Certain large problems cause segfaults

[GlenHenshaw]

Using Julia 1.6.1 on an M1 MacBook Air, macOS 11.3.1. This seems to crash only from a fresh Julia REPL, e.g. if I have already invoked solve() on a smaller problem, then this doesn't crash. But if I haven't then I get the following:

julia> using DataDrivenDiffEq;
julia> X = rand([0, 1], 128, 936);
julia> T = collect(LinRange(0, 4.367058580858928, 936));
julia> problem = DiscreteDataDrivenProblem(X, T);
julia> res = solve(problem, DMDSVD())

signal (11): Segmentation fault: 11
in expression starting at REPL[5]:1
unknown function (ip: 0x0)
Allocations: 683789624 (Pool: 683746676; Big: 42948); GC: 426
zsh: segmentation fault  Julia

I can also sometimes get the following error:

rosetta error: unexpectedly need to EmulateForward on a synchronous exception x86_rip=0x108e6ff20 arm_pc=0x10955cfd0 num_insts=4 inst_index=2 x86 instruction bytes: 0x56415741e5894855 0xfe8949d789415053
zsh: trace trap  julia

but not yet repeatedly with a minimal code.

[AlCap23]

Can you try a script as well ? I'll check if I can reproduce this on Ubuntu.

[GlenHenshaw]

When run as a script, the above code throws the following error:

rosetta error: unexpectedly need to EmulateForward on a synchronous exception x86_rip=0x10aa272e0 arm_pc=0x10b10b73c num_insts=6 inst_index=3 x86 instruction bytes: 0x56415741e5894855 0xec83485354415541
zsh: trace trap  julia error.jl

Not sure you will be able to replicate this on Ubuntu, as rosetta is the translation layer that M1 Macs use to run x86 code.

[AlCap23]

Could you use the operator_only = true kwarg? I run into a problem within building the solution, which might be related to the size of the problem.

[GlenHenshaw]

running:

res = solve(problem, DMDSVD(); operator_only = true)

does indeed clear the error, both from the REPL and from the script.

[AlCap23]

I'll add the kwargs for the solution within the next days. I think this might be due to the function building, but have to dig deeper :)

Yes. Definitely this.

[AlCap23]

Closed via #233