Hacktoberfest 2026：メンテナが10月に向けて印を付けた、オープンで初心者向けの issue。 Hacktoberfest の issue を見る

In-place broadcasting into a `VectorOfArray` mishandles arrays with fewer dimensions

オープン
  • 何か問題がありますか？
#680 コメント 0 件 リアクション 0 件 担当者 0 名 GitHub で見る

メンテナーはふだん 3 日以内に返信

まだ誰も着手していません。

評価

難易度
4/5
見積もり時間
3〜5日
初心者へのやさしさ
45/100
issue の種類
バグ
明瞭さ
明確に書かれている
活発さ
活発
技術スタック
julia
領域
data

調査の方向性

Start with unpack_voa in src/vector_of_array.jl, which slices every array argument along its last dimension, and with the BroadcastStyle methods for VectorOfArrayStyle that route v .= ... through it. The FastBroadcast extension's fast_materialize! for AbstractVectorOfSArray depends on the same slicing. Done means each case in the issue's table (vector, 1×2, fill(2.0), and 3×2 arguments, with both Vector and SVector inner arrays) matches Array(v) broadcasting, with tests added.

索引モデルが issue の本文から書いたものです。

説明

Since v4, a VectorOfArray is a proper AbstractArray (VectorOfArray([[1, 2, 3], [4, 5, 6]]) is a 3×2 array whose columns are the inner arrays), so broadcasting should match broadcasting with Array(v). In-place broadcasting into a VectorOfArray only does so if every array argument has the same number of dimensions as the VectorOfArray. Otherwise it errors or silently returns wrong results:

using RecursiveArrayTools

v = VectorOfArray([[1.0, 2.0, 3.0], [4.0, 5.0, 6.0]])
v .= v .+ [10.0, 20.0, 30.0]
Array(v)
# 3×2 Matrix{Float64}:
#  11.0  24.0
#  12.0  25.0
#  13.0  26.0
#
# but Array(v) .+ [10.0, 20.0, 30.0] is
#  11.0  14.0
#  22.0  25.0
#  33.0  36.0

For the same v (and the same with SVectors as inner arrays):

x v .= x .+ 1 v .= v .+ x
[10.0, 20.0, 30.0] error¹ wrong result (see above)
[10.0 20.0] (1×2) error¹ ✓
fill(2.0) ArgumentError: tuple length should be ≥ 0, got -1 same error
3×2 Matrix ✓ ✓

¹ DimensionMismatch: destination axes (Base.OneTo(3),) are not compatible with source axes ... with Vectors as inner arrays; MethodError: Cannot convert an object of type ... to an object of type Float64 with SVectors.

Out-of-place broadcasting such as v .+ x gives the correct results for vectors and matrices x (the broadcast style becomes DefaultArrayStyle{2}, so Base handles it). It fails only for fill(2.0), because BroadcastStyle(::VectorOfArrayStyle, ::DefaultArrayStyle{0}) returns the VectorOfArrayStyle, which treats zero-dimensional arrays like scalars.

The cause is unpack_voa (src/vector_of_array.jl):

function unpack_voa(x::AbstractArray{T, N}, i) where {T, N}
    return @view x[ntuple(x -> Colon(), N - 1)..., i]
end

It slices every array argument along its own last dimension, as if that dimension indexed the inner arrays. That holds only if ndims(x) equals the number of dimensions of the VectorOfArray. With Base's broadcasting semantics, an argument with fewer dimensions should be broadcast across the inner arrays as a whole (zero-dimensional arrays like scalars), and an argument whose last dimension has size 1 should not be indexed with i.

The FastBroadcast extension relies on the same assumption: fast_materialize!(::Serial, dst::AbstractVectorOfSArray, bc) builds each inner SArray from unpack_voa(bc, i) and needs exactly length(dst.u[i]) elements. For example, with FastBroadcast 1.4.0:

using RecursiveArrayTools, StaticArrays, FastBroadcast

v = VectorOfArray([SVector(1.0, 2.0, 3.0), SVector(4.0, 5.0, 6.0)])
x = [10.0, 20.0, 30.0]
@.. v = x + 1  # ERROR: Generator produced too few elements: Expected exactly 3 elements, but got 1

RecursiveArrayTools 4.5.3, Julia 1.13.1.

🤖 Generated with Claude Code

主要言語
Julia
スター
234
フォーク
77
平均マージ
1日 16時間
マージ済み PR（30日）
11

環境構築

はじめの一歩

  1. issue を最後まで読み、次にプロジェクトのコントリビューションガイドを読みます。
  2. 着手することを issue にコメントします — 二人が同じ作業をするのを防げます。
  3. リポジトリをフォークし、ブランチを切って変更します。
  4. issue 番号を参照したプルリクエストを送ります。

SciML/RecursiveArrayTools.jl のほかの issue

SciML/RecursiveArrayTools.jl の issue をすべて見る

似ている issue

Julia の issue をもっと見る

新しい issue をメールで受け取る

初心者向けの GitHub issue を短くまとめたダイジェスト。