[LLVMdev] C++ Exception Handling Problem
Nicolas Geoffray
nicolas.geoffray at lip6.fr
Tue Mar 10 08:40:43 PDT 2009
Hi Nyx,
Nyx wrote:
>
> Duncan Sands wrote:
>
>> Hi Nyx,
>>
>>
>>> I'm in the process of creating a JIT and I've run into a problem with
>>> exception handling. The situation I'm in is that my program will JIT
>>> functions, which will call native C++ functions (part of the run-time
>>> support). These native functions can throw exceptions. However, I don't
>>> actually want to handle these exceptions in the JITted functions. There
>>> are
>>> already try/catch blocks in place in the C++ function that calls the
>>> JITted
>>> functions.
>>>
>>> The problem I run into is that when exceptions are thrown, they aren't
>>> caught anywhere. My program crashes with an error message: "terminate
>>> called
>>> after throwing an instance of 'RunError'". Where RunError is the
>>> exception
>>> type. Hence, the exceptions aren't getting caught where they should be.
>>>
My guess here is that there are JITted functions in your stack. If
that's the case, you need to set the ExceptionHandling flag of LLVM on.
This will make the JIT generate dwarf tables for JITted code, and libgcc
will be able to unwind these functions.
Nicolas
>> does this happen in the non-jit context. For example, suppose rather than
>> JITing your bitcode you compile it down to object code, and arrange for it
>> to be called from C++ and call C++. Do exceptions then propagate and get
>> caught correctly?
>>
>> Ciao,
>>
>> Duncan.
>> _______________________________________________
>> LLVM Developers mailing list
>> LLVMdev at cs.uiuc.edu http://llvm.cs.uiuc.edu
>> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
>>
>>
>>
>
> By that I assume you mean compiling into a .o object file? I presume that
> would work, but I can't really do that right now because the code I produce
> contains pointers to dynamically allocated structures. What I'm doing right
> now is getting a function pointer from the execution engine and calling the
> JITted functions directly.
>
More information about the llvm-dev
mailing list