Forgetting to call Target::initialize_native(...) causes segfault about inkwell HOT 6 CLOSED

Thanks for reporting this! I think I'm already aware of this issue but I haven't been able to figure out a way to avoid the segfault and error because IIRC LLVM doesn't provide a way to check if any target has yet been initialized and also because LLVM's global state seems quite messy from an outside perspective so keeping track of it manually seemed really tricky if not impossible

from inkwell.

One possible soution might be to generate a ZST from those target initializers which is required to be passed into the EE creation method, which would ensure at least one initializer has been called (so long as the user can't create the ZST themselves)

from inkwell.

Yeah I've noticed this happens inside of

FWIW error string returned by LLVM in that particular case is "Unable to find target for this triple (no targets are registered)"... but comparing them would be a horrible idea and I shouldn't even think about that :)

Another interesting place is implementation of Clone itself:

That one says it calls verify to make sure that actual cloning doesn't segfault, but then if verify itself segfaults, what's the difference? Although removing it doesn't fix the issue either...

from inkwell.

As far as I can tell, calling initialize_native multiple times is no-op, as LLVM should already know whether target has been previously initialised.

As such, I wonder if it would make sense and be simpler to just call it as part of create_jit_execution_engine so that user wouldn't need to remember to do so?

from inkwell.

Hmm maybe. You can only run a JIT with a native target, right?

from inkwell.

As far as I understand, yeah. (Same for interpreter engine I guess.)

from inkwell.