Memory corruption in 0.9.3

[kbarros]

We have noticed intermittent crashes since upgrading from Spglib 0.9.2 to 0.9.3. Looks like the only real difference between these versions is the switch from enums to SumTypes: singularitti/Spglib.jl@819b7f6.

Perhaps there is memory corruption when reading an spglib_jll response into a sum type?

Frequently the result is a segfault, but here is one stacktrace we got that seems interpretable:

Failed to precompile Sunny [2b4a2ac8-8f8b-43e8-abf4-3cb0c45e8736] to "/home/runner/.julia/compiled/v1.9/Sunny/jl_GxA1Pq".
ERROR: LoadError: ArgumentError: cannot convert NULL to string
Stacktrace:
  [1] unsafe_string
    @ ./strings/string.jl:84 [inlined]
  [2] unsafe_string
    @ ./c.jl:193 [inlined]
  [3] get_error_message(code::Spglib.SpglibReturnCode)
    @ Spglib ~/.julia/packages/Spglib/rGUIW/src/error.jl:37
  [4] check_error
    @ ~/.julia/packages/Spglib/rGUIW/src/error.jl:50 [inlined]
  [5] get_spacegroup_type(hall_number::Int64)
    @ Spglib ~/.julia/packages/Spglib/rGUIW/src/symmetry.jl:291
  [...]

This appeared on our Github actions CI with Julia 1.9.4, Ubuntu, x86. Full trace here: https://github.com/SunnySuite/Sunny.jl/actions/runs/7629267203/job/20782289111?pr=217

I haven't seen the crashes yet on my Mac. On Linux/x86, reproducing might be as simple as ] add Sunny#spglib_crash and then using Sunny.

[MasonProtter]

This isn't a bug with SumTypes, it's just being used in a completely invalid way by that library. It's asserting to julia that it can interpret the return type from a ccall as a sum type, which encloses a union, so yeah that's almost guaranteed to cause memory corruption.

I would raise an issue in that library instead.

[kbarros]

Thank you @MasonProtter. My mistake, I meant to file this issue in the Spglib repo, but your input is valuable, thanks.

[MasonProtter]

Not a problem, happy I could help diagnose.