Hello,
I hope you missed my mails about Solaris 😉
We are currently testing whether we can compile on Solaris using GCC 4.9.2. We
were successful in doing so for Solaris x86 but we have hit a problem in Sparc.
It seems related to certain assumptions made around how threads propagates
exceptions:
The failing test is "examples/cpp/example_test.py: test_ssl_bad_name"
In examples/cpp/ssl.cpp [1], it spawns another thread and wait for it to throw
an exception.
However, if you check the thread documentation[2], it states the propagation of
exceptions is not guaranteed[3].
So I was wondering if the assumption that Proton makes is not a big dangerous
and should be changed?
Simpler Example
---------------------
#include <stdexcept>
#include <iostream>
#include <thread>
class example_cert_error : public std::runtime_error
{
public:
explicit example_cert_error(const std::string& s) :
std::runtime_error(s) {}
};
int main(int argc, char** argv) {
try {
std::thread th([](){throw example_cert_error("my error"); });
} catch (const example_cert_error& ce) {
std::cout << "Caught: " << ce.what() << std::endl;
}
return 0;
}
Solaris X86
---------------
$ g++49 -std=c++11 throw.cpp -o throw
$ ./throw
Caught: my error
Solaris Sparc
-----------------
$ g++49 -std=c++11 throw.cpp -o throw
$ ./throw
terminate called without an active exception
Abort (core dumped)
[1]: https://github.com/apache/qpid-proton/blob/0.16.x/examples/cpp/ssl.cpp#L178
[2]: http://en.cppreference.com/w/cpp/thread/thread
[3]: The top-level function may communicate its return value or an exception to
the caller
