Define promotion and conversion for CartesianIndices and LinearIndices #28084
Conversation
timholy
force-pushed
the
teh/cartinds_promotion
branch
from
d93aef9
to
63fe543
Compare
timholy
changed the title
base/indices.jl
Outdated
| @@ -357,6 +357,14 @@ LinearIndices(inds::NTuple{N,Union{<:Integer,AbstractUnitRange{<:Integer}}}) whe | |||
| LinearIndices(map(i->first(i):last(i), inds)) | |||
| LinearIndices(A::Union{AbstractArray,SimpleVector}) = LinearIndices(axes(A)) | |||
|
|
|||
| promote_type(::Type{NTuple{N,R1}}, ::Type{NTuple{N,R2}}) where {N,R1<:AbstractUnitRange,R2<:AbstractUnitRange} = | |||
| NTuple{N,promote_type(R1,R2)} | |||
There was a problem hiding this comment.
What makes tuples of unit ranges special compared to other tuples to justify this? (Other than that this comes handy in this PR?)
There was a problem hiding this comment.
Fair enough. But it's not wrong either. However, I could hide it behind a private method if you really think that's better.
There was a problem hiding this comment.
This method is running into some trouble too with ambiguity (really, typevar) tests. But I think it's a bug, see #28086.
There was a problem hiding this comment.
I'd lean towards making this a private method, too.
There was a problem hiding this comment.
Yeah. For fun I played with making it really general,
index 78a1954..f260cbf 100644
--- a/base/promotion.jl
+++ b/base/promotion.jl
@@ -236,6 +236,11 @@ promote_rule(::Type{Any}, ::Type{<:Any}) = Any
promote_rule(::Type{<:Any}, ::Type{Any}) = Any
promote_rule(::Type{Any}, ::Type{Any}) = Any
+function promote_rule(::Type{Tuple{R1,Vararg{R1,N}}}, ::Type{Tuple{R2,Vararg{R2,N}}}) where {R1,R2,N}
+ R = promote_type(R1, R2)
+ Tuple{R,Vararg{R,N}}
+end
+Somewhat surprisingly, only one test fails: https://github.com/JuliaLang/julia/blob/master/test/abstractarray.jl#L653
But I'll make it a private method, because that really would be a scary change.
timholy
force-pushed
the
teh/cartinds_promotion
branch
from
63fe543
to
28b8f77
Compare
Interesting that we hadn't yet noticed the lack of such fundamental methods. Porting packages is good!