http://gcc.gnu.org/bugzilla/show_bug.cgi?id=60645
--- Comment #5 from Chris Johns <chris at contemporary dot net.au> --- I do not know what the right thing to do is for something like libstdc++ with such a wide number of different users. I suspect a patch would be easy if the path to take was clear. In the RTEMS case raising a runtime error and halting the target is acceptable. The key create error is hard and related to a user's configuration. A failure in a live system should not happen because testing should expose any issues like this. I am not familiar with the failure modes of things like key create on other operating systems so treating it as a hard error may not be acceptable. I would also like to point out managing these errors means RTEMS can look at using the posix thread model for gcc and I see advantages in this happening.
