[PATCH 2/7] msvcrt: Clean up registered C++ object in handler.
Daniel Lehman
dlehman at esri.com
Tue May 30 17:59:14 CDT 2017
On 05/24/17 02:55, Daniel Lehman wrote:
> +static DWORD cxx_catch_cleanup(EXCEPTION_RECORD *rec, EXCEPTION_REGISTRATION_RECORD *frame,
> + CONTEXT *context,
> +EXCEPTION_REGISTRATION_RECORD **pdispatcher) {
> + if (rec->ExceptionFlags & (EH_UNWINDING | EH_EXIT_UNWIND))
> + {
> + thread_data_t *data = msvcrt_get_thread_data();
> + frame_info *cur;
> +
> + if (cxx_is_consolidate(rec))
> Is this condition really needed? Shouldn't we clean the object no matter what's the reason of unwind?
Yeah. That's covered by patch 7/7. The original code only cleaned up if consolidating.
Since this 2/7 patch was somewhat of a refactoring for the later patches, I kept the consolidate-only cleanup logic. I can merge them if you want
>> + if ((ULONG64)cur <= (ULONG64)frame)
> This condition is not working. It's making assumption about order of catch_frame and frame_info variables on stack while they are declared this way:
I see what you mean. If I forcefully reverse them, my tests crash; my version of gcc was always putting them in the same place on the stack, regardless of where they were declared
>> + EXCEPTION_REGISTRATION_RECORD catch_frame;
>> cxx_frame_info frame_info;
> Shouldn't the cxx_catch_cleanup just unregister the object that was registered in call_catch_block?
I didn't find a way to be call __CxxUnregisterObject on that one specifically (I'll try suggestions if you got em)
but the loop in cxx_catch_cleanup effective does this. when unwinding a nested exception, cxx_catch_cleanup will be called when unwinding call_catch_block, so all objects registered in call_catch_block and below will be cleaned up at that point. Unless the catch block itself explicitly calls __CxxRegisterObject, the only object on the list up to that frame will be the one from call_catch_block
> Here's a test case that demonstrate the problem with cur <= frame
> comparison:
>
> try {
> try { int *p = NULL; *p = 0x42; }
> catch (klass x) { throw 1; }
> } catch (int i) { }
>
> try { throw 1; }
> catch(...) {}
This crashes for me even on Windows because the SEGV is uncaught. It 'works' if I set an seh translator that throws int, but I get identical results on Wine with my series applied. Do I need to add something?
Thanks
daniel
More information about the wine-devel
mailing list