Deprecate vectorized big methods in favor of compact broadcast syntax #18512
Conversation
| broadcast(::typeof(big), r::StepRange) = big(r.start):big(r.step):big(last(r)) | ||
| broadcast(::typeof(big), r::FloatRange) = FloatRange(big(r.start), big(r.step), r.len, big(r.divisor)) | ||
| function broadcast(::typeof(big), r::LinSpace) | ||
| big(r.len) == r.len || error(string(r, ": too long for ", big)) |
There was a problem hiding this comment.
Might be better as an ArgumentError?
There was a problem hiding this comment.
Good call. Changing here and in #18495. Thanks!
| @@ -541,7 +541,7 @@ function float{T}(A::AbstractArray{T}) | |||
| convert(AbstractArray{typeof(float(zero(T)))}, A) | |||
| end | |||
|
|
|||
| for fn in (:float,:big) | |||
| for fn in (:float,) | |||
There was a problem hiding this comment.
You could probably just deloopify this since it's iterating over a single value, or so I would assume.
|
You always do such great work, Sacha! |
|
Thanks for the kind words @ararslan! 😊 |
| @@ -828,8 +828,8 @@ function complex{T}(A::AbstractArray{T}) | |||
| convert(AbstractArray{typeof(complex(zero(T)))}, A) | |||
| end | |||
|
|
|||
| big{T<:Integer,N}(A::AbstractArray{Complex{T},N}) = convert(AbstractArray{Complex{BigInt},N}, A) | |||
| big{T<:AbstractFloat,N}(A::AbstractArray{Complex{T},N}) = convert(AbstractArray{Complex{BigFloat},N}, A) | |||
| broadcast{T<:Integer,N}(::typeof(big), A::AbstractArray{Complex{T},N}) = convert(AbstractArray{Complex{BigInt},N}, A) | |||
There was a problem hiding this comment.
We could possibly document that broadcast can alias, like we do for convert. #17519
There was a problem hiding this comment.
Ref. ongoing discussion in #18495, and particularly #18495 (comment). Agreed, that is :).
|
Rebased. ( |
|
Subsumed by #19791. |
This PR deprecates all (?) remaining vectorized
bigmethods in favor of compact broadcast syntax. Ref. #16285, #17302, and #18495.As with #18495, to work around #18462 I made the
Broadcastmoduleimportrather thanexportbroadcast. I will correct this if / when a better solution for #18462 appears.I imagine #17302 was a bit of a bear to review. So for devectorization of manually vectorized functions, I'm shooting for small, atomic PRs with the hope that that will ease review. If instead I should bundle them, please let me know! Best!