[cfe-dev] Crash with C++exception. Bug in codegen, or in the developer ?

[John McCall]

On Jul 13, 2012, at 12:17 PM, Jean-Daniel Dupas wrote:
> While running the following code,  it crashes due to an invalid free operation:
> 
> exc(8657) malloc: *** error for object 0x7f97c14039e0: pointer being freed was not allocated
> *** set a breakpoint in malloc_error_break to debug
> fish: Job 1, './exc' terminated by signal SIGABRT (Abort)
> 
> --- --- --- --- --- 
> #include <string>
> 
> class Error {
> public:
>   explicit inline Error(const std::string &str) noexcept {}
> };
> 
> std::string _cxx_strfmt() noexcept { return std::string(); }
> 
> int main(int argc, char **argv) {
>   try {
>     throw Error(_cxx_strfmt());
>   } catch (...) {}
> }
> ------
> 
> compiled using 
> clang++ -std=c++11  exc.cpp
> 
> I can't see what is wrong with this code, but if I remove the noexcept declaration (either one), it stop crashing and run as expected.
> 
> Is this something I'm doing wrong, or this is a subtle clang bug ? 
> 
> I attach the llvm ir generate with noexcept (exc-crash.s) and the one without (exc.s)
> 
> If this is a bug in clang, tell me, I will fill a report.

It's a clang bug;  somehow we're unconditionally freeing the exception
in this case.

John.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-dev/attachments/20120713/750aea3d/attachment.html>