Andi Kleen:
> I would rather make it XFAIL and also open a PR, after all it is a
limitation that could (and should) be fixed.
Huh, I didn't see this thread in time.
For the special case that the exception handling code in the caller is
empty - which is what mustcall3.C and mustcall5.C are testing -
we can handle tailcalls by modifying the sjlj test to test for empty
exception code too.
To avoid unwanted code and constant duplication, I've refactored the
code a bit. See:
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=121041
