Segmentation fault with besselk and "large" inputs on Julia version 1.11.5

[Alex-Adam25]

Hello,
I encountered a segmentation fault when using the modified Bessel function of the second kind, besselk(nu,x) , with nu = 2, x ≥ 19.7, and 128-bit precision (ArbReal{128}) in Julia 1.11.5. Here is an example

$ julia +1.11.5 -q
julia> using ArbNumerics

julia> ArbNumerics.besselk(ArbReal{128}("2"), ArbReal{128}("19.7"))

Segmentation fault

However, when using Julia 1.11.4, the computation succeeds without issues:

$ julia +1.11.4 -q
julia> using ArbNumerics

julia> ArbNumerics.besselk(ArbReal{128}("2"), ArbReal{128}("19.7"))
8.6204978292581032941664283046893573e-10

Other combinations of (nu, x) that produce small output values also trigger the segmentation fault.

I have traced the crash to the ccall in .julia/packages/ArbNumerics/gdpAn/src/float/bessel.jl (line 20) but I don't know enough to proceed debugging it.
Here is the ccall causing the segfault directly:

julia +1.11.5 -q
julia> using ArbNumerics

julia> ccall(ArbNumerics.@libarb(acb_hypgeom_bessel_k),Cvoid,(Ref{ArbComplex}, Ref{ArbComplex}, Ref{ArbComplex}, Clong),
       ArbComplex{128}(),ArbComplex{128}(2),ArbComplex{128}(30),128)

[555570] signal 11 (1): Segmentation fault
in expression starting at REPL[2]:1
Allocations: 1 (Pool: 1; Big: 0); GC: 0
Segmentation fault

[JeffreySarnoff]

Works with Julia 1.12-beta4
I really do not know how to troubleshoot issues tied to a specific julia release (maybe other than the one I am using (albeit pre-release)). More to the point, I do not have expertise with these sort of internals for julia. I sent your case to Slack#internals. I will let you know if they have some insight.

[JeffreySarnoff]

Slack#internals requests you run
julia> versioninfo()
and post the output

[JeffreySarnoff]

Here is the slack thread -- add to it
https://app.slack.com/client/T68168MUP/C688QKS7Q

[JeffreySarnoff]

from slack
This might be the same issue we had with foreign threads and julia allocation functions. It should be fixed in 1.11.6: JuliaLang/julia#58202
disabling the flint threads before the call seems to help
julia> @CCall ArbNumerics.FLINT_jll.libflint_path.flint_set_num_threads(1::Cint)::Cvoid

julia> ArbNumerics.besselk(ArbReal{128}("2"), ArbReal{128}("19.7"))
8.6204978292581032941664283046893573e-10
You could check if it is fixed with the build from the 1.11 backports PR.

try that?

It should be fixed in 1.11.6

[Alex-Adam25]

This is my versioninfo:

julia> versioninfo()
Julia Version 1.11.5
Commit 760b2e5b73 (2025-04-14 06:53 UTC)
Build Info:
Official https://julialang.org/ release
Platform Info:
OS: Windows (x86_64-w64-mingw32)
CPU: 8 × AMD Ryzen 5 2500U with Radeon Vega Mobile Gfx
WORD_SIZE: 64
LLVM: libLLVM-16.0.6 (ORCJIT, znver1)
Threads: 1 default, 0 interactive, 1 GC (on 8 virtual cores)

But I can confirm that on the backports-release-1.11 the bug is gone.
So I think this is fine then.

[JeffreySarnoff]

good