WIP Return `SUnitRange` from `indices(::StaticArray)` #150

andyferris · 2017-04-27T05:35:34Z

Needs testing.

Eventually we might be able to move to generic indexing like OffsetArrays. But even without that, it's nice to be able to discuss and introspect the indices (at compile time).

@timholy are there some caveats to doing this? (I already ran into a few surprising things - perhaps you know of any other gotchas?)

Needs testing

c42f · 2017-04-27T05:42:46Z

Looks sensible so far.

andyferris · 2017-04-27T05:46:59Z

I somehow broke a macro with this code...

julia> macroexpand(:(@SArray([1 for x = SVector(1,2), y = SVector(1,2,3)])))
:($(Expr(:error, StackOverflowError())))

andyferris · 2017-04-27T05:54:55Z

whow - so much stack overflow

[119997] != at ./operators.jl:129 [inlined]
[119998] ==(::StaticArrays.SUnitRange{1,2}, ::StaticArrays.SUnitRange{1,2}) at ./abstractarray.jl:1513
[119999] ==(::Tuple{StaticArrays.SUnitRange{1,2}}, ::Tuple{StaticArrays.SUnitRange{1,2}}) at ./tuple.jl:267

andyferris · 2017-04-27T06:06:25Z

Getting deeper... now it's reduced_indices0 and conversions to Base.Slice. Oh dear.

andyferris · 2017-04-27T06:12:36Z

it appears to me that we should make AbstractUnitRange into a trait...

andyferris · 2017-04-27T06:26:38Z

The alternative here is to define a new AbstractUnitRange which is a static version of OneTo, such that these are the indices of a static array.

(My eventual "take over the world" goal here is to use a trait to describe if an array is statically sized by asking the question if all it's indices are static, and organize dispatch around this).

timholy

💯 I think this is a great idea, independent of the question of whether one should support non-1 indexing. (Yes, I want it. For example, in ImageFiltering I'd like to define the Sobel filter, which can be factored along individual dimensions into fixed 3-vectors, as having indices -1:1.)

timholy · 2017-04-27T09:12:39Z

src/SUnitRange.jl

+(==)(::SUnitRange{1, L}, r::Base.OneTo) where {L} = L == r.stop
+(==)(::SUnitRange, ::Base.OneTo) = false
+(==)(r::Base.OneTo, ::SUnitRange{1, L}) where {L} = L == r.stop
+(==)(::Base.OneTo, ::SUnitRange) = false


You may also want a variant for generic AbstractUnitRange.

timholy · 2017-04-27T09:14:08Z

src/SUnitRange.jl

+next(r::SUnitRange, i) = (r[i], i+1)
+done(r::SUnitRange{Start, L}, i) where {Start, L} = i > L
+
+@pure Base.UnitRange{Int}(::SUnitRange{Start, L}) where {Start, L} = Start : (Start + L - 1)


Any particular reason this needs to be @pure?

timholy · 2017-04-27T09:21:18Z

src/SUnitRange.jl

+done(r::SUnitRange{Start, L}, i) where {Start, L} = i > L
+
+@pure Base.UnitRange{Int}(::SUnitRange{Start, L}) where {Start, L} = Start : (Start + L - 1)
+@pure Base.Slice(::SUnitRange{Start, L}) where {Start, L} = Base.Slice(Start : (Start + L - 1)) # TODO not optimal


This is not wrong but is perhaps pushing the limits. Base.Slice is a (deliberately) limited version of IdentityRange, and we only say that two arrays are equal if they have both the same values and the same indices. Since this is a constructor rather than a conversion, you might be able to get away with this, but I'd also add

convert(::Type{Base.Slice}, r::SUnitRange{1,L}) = Base.Slice(r) convert(::Type{Base.Slice}, r::SUnitRange) = error("$r cannot be converted to Base.Slice, consider `Base.Slice(r)` instead")

(convert has an implicit promise of preserving at least approximate equality relationships, see the many discussions about whether it's OK to strip physical units e.g., convert(Int, 3meter).)

timholy · 2017-04-27T09:22:04Z

src/SUnitRange.jl

+@pure Base.UnitRange{Int}(::SUnitRange{Start, L}) where {Start, L} = Start : (Start + L - 1)
+@pure Base.Slice(::SUnitRange{Start, L}) where {Start, L} = Base.Slice(Start : (Start + L - 1)) # TODO not optimal
+
+function Base.checkindex(::Type{Bool}, ::SUnitRange{Start, L}, r::UnitRange{Int}) where {Start, L}


Any particular reason not to define one method for AbstractUnitRange?

timholy · 2017-04-27T09:31:00Z

it appears to me that we should make AbstractUnitRange into a trait...

Oh, wow, I see what you mean. Yikes, this is indeed an issue. I think your plan to define a whole new static type that subtypes from AbstractUnitRange is probably the best solution for now.

andyferris · 2017-04-27T10:22:21Z

Cool - thanks Tim for the feedback! You've confirmed everything I thought you would say ;)

timholy · 2017-04-27T10:44:44Z

I take predictability as a complement. (In great contrast to my current president...)

vlc1 · 2020-01-15T16:08:45Z

I'm really interested in using SUnitRanges, and was wondering whether this PR had led to anything?

Also, is there a reason why SOneTo is a subtype of AbstractUnitRange while SUnitRange derives from StaticVector?

andyferris · 2020-01-16T07:02:55Z

@c42f, thoughts on whether we are close to ready for doing this?

Also, is there a reason why SOneTo is a subtype of AbstractUnitRange while SUnitRange derives from StaticVector?

