Assert statements don't properly mark their message arguments as used causing the linker to fail.

[MinekPo1]

Minimal example:

import std::io;

fn String foo() {
    return "hello from foo!";
}

// c3c complains about using assert(false), so we need to trick it
fn bool bar() {
    return false;
}

fn void main() {
    io::printn("Hello, World!");
    assert(bar(), "%s", foo());
}

The linker error message is undefined reference to .foo', we can further confirm this is the case by looking at the object file, where we find foo` referenced but undefined:

000000000 V $ct.char
0000000000000000 V $ct.sa$char
0000000000000000 V $ct.String
0000000000000003 r .file
000000000000000b r .func
0000000000000000 r .L.str
0000000000000000 T <module>_fn.bar
                 U <module>_fn.foo
0000000000000010 T <module>_fn.main
00000000000001a0 T main
0000000000000000 r .panic_msg
                 U std.core.builtin.panicf
                 U std.io.File.flush
                 U std.io.File.write
                 U std.io.File.write_byte
                 U std.io.stdout

Tested with c3c 0.6.2, with llvm 17.0.6 targeting x86_64-pc-linux-gnu (TODO: test with a master build)

Not sure if its the correct pass, but in sema_liveness.c I have found one example of a semantic pass not descending into the message arguments, and "liveness" sounds like it looks for dead code, which feels right.

[lerno]

This was indeed sema_liveness. It should now be patched. Please close this when you have verified that it works as intended.