StackOverflowError when taking matrix exp with big matrix

[daviehh]

Reduced to a minimum example:

using LinearAlgebra

t = 2.1
hx = rand(125, 125)
u = exp(-1im * hx * t)

println(u[1, 1])

running this script in julia repl (outside of vscode) is ok, but when run in vscode it errors with just

ERROR: LoadError: StackOverflowError:
in expression starting at /test/sof.jl:5

with no more info...

my julia versioninfo():

julia> versioninfo()
Julia Version 1.7.0-rc2
Commit f23fc0d27a (2021-10-20 12:45 UTC)
Platform Info:
OS: macOS (x86_64-apple-darwin21.1.0)
CPU: Intel(R) Core(TM) i5-8259U CPU @ 2.30GHz
WORD_SIZE: 64
LIBM: libopenlibm
LLVM: libLLVM-12.0.1 (ORCJIT, skylake)
Environment:
JULIA_MATHLINK = /Applications/Mathematica.app/Contents/Frameworks/mathlink.framework
JULIA_MATHKERNEL = /Applications/Mathematica.app/Contents/MacOS/MathKernel

julia-vscode: this issue is present in both the normal (v1.4.3) and insider version (v1.5.2)

Most likely not related to io or variable names, since the calculation above was originally inside a function with no prints, etc.

[daviehh]

A bit more digging: the error info above was observed with julia 1.7 rc2; sometimes (?) it also gives

ERROR: LoadError: ReadOnlyMemoryError()

No errors with julia 1.6.3. Computer has 16 G ram, and task manager shows normal memory usage.

[pfitzseb]

Very weird. Unfortunately I can't repro this on my machine (with 16GB RAM as well). MacOS stackoverflows are notorious for not being particularly informative, but I'm not sure I can be of much help debugging this. Couple of pointers:

• Are you running the same sysimg in both cases?
• Did you load any packages in either case? It's possible something interacts weirdly with the modules loaded into the VS Code REPL.

[daviehh]

Thanks for getting back! No custom sysimg, no other package other than LinearAlgebra; also tried with the connect external REPL by pasting the code provided, with fresh julia with no startup file (also tried with an empty ~.julia folder), still StackOverflowError; notice no error with smaller matrix below:

In addition,

julia> n = 100
100

julia> exp(rand(ComplexF64, n, n));
ERROR: StackOverflowError:

but exp(rand(n, n)) works fine (Float64 matrix is fine, but not complex-valued ones)...

FWIW, also noticing this with ijulia/jupyter notebooks, but not Pluto.jl if it helps troubleshooting

[daviehh]

Closing, since it appears to be a julia issue and not related to vscode: without vscode stuff, the vanilla julia repl just has StackOverflowError with a higher n:

julia> using LinearAlgebra

julia> n = 300
300

julia> m = rand(ComplexF64, n, n);

julia> mex = exp(m);
ERROR: StackOverflowError:

[daviehh]

link: JuliaLang/LinearAlgebra.jl#886