We'd like to add support for C++ exceptions (yes...) to our embedded firmware. Up until now we've been linking against libsupc++_nano.a and it has worked great. However, we've found that it does not work for throwing C++ exceptions. The problem appears to be none of the ARM.exidx or ARM.extab linker sections are generated during the build of libsupc++_nano; if you use the arm-none-eabi-objdump tool you can see that none of the sections are there vs. for the full libsupc++.a library they are all over the place.
It's unfortunate that the exception tables aren't being built as part of libsupc++_nano.a because it apparently contains all of the exception-handling/unwinding code and links just fine when you try to use it in a program that throw exceptions. What happens in this case is that __cxa_throw() is invoked by a C++ "throw" statement and then the stack unwinding code tries to look up the EIT entry for a return address in __cxa_throw(). However, since libsupc++_nano was built without exception tables, no such EIT entry exists and std::terminate() is invoked instead.
When we link against the full libsupc++.a, exceptions do work. However, this has the downside of pulling in a huge amount of extra code that we really don't care about. It seems like C++ name demangling gets pulled in and this bloats the binary by nearly 30K. I think this is because the full libsupc++ tries to write exception information to STDOUT whereas libsupc++_nano is silent. When linking against the full libsupc++ we also have to provide things like _read, _write, _lseek, _isatty, etc. - I think all so that std::terminate() can print something like "uncaught exception of type foo::bar::baz" before aborting.
We think it would be really great if libsupc++_nano.a could be built such that exception handling tables are also included in the library. The exception tables themselves wouldn't significantly change the size of the binary, and they can be discarded by the linker for users that do not use exception handling code.
View all questions in GNU Toolchain forum