Improve is_instanceof inference

[iluuu1994]

When a class is final, it may be treated as !is_instanceof.

[dstogov]

The modification of the __is_instanceof should be done before the check if it was changed.

diff --git a/Zend/Optimizer/zend_inference.c b/Zend/Optimizer/zend_inference.c
index af6c301fd934..f3988806e739 100644
--- a/Zend/Optimizer/zend_inference.c
+++ b/Zend/Optimizer/zend_inference.c
@@ -2005,8 +2005,13 @@ static uint32_t get_ssa_alias_types(zend_ssa_alias_kind alias) {
 		if (var >= 0) {													\
- 			if (ssa_var_info[var].ce != (_ce) ||                        \
- 			    ssa_var_info[var].is_instanceof != (_is_instanceof)) {  \
-				ssa_var_info[var].ce = (_ce);						    \
-				ssa_var_info[var].is_instanceof = (_is_instanceof);     \
+			zend_class_entry *__ce = (_ce);							\
+			bool __is_instanceof = (_is_instanceof);				\
+			if (__ce && (__ce->ce_flags & ZEND_ACC_FINAL)) {		\
+				__is_instanceof = false;							\
+			}														\
+ 			if (ssa_var_info[var].ce != __ce ||                        \
+ 			    ssa_var_info[var].is_instanceof != __is_instanceof) {  \
+				ssa_var_info[var].ce = __ce;							\
+				ssa_var_info[var].is_instanceof = __is_instanceof;		\
 				if (update_worklist) { 									\
 					add_usages(op_array, ssa, worklist, var);			\
 				}														\

[iluuu1994]

Fair enough. I don't think this should make a difference because the check is based on ce, which would need to change too for the outcome of the check to be different. But your siggestion might be easier to understand.

[dstogov]

Fair enough. I don't think this should make a difference because the check is based on ce, which would need to change too for the outcome of the check to be different. But your siggestion might be easier to understand.

The old code might never reach the fixed point and repeatable call add_usages().

[iluuu1994]

The old code might never reach the fixed point and repeatable call add_usages().

Yeah, you're right.