crash when using zig cc to compile

[zenspider]

Zig Version

0.12.0-dev.105+f1992a39a

Steps to Reproduce and Observed Behavior

given:

#include <stdio.h>
#include <stdlib.h>
#include <netdb.h>
#include <sys/socket.h>
#include <arpa/inet.h>

int main(int argc, char **argv)
{
  struct sockaddr_in sap;
  struct hostent* hp = gethostbyname("www.google.com");
  sap.sin_addr = *(struct in_addr *)hp->h_addr;         // BUG IS HERE
  printf("IP Address: %s\n", inet_ntoa(sap.sin_addr));

  exit(0);
}

compiling and running with zig cc crashes:

7985 % zig version; zig cc goog.c && ./a.out
0.12.0-dev.105+f1992a39a
Trace/BPT trap: 5

running with lldb points at the commented line above

Expected Behavior

same as llvm's cc:

7987 % cc --version ; cc goog.c && ./a.out
Apple clang version 15.0.0 (clang-1500.0.40.1)
Target: arm64-apple-darwin23.1.0
Thread model: posix
InstalledDir: /Library/Developer/CommandLineTools/usr/bin
IP Address: 142.251.215.228

[zenspider]

just force updated and confirmed the bug still exists:

7990 % zig version ; zig cc goog.c && ./a.out
0.12.0-dev.1746+19af8aac8
Trace/BPT trap: 5

[tenderlove]

I'm not sure how helpful this is, but I disassembled the program and it seems zig cc is emitting a brk instruction (and I guess a useless jump right before it?):

(lldb) disassemble --start-address `0x10000071c - 32` --count 15
a.out`main:
    0x1000006fc <+292>: cset   w8, eq
    0x100000700 <+296>: tbnz   w8, #0x0, 0x10000071c     ; <+324> at goog.c:11:37
    0x100000704 <+300>: b      0x100000708               ; <+304> at goog.c
    0x100000708 <+304>: ldr    x8, [sp, #0x10]
    0x10000070c <+308>: ands   x8, x8, #0x7
    0x100000710 <+312>: cset   w8, eq
    0x100000714 <+316>: tbnz   w8, #0x0, 0x100000720     ; <+328> at goog.c
    0x100000718 <+320>: b      0x10000071c               ; <+324> at goog.c:11:37
->  0x10000071c <+324>: brk    #0x5516
    0x100000720 <+328>: ldr    x8, [sp, #0x18]
    0x100000724 <+332>: ldr    x8, [x8]
    0x100000728 <+336>: ldr    w8, [x8]
    0x10000072c <+340>: stur   w8, [x29, #-0x1c]
    0x100000730 <+344>: ldur   w8, [x29, #-0x1c]
    0x100000734 <+348>: mov    x0, x8

I don't know the compiler internals, but this definitely seems like a compiler bug 😅

[ehaas]

That's -fsanitize=undefined tripping (zig cc enables it by default; clang does not). Casting a pointer to one with a larger alignment is undefined behavior in C.

Edit: it looks like the implementation itself might be broken? https://developer.apple.com/forums/thread/116121 - in general it's not recommended to use gethostbyname anyway, it's a legacy function that doesn't support ipv6

[zenspider]

Workaround: add -fno-sanitize=undefined:

8108 % zig cc -fno-sanitize=undefined goog.c ; ./a.out
IP Address: 142.251.33.68

ETA:

__attribute__((no_sanitize("undefined")))
int main(int argc, char **argv) { ... }