assert! terminates rusti REPL

[danluu]

Perhaps this is expected behavior in order to maintain consistency, but, as a newcomer to rust, this surprised me, and it seems mildly inconvenient.

rusti> assert!(false); rust: task failed at 'assertion failed: false', <anon>:34 terminate called throwing an exception Abort trap: 6

By comparison, other languages I'm familiar with throw an error and then return you to the REPL:
Scala:
scala> assert (false) java.lang.AssertionError: assertion failed ... scala>

Ocaml:
utop $ assert(0 == 1);; Exception: (Assert_failure //toplevel// 1 0) utop $

Python:
p> assert(0) Traceback (most recent call last): File "<stdin>", line 1, in <module> AssertionError p>

[alexcrichton]

I investigated this for a bit, and I came up with this test case causing the same behavior:

fn main() { assert!(false); }

And then running rustc -Z jit ./foo.rs.

It appears that when failure is generated the c++ exception isn't caught in the normal spot. I know very little about C++ exceptions, but this definitely has something to do with LLVM and JIT code.

I googled around and found a very similar problem in a few different spots. They all pointed to the same answer of enabling exception handling during JIT code generation, but rust definitely does that, so I think that something else is going wrong here. I guess this could be an LLVM bug, but I doubt that. I figured I'd post my findings here anyway.

[alexcrichton]

In other news, using the LLVM branch brson/llvm/wip rebased on top of chapuni/llvm/master (yielding my branch) fixes this issue.

This makes me think that it's purely awaiting an LLVM upgrade, or in other words this is blocked on #7115

[emberian]

@alexcrichton this isn't fixed, but #7115 seems to have landed.

[emberian]

Actually I take that back my rustc might not be new enough..

[bluss]

With the llvm upgrade, I still see an abort on failure. I don't think assert is different from plain fail!

$ rusti
WARNING: The Rust REPL is experimental and may be
unstable. If you encounter problems, please use the
compiler instead. Type :help for help.
rusti> fail!(); 1
rust: task failed at 'explicit failure', <anon>:17
terminate called after throwing an instance of 'rust_task*'
Avbruten (SIGABRT)

[alexcrichton]

I specifically remember this being fixed at some point, although we may not have pulled in enough LLVM fixes to make it a thing. Regardless, yes, this was not fixed by #7115 and may need more LLVM patches, but we may have regressed to the point that it's our problem now as well.

[alexcrichton]

@blake2-ppc, that looks like the same failure as before though. Basically the JIT code is emitted in such a way that you can't throw an exception from one end to the other. The jit code is calling out into C++ which is thrown an exception to terminate the task, but the C++ code on the other side of the JIT code isn't catching it.

As soon as I can get a building rust with LLVM-head I'll see if this is fixed on it.