Yes - you can only have one supertype. Really we need multiple inheretence or traits or interfaces or something to do this better. Given we have to choose one... trust me - it's far easier to replicate AbstractUnitRange functionality, rather than StaticVector. Actually, I'm guessing it's just because this is a very old PR at this point... :)

vlc1 · 2020-01-16T10:05:59Z

Thanks for your feedback. It sounds that doing things right requires refactoring quite a few things, I'm not sure I have the skills to do this.

However a quick way to get what I need would be to start from the static-indices branch while making SUnitRange a subtype of StaticVector (for the reasons you mentioned).

Would that make sense?

c42f · 2020-01-17T06:08:50Z

@c42f, thoughts on whether we are close to ready for doing this?

Last I looked, I just had the general feeling that these changes are both highly desirable and have the potential to be somewhat breaking. Other than that, I feel like I don't have any special insight on how hard it would be to do this, and I'd only gain some by trying to do the actual work ;-)

hyrodium · 2022-04-21T13:51:06Z

I'll close this PR because:

This seems stale.
Currently, axes(::StaticArray) returns tuple of SOneTo. (indices was deprecated in Julia v0.7.0, and removed in Julia v1.0.0.)
- Note that on Julia v0.6.0, indices(::SMatrix) returns tuple of Base.OneTo.

julia> versioninfo()
Julia Version 0.6.0
Commit 9036443 (2017-06-19 13:05 UTC)
Platform Info:
  OS: Linux (x86_64-pc-linux-gnu)
  CPU: AMD Ryzen 7 2700X Eight-Core Processor
  WORD_SIZE: 64
  BLAS: libopenblas (USE64BITINT DYNAMIC_ARCH NO_AFFINITY Barcelona)
  LAPACK: libopenblas64_
  LIBM: libopenlibm
  LLVM: libLLVM-3.9.1 (ORCJIT, generic)

julia> m = @SMatrix rand(3,3)
3×3 StaticArrays.SArray{Tuple{3,3},Float64,2,9}:
 0.277928  0.778035  0.645978
 0.315307  0.682505  0.176565
 0.984043  0.505622  0.843649

julia> indices(m)
(Base.OneTo(3), Base.OneTo(3))

julia> versioninfo()
Julia Version 1.7.2
Commit bf53498635 (2022-02-06 15:21 UTC)
Platform Info:
  OS: Linux (x86_64-pc-linux-gnu)
  CPU: AMD Ryzen 7 2700X Eight-Core Processor
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-12.0.1 (ORCJIT, znver1)

julia> m = @SMatrix rand(3,3)
3×3 SMatrix{3, 3, Float64, 9} with indices SOneTo(3)×SOneTo(3):
 0.994084   0.435498   0.579111
 0.120314   0.935061   0.435616
 0.0265949  0.0811534  0.772815

julia> axes(m)
(SOneTo(3), SOneTo(3))

WIP Return SUnitRange from indices()

4de8e01

Needs testing

More WIP. Still broken...

a6e65af

timholy reviewed Apr 27, 2017

View reviewed changes

c42f added the design speculative design related issue label Aug 1, 2019

This was referenced Aug 1, 2019

Release 1.0.0 #532

Closed

WIP: static ranges for smarter macro comprehensions #30

Closed

c42f added this to the 1.0 milestone Sep 28, 2019

c42f added the feature features and feature requests label Oct 1, 2019

andyferris added the up for grabs Implement me, I'm yours! label Jan 17, 2020

c42f removed this from the 1.0 milestone Nov 23, 2020

hyrodium mentioned this pull request Apr 21, 2022

StaticArrays.SUnitRange <: AbstractUnitRange should be true? #978

Open

hyrodium closed this Apr 21, 2022

andreasnoack deleted the static-indices branch June 24, 2025 06:06

WIP Return SUnitRange from indices(::StaticArray) #150

WIP Return SUnitRange from indices(::StaticArray) #150

Uh oh!

Conversation

andyferris commented Apr 27, 2017

Uh oh!

c42f commented Apr 27, 2017

Uh oh!

andyferris commented Apr 27, 2017 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

andyferris commented Apr 27, 2017

Uh oh!

andyferris commented Apr 27, 2017 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

andyferris commented Apr 27, 2017

Uh oh!

andyferris commented Apr 27, 2017

Uh oh!

timholy left a comment

Choose a reason for hiding this comment

Uh oh!

timholy Apr 27, 2017

Choose a reason for hiding this comment

Uh oh!

timholy Apr 27, 2017

Choose a reason for hiding this comment

Uh oh!

timholy Apr 27, 2017

Choose a reason for hiding this comment

Uh oh!

timholy Apr 27, 2017

Choose a reason for hiding this comment

Uh oh!

timholy commented Apr 27, 2017

Uh oh!

andyferris commented Apr 27, 2017

Uh oh!

timholy commented Apr 27, 2017

Uh oh!

vlc1 commented Jan 15, 2020

Uh oh!

andyferris commented Jan 16, 2020 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

vlc1 commented Jan 16, 2020

Uh oh!

c42f commented Jan 17, 2020

Uh oh!

hyrodium commented Apr 21, 2022

Uh oh!

Uh oh!

WIP Return `SUnitRange` from `indices(::StaticArray)` #150

WIP Return `SUnitRange` from `indices(::StaticArray)` #150

andyferris commented Apr 27, 2017 •

edited

Loading

andyferris commented Apr 27, 2017 •

edited

Loading

andyferris commented Jan 16, 2020 •

edited

Loading