Merge pull request #63 from JuliaFolds2/size_for_scatter

size for RoundRobin()

Author: lmiq

.github/workflows/ci.yml

@@ -12,17 +12,13 @@ jobs:
       fail-fast: false
       matrix:
         version:
-          - '1.6' # remove at some point in the future, if needed.
           - 'lts' # currently 1.10
           - '1' # latest stable 1.x release
           - 'pre'
         os:
           - ubuntu-latest
           - macOS-latest
           - windows-latest
-        exclude:
-          - version: '1.6'
-            os: macOS-latest
     steps:
       - uses: actions/checkout@v4
       - uses: julia-actions/setup-julia@v2

.github/workflows/downgrade.yml

@@ -15,7 +15,7 @@ jobs:
     runs-on: ubuntu-latest
     strategy:
       matrix:
-        version: ['1.6']
+        version: ['1.10']
     steps:
       - uses: actions/checkout@v4
       - uses: julia-actions/setup-julia@v2

CHANGELOG.md

@@ -1,9 +1,11 @@
 ChunkSplitters.jl Changelog
 ===========================
 
-Version 3.1.3-DEV
+Version 3.2.0-DEV
 -------------
+- ![FEATURE][badge-feature] Implement `size` option for `split=RoundRobin()`.
 - ![INFO][badge-info] Improve documentation of multithreading by adding `index_chunks` examples.
+- ![INFO][badge-info] Drop support for Julia 1.6 (Requires 1.10.0 now).
 
 Version 3.1.2
 -------------

Project.toml

@@ -10,10 +10,10 @@ Aqua = "0.8.5"
 BenchmarkTools = "1"
 Documenter = "1"
 OffsetArrays = "1"
-Test = "1.6"
+Test = "1.10"
 TestItemRunner = "1"
 TestItems = "1"
-julia = "1.6"
+julia = "1.10"
 
 [extras]
 Aqua = "4c88cf16-eb10-579e-8560-4a9242c79595"

src/internals.jl

@@ -94,7 +94,13 @@ Base.firstindex(::AbstractChunks) = 1
 Base.lastindex(c::AbstractChunks) = length(c)
 
 Base.length(c::AbstractChunks{T,FixedCount,S}) where {T,S} = c.n
-Base.length(c::AbstractChunks{T,FixedSize,S}) where {T,S} = cld(length(c.collection), max(1, c.size))
+Base.length(c::AbstractChunks{T,FixedSize,Consecutive}) where {T} = cld(length(c.collection), max(1, c.size))
+function Base.length(c::AbstractChunks{T,FixedSize,RoundRobin}) where {T}
+    l = length(c.collection)
+    l == 0 && return 0
+    s = min(l, max(1, c.size))
+    return fld(l, s) + rem(l, s)
+end
 
 Base.getindex(c::IndexChunks{T,C,S}, i::Int) where {T,C,S} = getchunkindices(c, i)
 Base.getindex(c::ViewChunks{T,C,S}, i::Int) where {T,C,S} = @view(c.collection[getchunkindices(c, i)])
@@ -158,7 +164,11 @@ function getchunkindices(c::AbstractChunks{T,C,S}, ichunk::Integer) where {T,C,S
             size = c.size
             l = length(c.collection)
             size = min(l, size) # handle size>length(c.collection)
-            n = cld(l, size)
+            if S == Consecutive
+                n = cld(l, size)
+            else # RoundRobin
+                n = fld(l, size) + rem(l, size)
+            end
         end
     else 
         n = 0
@@ -192,7 +202,18 @@ function _getchunkindices(::Type{FixedSize}, ::Type{Consecutive}, collection, ic
 end
 
 function _getchunkindices(::Type{FixedSize}, ::Type{RoundRobin}, collection, ichunk; size, kwargs...)
-    throw(ArgumentError("split=RoundRobin() not yet supported in combination with size keyword argument."))
+    l = length(collection)
+    fi = firstindex(collection)
+    step = fld(l, size)
+    if ichunk <= step
+        first = fi + ichunk - 1
+        last = first + (size - 1) * step
+        return first:step:last
+    else
+        k = ichunk - step
+        pos = fi + step * size + k - 1
+        return pos:step:pos
+    end
 end
 
 end # module

test/runtests.jl

@@ -5,28 +5,14 @@ using TestItems: @testitem, @testsnippet
 
 @testsnippet Testing begin
     function test_index_chunks(; array_length, n, size, split, result)
-        if n === nothing
-            d, r = divrem(array_length, size)
-            nchunks = d + (r != 0)
-        elseif size === nothing
-            nchunks = n
-        else
-            throw(ArgumentError("both n and size === nothing"))
-        end
         c = index_chunks(rand(Int, array_length); n=n, size=size, split=split)
+        nchunks = length(c)
         ranges = [c[i] for i in 1:nchunks]
         all(ranges .== result)
     end
 
     function sum_parallel(x, n, size, split, which)
-        if n === nothing
-            d, r = divrem(length(x), size)
-            nchunks = d + (r != 0)
-        elseif size === nothing
-            nchunks = n
-        else
-            throw(ArgumentError("both n and size === nothing"))
-        end
+        nchunks = length(index_chunks(x; n=n, size=size, split=split))
         s = zeros(eltype(x), nchunks)
         if which == "index_chunks"
             Threads.@threads for (ichunk, range) in enumerate(index_chunks(x; n=n, size=size, split=split))
@@ -128,7 +114,17 @@ end
     @test collect.(index_chunks(x; n=3, split=RoundRobin())) == [[-1, 2, 5], [0, 3], [1, 4]]
 
     # FixedSize
-    @test_throws ArgumentError collect(index_chunks(1:10; size=2, split=RoundRobin())) # not supported (yet?)
+    @test collect(index_chunks([1,2,3,4,5,6,7]; size=2, split=RoundRobin())) == [[1,4], [2,5], [3,6], [7]]
+    @test collect(chunks(['a','b','c','d','e','f','g']; size=2, split=RoundRobin())) == [['a','d'], ['b','e'], ['c','f'], ['g']]
+    @test collect(index_chunks(1:7; size=2, split=RoundRobin())) == [1:3:4, 2:3:5, 3:3:6, 7:3:7]
+    @test collect(index_chunks(1:6; size=2, split=RoundRobin())) == [1:3:4, 2:3:5, 3:3:6]
+    @test collect(index_chunks(1:10; size=3, split=RoundRobin())) == [1:3:7, 2:3:8, 3:3:9, 10:3:10]
+    @test collect(index_chunks(1:9; size=3, split=RoundRobin())) == [1:3:7, 2:3:8, 3:3:9]
+    x = OffsetArray(1:7, -1:5)
+    @test collect(index_chunks(x; size=2, split=RoundRobin())) == [-1:3:2, 0:3:3, 1:3:4, 5:3:5]
+    @test test_sum(; array_length=7, n=nothing, size=2, split=RoundRobin())
+    @test test_sum(; array_length=15, n=nothing, size=4, split=RoundRobin())
+    @test test_sum(; array_length=117, n=nothing, size=10, split=RoundRobin())
 end
 
 @testitem "check input argument errors" begin