Fix i386 release build warning

[iluuu1994]

--disable-debug is added temporarily to catch the error in CI.

https://github.com/php/php-src/actions/runs/11734433568/job/32690495546#logs

[cmb69]

php-src/main/getopt.c

Lines 87 to 100 in 3b609ab

	if ((argv[*optind][0] == '-') && (argv[*optind][1] == '-')) {
	const char *pos;
	size_t arg_end = strlen(argv[*optind])-1;
	/* '--' indicates end of args if not followed by a known long option name */
	if (argv[*optind][2] == '\0') {
	(*optind)++;
	return(EOF);
	}
	arg_start = 2;
	/* Check for <arg>=<val> */
	if ((pos = php_memnstr(&argv[*optind][arg_start], "=", 1, argv[*optind]+arg_end)) != NULL) {

If only -- is given, arg_start==2 and arg_end==1.

[iluuu1994]

@cmb69 I think the issue is the hard-coded 1 needle length in php_memnstr(). php_memnstr() passes needle_len-2 to memcmp, which would underflow, but at that point needle_len == 1 was already handled. When inlining php_memnstr(), I would expect GCC to discard the needle_len > 1 branch altogether, but for some reason it does not.

Maybe I'm on the wrong track. I don't have a 32-bit build set up locally.

[cmb69]

Why do we even use php_memnstr() here instead of memchr()? disregard

[iluuu1994]

That's a good question. We don't use php_memnstr() with a hard-coded needle_len=1 anywhere else, which is likely why this issue only occurs here.

[cmb69]

Oops, I'm confused. Shouldn't do too much when bisecting. ;)

Anyhow, relevant commit: 024fc2c (2007!)

[iluuu1994]

Your suggestion seems to work 🙂 Thanks!

[cmb69]

Certainly no ABI break here.

[cmb69]

The change to getopt.c looks good to me (might make sense to apply to lower branches too; at least PHP-8.4).