Sporadic test failure comparing extrema and minimum, maximum

[Zentrik]

The test at

julia/test/reducedim.jl

Line 614 in aad7245

@test all(x -> isequal(x[1], x[2:3]), zip(vext,vmin,vmax))

which is called by

julia/test/reducedim.jl

Line 621 in aad7245

@testset "NaN/missing test for extrema with dims #43599" begin

fails for me depending on what random matrices get generated.

E.g.

julia> a = Union{Missing, Int}[missing -3.0 3.0; -1.0 missing 0.0; missing 2.0 -3.0]
3×3 Matrix{Union{Missing, Int64}}:
   missing  -3          3
 -1           missing   0
   missing   2         -3

julia> dims = (1, 2)
(1, 2)

julia> vext = extrema(a; dims)
1×1 Matrix{Tuple{Union{Missing, Int64}, Union{Missing, Int64}}}:
 (missing, missing)

julia> vmin, vmax = minimum(a; dims), maximum(a; dims)
(Union{Missing, Int64}[-3;;], Union{Missing, Int64}[3;;])

julia> all(x -> isequal(x[1], x[2:3]), zip(vext,vmin,vmax))
false

julia> versioninfo()
Julia Version 1.12.0-DEV.379
Commit aad7245853 (2024-04-20 23:40 UTC)
Build Info:
  Official https://julialang.org/ release
Platform Info:
  OS: Windows (x86_64-w64-mingw32)
  CPU: 16 × AMD Ryzen 7 5700U with Radeon Graphics
  WORD_SIZE: 64
  LLVM: libLLVM-16.0.6 (ORCJIT, znver2)
Threads: 1 default, 0 interactive, 1 GC (on 16 virtual cores)

I don't see this on 1.10.2 but I do on 1.11.0-alpha1, 1.11.0-beta1 and master.

[Seelengrab]

Seems to be due to a change to minimum/maximum, which seem to ignore missing.

v1.10.1:

julia> ex = Union{Missing, Int64}[missing -2; missing -2];

julia> dims = (1,2)
(1, 2)

julia> extrema(ex; dims)
1×1 Matrix{Tuple{Union{Missing, Int64}, Union{Missing, Int64}}}:
 (missing, missing)

julia> minimum(ex; dims)
1×1 Matrix{Union{Missing, Int64}}:
 missing

julia> maximum(ex; dims)
1×1 Matrix{Union{Missing, Int64}}:
 missing

julia> versioninfo()
Julia Version 1.10.1
Commit 7790d6f0641 (2024-02-13 20:41 UTC)
Build Info:
  Official https://julialang.org/ release
Platform Info:
  OS: Linux (x86_64-linux-gnu)
  CPU: 24 × AMD Ryzen 9 7900X 12-Core Processor
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-15.0.7 (ORCJIT, znver3)
Threads: 1 default, 0 interactive, 1 GC (on 24 virtual cores)
Environment:
  JULIA_PKG_USE_CLI_GIT = true

v1.12:

julia> ex = Union{Missing, Int64}[missing -2; missing -2];

julia> dims = (1,2)
(1, 2)

julia> extrema(ex; dims)
1×1 Matrix{Tuple{Union{Missing, Int64}, Union{Missing, Int64}}}:
 (missing, missing)

julia> minimum(ex; dims)
1×1 Matrix{Union{Missing, Int64}}:
 -2

julia> maximum(ex; dims)
1×1 Matrix{Union{Missing, Int64}}:
 -2

julia> versioninfo()
Julia Version 1.12.0-DEV.306
Commit 0221ebc9b1 (2024-04-06 08:13 UTC)
Platform Info:
  OS: Linux (x86_64-pc-linux-gnu)
  CPU: 24 × AMD Ryzen 9 7900X 12-Core Processor
  WORD_SIZE: 64
  LLVM: libLLVM-16.0.6 (ORCJIT, znver4)
Threads: 23 default, 1 interactive, 23 GC (on 24 virtual cores)
Environment:
  JULIA_PKG_USE_CLI_GIT = true

[KristofferC]

I bisected this to 8db1294

~/julia$ manyjulias 8db129472ea88a9c5c4a6 test.jl 
# [ Info: Translated requested revision 8db129472ea88a9c5c4a6 to commit 8db129472ea88a9c5c4a6e94dff7b330723aed28
minimum(ex; dims) = Union{Missing, Int64}[-3;;]

~/julia$ manyjulias 8db129472ea88a9c5c4a6~1 test.jl 
# [ Info: Translated requested revision 8db129472ea88a9c5c4a6~1 to commit 3d8b10770a7a7317dde13a
minimum(ex; dims) = Union{Missing, Int64}[missing;;]

cc @vtjnash

[KristofferC]

I think this is a codegen / inference bug because JuliaInterpreter gets it right:

julia> using JuliaInterpreter

julia> @interpret minimum(ex; dims)
1×1 Matrix{Union{Missing, Int64}}:
 missing

julia> minimum(ex; dims)
1×1 Matrix{Union{Missing, Int64}}:
 -2

[vtjnash]

That seems possible. 8db1294 rewrote some of the memory load/store code to try to make it share more code with the other paths, but the way it stores the type selector bit and computes alignments is much different, so I might have missed a place that needed to be updated