Zend: Normalize empty string to NULL in OnUpdateStrNotEmpty

[arshidkv12]

#23082

[nicolas-grekas]

This fixes the reported symptom, but OnUpdateStrNotEmpty is shared: output_handler, unserialize_callback_func, arg_separator.output, arg_separator.input, doc_root, sys_temp_dir, plus session.name and zend_test.not_empty_str_test. Making it accept empty and store NULL applies to all of them.

arg_separator.output is PHP_INI_ALL and its consumer has no NULL check (ext/standard/http.c#L118). With this patch applied:

$ php -r 'var_dump(ini_set("arg_separator.output", "")); var_dump(http_build_query(["a"=>1,"b"=>2]));'
string(1) "&"
Segmentation fault (core dumped)

On master that ini_set() returns false and nothing breaks.

It should also fail ext/zend_test/tests/zend_ini_str_validator_basic.phpt, which asserts ini_set("zend_test.not_empty_str_test", "") returns bool(false). That test exists to pin the difference between the two handlers.

OnUpdateStr and OnUpdateStrNotEmpty are a deliberate pair. c82acef just picked the wrong one for unserialize_callback_func, so the fix belongs on the entry:

-	STD_PHP_INI_ENTRY("unserialize_callback_func",	NULL,	PHP_INI_ALL,	OnUpdateStrNotEmpty,	unserialize_callback_func,	php_core_globals,	core_globals)
+	STD_PHP_INI_ENTRY("unserialize_callback_func",	NULL,	PHP_INI_ALL,	OnUpdateStr,	unserialize_callback_func,	php_core_globals,	core_globals)

 		/* Check for unserialize callback */
-		if (PG(unserialize_callback_func) == NULL) {
+		if (PG(unserialize_callback_func) == NULL || ZSTR_LEN(PG(unserialize_callback_func)) == 0) {
 			incomplete_class = 1;
 			ce = PHP_IC_ENTRY;
 			break;
 		}

The second hunk restores the empty check c82acef dropped, since OnUpdateStr stores an empty zend_string rather than NULL.

I built and checked that: the GH-23082 reproducer matches 8.5 output, arg_separator.output still rejects empty, the 13 tests in ext/standard/tests/serialize/unserialize_callback_func/ pass, and the Symfony suites that surfaced this are green again.

Your gh23082.phpt applies unchanged to that approach and is worth keeping.

[arshidkv12]

Thanks! You're right. Changing OnUpdateStrNotEmpty affects unrelated INI directives. Switching only unserialize_callback_func to OnUpdateStr and treating both NULL and "" as "no callback" fixes the regression without changing other directives.

[Girgias]

This is not the correct fix.

[arshidkv12]

Ok. Thank you