Assertion failure during undefined variable error reporting

[maleadt]

As observed on PkgEval during CUDSS testing: https://s3.amazonaws.com/julialang-reports/nanosoldier/pkgeval/by_date/2025-01/14/CUDSS.primary.log

This is a package that depends on CUDA, which isn't available on PkgEval, so the libcuda global variable as used by the Clang.jl-generated ccalls generates an undefined variable exception. However, that now triggers an assertion:

julia: /source/src/gc-stock.c:721: jl_gc_small_alloc_inner: Assertion `__extension__ ({ __auto_type __atomic_load_ptr = (&ptls->gc_state); __typeof__ (*__atomic_load_ptr) __atomic_load_tmp; __atomic_load (__atomic_load_ptr, &__atomic_load_tmp, (memory_order_relaxed)); __atomic_load_tmp; }) == 0' failed.

[193] signal 6 (-6): Aborted
in expression starting at /home/pkgeval/.julia/packages/CUDSS/L1Cat/test/runtests.jl:15
unknown function (ip: 0x78a467d5cebc) at /lib/x86_64-linux-gnu/libc.so.6
gsignal at /lib/x86_64-linux-gnu/libc.so.6 (unknown line)
abort at /lib/x86_64-linux-gnu/libc.so.6 (unknown line)
unknown function (ip: 0x78a467cf8394) at /lib/x86_64-linux-gnu/libc.so.6
__assert_fail at /lib/x86_64-linux-gnu/libc.so.6 (unknown line)
jl_gc_small_alloc_inner at /source/src/gc-stock.c:721
jl_gc_small_alloc_noinline at /source/src/gc-stock.c:783 [inlined]
jl_gc_alloc_ at /source/src/gc-stock.c:797
ijl_new_struct at /source/src/datatype.c:1558
ijl_undefined_var_error at /source/src/rtutils.c:152
macro expansion at /home/pkgeval/.julia/packages/CUDA/L1qZp/lib/utils/call.jl:214 [inlined]
macro expansion at /home/pkgeval/.julia/packages/CUDSS/L1Cat/src/libcudss.jl:229 [inlined]
#cudssGetProperty##0 at /home/pkgeval/.julia/packages/CUDA/L1qZp/lib/utils/call.jl:35 [inlined]

Introduced somewhere in 4250be8...71bfbb3, maybe as part of the binding changes (cc @Keno)?

[vtjnash]

This is a CUDA bug, since @ccall(jl_gc_safe_enter()::Int8) is UB to exist inside a julia program (not a legal function to call ever)

https://github.com/JuliaGPU/CUDA.jl/blob/d07a24572814efef6691005bd33ae7fd5f978f49/lib/utils/call.jl#L214

[maleadt]

@ccall(jl_gc_safe_enter()::Int8) is UB to exist inside a julia program (not a legal function to call ever)

Well, we need it or we deadlock. Could you provide an alternative then, given that we don't have #49933?

cc @vchuravy, x-ref JuliaGPU/CUDA.jl#2262

[vtjnash]

There is no way to do this that isn't UB, other than to finish implementing #49933