I just noticed that my recent change to prevent PG_RE_THROW from dying
if there's noplace to longjmp to has provoked a whole lot of warnings
that were not there before. Apparently this is because gcc understands
that siglongjmp() never returns, but is not equally clueful about
pg_re_throw().
We can fix this for gcc by putting __attribute__((noreturn)) on the
declaration of pg_re_throw(), but what about other compilers?
regards, tom lane