-Wreturn-type should not emit warnings when non-termination can be derived indirectly

[EdSchouten]

Bugzilla Link	11542
Version	3.0
OS	FreeBSD
CC	@efriedma-quic,@tkremenek

Extended Description

I suspect this bug is similar to bug 9599, but not identical, so I'm filing this anyway.

The following code generates a -Wreturn-type warning:

__attribute__((__noreturn__)) void bar(void);

static void
foo(void)
{

        bar();
}

int
f(void)
{

        foo();
}

foo.c:15:1: warning: control reaches end of non-void function [-Wreturn-type]

If I change the call to foo() to bar(), it obviously work, but in this case the compiler should be to derive to a certain extent that foo() isn't going to return anyway. This information should be available anyway, as for this specific C file, foo() is even inlined into bar() completely.

[EdSchouten]

Stated differently; in theory (disregarding things like in-line assembly, the halting problem, etc) there should be no need at all to add attribute((noreturn)) to any static function. It should be able to derive this information from the call graph.

[EdSchouten]

It must be noted that it also raises this warning for pieces of code like the following:

static void
foo(void)
{

        for (;;);
}

int
f(void)
{

        foo();
}

[philnik777]

This seems to me more like we should enable -Wmissing-noreturn by default to propagate [[noreturn]], or at least with -Wall or -Wextra.

[AaronBallman]

Resolving as a duplicate of #111509