One thing to note: it's not always working. E.g. it doesn't work if we have a SIGSEGV but it does if we have guarantee(false). It probably has something to do with signal handling.

We also encountered the confusing situation in crash_or_trap where raise(SIGTRAP) didn't abort the process. Maybe it is now swallowed up by MacOS. So now the crash_or_trap macro is the same as crash -- it calls abort() after the BREAKPOINT line.

If you're running a debugger on some system and want to do something crazy like step forward, well, then, you can adjust the macro accordingly.

I assume the reason for this MacOS change is somehow related to how we block signals on non-main threads in thread_pool.cc. Maybe we just ignore SIGTRAP now on MacOS. It didn't trap when running in lldb either.

(I tested a simple executable that calls raise(SIGTRAP) in a separate thread and that did interrupt the whole process. I am not going to investigate further.)

• I have read and agreed to the RethinkDB Contributor License Agreement http://rethinkdb.com/community/cla/