(S) into compilation time?

On Jan 30, 2020, at 1:36 PM, Chris Foster ***@***.***> wrote: what is the difference (and disadvantage) using the following instead? The fact this is @generated is likely a historical artifact from a time when the ***@***.*** annotation didn't exist (keep in mind that StaticArrays dates from julia 0.5 or so). You'd need to read the history to know for sure. At this stage we could very likely replace this with a normal function.

but the second one is really mysterious to me.

…

why don't we define the following? @inline SDiagonal [...] Again, likely historical (due to the syntax changes in the julia 0.6 -> 1.0 transition). — You are receiving this because you modified the open/close state. Reply to this email directly, view it on GitHub <#721?email_source=notifications&email_token=AIABXJO264VVIBR6DNQD2ODRAJRPTA5CNFSM4KNEJWMKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEKJX36A#issuecomment-580091384>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AIABXJNEDGKY4J6O5NA3GMLRAJRPTANCNFSM4KNEJWMA>.

[c42f]

maybe using generated function could push the computation of tuple_prod(S) into compilation time

Indeed, that's exactly what that $ does there: compute tuple_prod as part of the generator and splice the result into the AST. But it's also very likely unnecessary because tuple_prod is now marked @pure (so the compiler will do the same thing without the @generated compile time overhead).

the second one is really mysterious to me

The syntax for constructors of parametric types was very different in julia 0.5. If you read the old manuals for julia 0.6 vs 1.0 you may understand.

As I said: it's likely purely historical. I think your proposed rewording would be equivalent but clearer in the new syntax.