Julia 1.9 RC1 segfaults with this code

[prittjam]

Julia Version 1.9.0-rc1
Commit 3b2e0d8 (2023-03-07 07:51 UTC)
Platform Info:
OS: Linux (x86_64-linux-gnu)
CPU: 12 × Intel(R) Core(TM) i7-8850H CPU @ 2.60GHz
WORD_SIZE: 64
LIBM: libopenlibm
LLVM: libLLVM-14.0.6 (ORCJIT, skylake)
Threads: 8 on 12 virtual cores
Environment:
LD_LIBRARY_PATH =
JULIA_NUM_THREADS = 8

using StaticArrays

abstract type VgtkNum end

struct VgtkInt{T <: Integer} <: VgtkNum
    int::T
end

VgtkInt(::Val{N}) where N <: Integer = VgtkInt(N)
VgtkInt(x::AbstractFloat) = VgtkInt(trunc(Int, x))

struct VgtkFloat{T <: AbstractFloat} <: VgtkNum
   float::T
end

VgtkFloat(x::Integer) = VgtkFloat(float(x))

value(x::VgtkInt) = x.int
value(x::VgtkFloat) = x.float 

Base.:(+)(a::VgtkInt, b::VgtkInt) = VgtkInt(a.int + b.int)
Base.:(+)(a::VgtkFloat, b::VgtkFloat) = VgtkFloat(a.float + b.float)
Base.:(+)(a::VgtkNum, b::VgtkNum) = VgtkFloat(value(a) + value(b))

neltype(x) = neltype(typeof(x))
neltype(::Type{T}) where T <: AbstractArray = neltype(eltype(T))
neltype(::Type{T}) where T = T


const PointRpN{N, T <: AbstractFloat} = SVector{value(VgtkInt(Val(N))+VgtkInt(1)), T} 
const PointRp2{T <: AbstractFloat} = PointRpN{2, T}
const LineRp2{T <:  AbstractFloat} = SVector{3, T}

[KristofferC]

Bisected to 7ad0e3d. cc @JeffBezanson

[jakobnissen]

Reduced example with no deps:

struct A{T} end
const B{N} = A(Val(N))

[gbaraldi]

Just for reference, on 1.8 this errors with

julia> struct A{T} end

julia> const B{N} = A(Val(N))
ERROR: UndefVarError: x not defined
Stacktrace:
 [1] Val{N}()
   @ Base ./essentials.jl:712
 [2] Val(x::TypeVar)
   @ Base ./essentials.jl:714
 [3] top-level scope
   @ REPL[2]:1

[vtjnash]

Calling Val(N) where N may trick Julia into creating an instance of an invalid DataType. But that is not really a new regression.

[gbaraldi]

It did use to error before and now it segfaults, so I would say this is a regression. The segfault is because we call a broken tuple.