Using -fno-unwind-tables in conjunction with -fno-exceptions

Question

What is the benefit of using -fno-unwind-tables in addtion to -fno-exceptions – especially on a (freestanding) C++ Embedded Systems?

According to Practical Guide to Bare Metal C++ — § Exceptions there should be both used:

It is possible to forbid usage of throw statements by providing certain options to the compiler. For GNU compiler (gcc) please use -fno-exceptions in conjunction with -fno-unwind-tables options.

However there's no explanation what -fno-unwind-tables acutally does.

Answer 1

Per the GCC documentation, it suppresses the generation of static unwind tables (as opposed to complete exception-handling code), the main benefit of which is likely to be smaller object files due to not containing that data. Note that such tables might be used for things like stack-walking code for backtrace generation even in the absence of full-blown exception-handling (e.g. in C code).

I don't know for sure, but I'd imagine that in the case where, by default, full exception-handling code is already disabled but static unwind tables are enabled, -fno-exceptions probably doesn't affect the latter, hence using both options covers all bases. In other words, the compiler may want to generate one or the other, but when size and efficiency matter you definitely want it to do neither.

---

Having hacked about with the makefile of some hapless C++ project (already using -fno-exceptions) to cross-compile it 3 ways, it's apparent which way my particular toolchain defaults:

default                        : 576KB
CPPFLAGS += -fno-unwind-tables : 576KB
CPPFLAGS += -funwind-tables    : 580KB

The tables themselves reside in an additional .ARM.extab ELF section in the latter - of course, for an embedded project you may well have a finely-tuned linker script which would end up discarding that anyway, but hey, still better not to waste time and effort generating it in the first place.