CI failure on julia version 1.12.0-rc3 with IntervalArithmetic.jl

[OlivierHnt]

Over at IntervalArithmetic.jl, we are witnessing CI failure for the Julia version 1.12.0-rc3 (as well as nightly).

Everything works fine on Julia versions 1.10.x, 1.11.x and 1.12-rc2.

The beginning of the error message is:

LLVM ERROR: out of memory
Allocation failed

[2506] signal 6 (-6): Aborted

I do not know if this is relevant, but the only "odd thing" we do in the library is a use of @eval to allow the user to configure some algorithm behavior.
So something in the vein of

struct A{T} end

foo(::A{:fast}, a) = "use fast algorithm"
foo(::A{:slow}, a) = "use slow algorithm"

foo(a) = foo(default(), a)

default() = A{:fast}()

configure(algo) = @eval default() = A{$(QuoteNode(algo))}()

and then

julia> foo(1)
"use fast algorithm"

julia> configure(:slow)
default (generic function with 1 method)

julia> foo(1)
"use slow algorithm"

[DilumAluthge]

Can you reproduce locally?

[DilumAluthge]

Also, are you only seeing this on 32-bit builds of Julia? Or do you see it with both 32-bit and 64-bit?

[OlivierHnt]

It seems to fail on both x86 and x64, though, on x64 it looks like it hangs indefinitely.

I will try to test this locally, and report back.

[OlivierHnt]

It fails locally as well:

julia> versioninfo()
Julia Version 1.12.0-rc3
Commit 7522b240144 (2025-09-26 07:42 UTC)
Build Info:
  Official https://julialang.org release
Platform Info:
  OS: macOS (arm64-apple-darwin22.4.0)
  CPU: 11 × Apple M3 Pro
  WORD_SIZE: 64
  LLVM: libLLVM-18.1.7 (ORCJIT, apple-m3)
  GC: Built with stock GC
Threads: 1 default, 1 interactive, 1 GC (on 5 virtual cores)

The first error line is

julia(7277,0x1f6940800) malloc: Failed to allocate segment from range group - out of space

[DilumAluthge]

Could you do a Git bisect?

[Kolaru]

The problem is the splatting of a vector of 256 elements here: https://github.com/JuliaIntervals/IntervalArithmetic.jl/blob/544c0005b96e70c026ebc306ebeeb7e7fed63274/test/interval_tests/multidim.jl#L125

vb4 = mince(fill(-1..1, 4), 4)   # 256-element Vector{Vector{Interval{Float64}}}
hull.(vb4...)  # hangs on nightly, 10s of compilation on 1.11.6

We can definitely fix the tests to avoid it, I am not sure to which extent it is supposed to be supported 😬

[adienes]

my understanding is that big splat arguments are "supposed" to be supported, but in practice are fragile. there are several open issues for similar failure cases. see also: #38364 (comment)

[OlivierHnt]

Is the bisect still needed now that the culprit has been identified?
If so, I don't know how a git bisect of Julia is done, is there some reference about this?