Backport fix for HASH/PACKED array inference through MAY_BE_ARRAY_EMPTY flag (#12591)

* Fixed HASH/PACKED array inference through MAY_BE_ARRAY_EMPTY flag

This fixes GH-12527

* typo

Author: dstogov

Zend/Optimizer/ssa_integrity.c

@@ -170,10 +170,10 @@ void ssa_verify_integrity(zend_op_array *op_array, zend_ssa *ssa, const char *ex
 			}
 		} FOREACH_PHI_USE_END();
 
-		if ((type & MAY_BE_ARRAY_KEY_ANY) && !(type & MAY_BE_ARRAY_OF_ANY)) {
+		if ((type & (MAY_BE_ARRAY_KEY_ANY-MAY_BE_ARRAY_EMPTY)) && !(type & MAY_BE_ARRAY_OF_ANY)) {
 			FAIL("var " VARFMT " has array key type but not value type\n", VAR(i));
 		}
-		if ((type & MAY_BE_ARRAY_OF_ANY) && !(type & MAY_BE_ARRAY_KEY_ANY)) {
+		if ((type & MAY_BE_ARRAY_OF_ANY) && !(type & (MAY_BE_ARRAY_KEY_ANY-MAY_BE_ARRAY_EMPTY))) {
 			FAIL("var " VARFMT " has array value type but not key type\n", VAR(i));
 		}
 		if ((type & MAY_BE_REF) && ssa->var_info[i].ce) {

Zend/Optimizer/zend_dump.c

@@ -239,21 +239,34 @@ static void zend_dump_type_info(uint32_t info, zend_class_entry *ce, int is_inst
 		}
 		if (info & MAY_BE_ARRAY) {
 			if (first) first = 0; else fprintf(stderr, ", ");
-			if (!(info & MAY_BE_ARRAY_KEY_STRING) || (info & MAY_BE_PACKED_GUARD)) {
-				if (MAY_BE_PACKED_ONLY(info)) {
-					if (info & MAY_BE_PACKED_GUARD) {
-						fprintf(stderr, "!");
-					}
+			if (info & MAY_BE_PACKED_GUARD) {
+				fprintf(stderr, "!");
+			}
+			if (MAY_BE_EMPTY_ONLY(info)) {
+				fprintf(stderr, "empty ");
+			} else if (MAY_BE_PACKED_ONLY(info)) {
+				fprintf(stderr, "packed ");
+			} else if (MAY_BE_HASH_ONLY(info)) {
+				fprintf(stderr, "hash ");
+			} else if ((info & MAY_BE_ARRAY_KEY_ANY) != MAY_BE_ARRAY_KEY_ANY && (info & MAY_BE_ARRAY_KEY_ANY) != 0) {
+				bool afirst = 1;
+				fprintf(stderr, "[");
+				if (info & MAY_BE_ARRAY_EMPTY) {
+					if (afirst) afirst = 0; else fprintf(stderr, ", ");
+					fprintf(stderr, "empty ");
+				}
+				if (MAY_BE_PACKED(info)) {
+					if (afirst) afirst = 0; else fprintf(stderr, ", ");
 					fprintf(stderr, "packed ");
-				} else if (MAY_BE_HASH_ONLY(info)) {
-					if (info & MAY_BE_PACKED_GUARD) {
-						fprintf(stderr, "!");
-					}
+				}
+				if (MAY_BE_HASH(info)) {
+					if (afirst) afirst = 0; else fprintf(stderr, ", ");
 					fprintf(stderr, "hash ");
 				}
+				fprintf(stderr, "] ");
 			}
 			fprintf(stderr, "array");
-			if ((info & MAY_BE_ARRAY_KEY_ANY) != 0 &&
+			if ((info & (MAY_BE_ARRAY_KEY_LONG|MAY_BE_ARRAY_KEY_STRING)) != 0 &&
 			    ((info & MAY_BE_ARRAY_KEY_LONG) == 0 ||
 			     (info & MAY_BE_ARRAY_KEY_STRING) == 0)) {
 				bool afirst = 1;

Zend/Optimizer/zend_func_info.c

@@ -61,7 +61,7 @@ static uint32_t zend_range_info(const zend_call_info *call_info, const zend_ssa
 		uint32_t t2 = _ssa_op1_info(op_array, ssa, call_info->arg_info[1].opline,
 			&ssa->ops[call_info->arg_info[1].opline - op_array->opcodes]);
 		uint32_t t3 = 0;
-		uint32_t tmp = MAY_BE_RC1 | MAY_BE_ARRAY;
+		uint32_t tmp = MAY_BE_RC1 | MAY_BE_ARRAY | MAY_BE_ARRAY_EMPTY;
 
 		if (call_info->num_args == 3) {
 			t3 = _ssa_op1_info(op_array, ssa, call_info->arg_info[2].opline,
@@ -87,7 +87,7 @@ static uint32_t zend_range_info(const zend_call_info *call_info, const zend_ssa
 		return tmp;
 	} else {
 		/* May throw */
-		return MAY_BE_RC1 | MAY_BE_ARRAY | MAY_BE_ARRAY_PACKED | MAY_BE_ARRAY_OF_LONG | MAY_BE_ARRAY_OF_DOUBLE | MAY_BE_ARRAY_OF_STRING;
+		return MAY_BE_RC1 | MAY_BE_ARRAY | MAY_BE_ARRAY_EMPTY | MAY_BE_ARRAY_PACKED | MAY_BE_ARRAY_OF_LONG | MAY_BE_ARRAY_OF_DOUBLE | MAY_BE_ARRAY_OF_STRING;
 	}
 }
 
