## Saturday, December 22, 2012

I learned something recently. One should be careful with catch(...). And one should be very careful with pthread_cancel.

The catch-all has a limited uses of course, since you usually know the types of exception to expect. However, there is the rare occasion for its use. What I learned is that you must always re-throw the caught exception. If you don't, pthread_cancel and pthread_exit may break.

The reason is that on Linux, those functions work by throwing a special exception. This exception is special and does not derive from std::exception. You must re-throw it. It is possible to catch this exception alone and re-throw. In order to do that, see this post.

The same logic also implies that when using pthread_cancel, you should not use functions which serve as cancellation points in destructors. In other words, it is best not to use pthread_cancel.

In summary:

• When using catch(...), always re-throw.
• Do not use pthread_cancel.

Note: the same does not appear to be true on Solaris.

References: Ulrich Drepper's blogKenneth's blog.