I see, didn't know that, I'll add that to the commit to test that properly :)

Unrelated tests (intptrcast.Casting ints,variable.32-bit tracepoint arg) still failing, I put them back into the disable list.

Unrelated tests (intptrcast.Casting ints,variable.32-bit tracepoint arg) still failing, I put them back into the disable list.

Could these be caused by an issue similar to this one?

It's highly probable, opt reports an invalid cast for the first one and an invalid store for the second one.

It's highly probable, opt reports an invalid cast for the first one and an invalid store for the second one.

Any chance you'd be able to fix it as a part of this PR?

Lets get this merged asap.

For future work it would be nice to:

have a codegen test that covers this, that way the llvm-as check will find issues like these. ALthough if we managed to just validate everything the codegen test becomes useless anyway.

I wonder if the code becomes better if we convert the current monster into a AsyncEvent::Join.

int value_size = 8 + 8 + bpftrace.join_argnum_ * bpftrace.join_argsize_;
auto map = std::make_unique<T>(
    "join", BPF_MAP_TYPE_PERCPU_ARRAY, 4, value_size, 1, 0);

If instead we had:

struct Watchpoint
{
  uint64_t action_id;
  uint64_t join_id;
  uint8_t data;

  std::vector<llvm::Type*> asLLVMType(ast::IRBuilderBPF& b);
} __attribute__((packed));

we could just use geps in a logical way to access elements instead of all the inline pointer casts.

But thats all for future work :)