@@ -120,7 +120,12 @@ uint32_t zend_get_internal_func_info(
 	if (info->info_func) {
 		return call_info ? info->info_func(call_info, ssa) : 0;
 	} else {
-		return info->info;
+		uint32_t ret = info->info;
+
+		if (ret & MAY_BE_ARRAY) {
+			ret |= MAY_BE_ARRAY_EMPTY;
+		}
+		return ret;
 	}
 }
 

Zend/Optimizer/zend_inference.c

@@ -2105,16 +2105,22 @@ ZEND_API uint32_t ZEND_FASTCALL zend_array_type_info(const zval *zv)
 		tmp |= MAY_BE_RCN;
 	}
 
-	ZEND_HASH_FOREACH_STR_KEY_VAL(ht, str, val) {
-		if (str) {
-			tmp |= MAY_BE_ARRAY_KEY_STRING;
-		} else {
-			tmp |= MAY_BE_ARRAY_KEY_LONG;
-		}
-		tmp |= 1 << (Z_TYPE_P(val) + MAY_BE_ARRAY_SHIFT);
-	} ZEND_HASH_FOREACH_END();
-	if (HT_IS_PACKED(ht)) {
-		tmp &= ~(MAY_BE_ARRAY_NUMERIC_HASH|MAY_BE_ARRAY_STRING_HASH);
+	if (zend_hash_num_elements(ht) == 0) {
+		tmp |=  MAY_BE_ARRAY_EMPTY;
+	} else if (HT_IS_PACKED(ht)) {
+		tmp |= MAY_BE_ARRAY_PACKED;
+		ZEND_HASH_PACKED_FOREACH_VAL(ht, val) {
+			tmp |= 1 << (Z_TYPE_P(val) + MAY_BE_ARRAY_SHIFT);
+		} ZEND_HASH_FOREACH_END();
+	} else {
+		ZEND_HASH_MAP_FOREACH_STR_KEY_VAL(ht, str, val) {
+			if (str) {
+				tmp |= MAY_BE_ARRAY_STRING_HASH;
+			} else {
+				tmp |= MAY_BE_ARRAY_NUMERIC_HASH;
+			}
+			tmp |= 1 << (Z_TYPE_P(val) + MAY_BE_ARRAY_SHIFT);
+		} ZEND_HASH_FOREACH_END();
 	}
 	return tmp;
 }
@@ -2222,6 +2228,7 @@ static uint32_t assign_dim_array_result_type(
 	if (tmp & MAY_BE_ARRAY_KEY_ANY) {
 		tmp |= (value_type & MAY_BE_ANY) << MAY_BE_ARRAY_SHIFT;
 	}
+	tmp &= ~MAY_BE_ARRAY_EMPTY;
 	return tmp;
 }
 
@@ -3656,6 +3663,9 @@ static zend_always_inline zend_result _zend_update_type_info(
 							break;
 						}
 					}
+					if (opline->opcode != ZEND_FETCH_DIM_FUNC_ARG) {
+						tmp &= ~MAY_BE_ARRAY_EMPTY;
+					}
 				}
 				if (((tmp & MAY_BE_ARRAY) && (tmp & MAY_BE_ARRAY_KEY_ANY))
 				 || opline->opcode == ZEND_FETCH_DIM_FUNC_ARG
@@ -3829,7 +3839,7 @@ static zend_always_inline zend_result _zend_update_type_info(
 			UPDATE_SSA_TYPE(MAY_BE_LONG, ssa_op->result_def);
 			break;
 		case ZEND_FUNC_GET_ARGS:
-			UPDATE_SSA_TYPE(MAY_BE_RC1|MAY_BE_RCN| MAY_BE_ARRAY | MAY_BE_ARRAY_PACKED | MAY_BE_ARRAY_OF_ANY, ssa_op->result_def);
+			UPDATE_SSA_TYPE(MAY_BE_RC1|MAY_BE_RCN|MAY_BE_ARRAY|MAY_BE_ARRAY_EMPTY|MAY_BE_ARRAY_PACKED|MAY_BE_ARRAY_OF_ANY, ssa_op->result_def);
 			break;
 		case ZEND_GET_CLASS:
 		case ZEND_GET_CALLED_CLASS:

Zend/Optimizer/zend_inference.h

@@ -29,7 +29,6 @@
 #define MAY_BE_PACKED_GUARD         (1<<27) /* needs packed array guard */
 #define MAY_BE_CLASS_GUARD          (1<<27) /* needs class guard */
 #define MAY_BE_GUARD                (1<<28) /* needs type guard */
-//#define MAY_BE_IN_REG               (1<<29) /* deprecated and not used */
 
 #define MAY_HAVE_DTOR \
 	(MAY_BE_OBJECT|MAY_BE_RESOURCE \

Zend/zend_type_info.h

@@ -59,15 +59,17 @@
 #define MAY_BE_ARRAY_PACKED         (1<<21)
 #define MAY_BE_ARRAY_NUMERIC_HASH   (1<<22) /* hash with numeric keys */
 #define MAY_BE_ARRAY_STRING_HASH    (1<<23) /* hash with string keys */
+#define MAY_BE_ARRAY_EMPTY          (1<<29)
 
 #define MAY_BE_ARRAY_KEY_LONG       (MAY_BE_ARRAY_PACKED | MAY_BE_ARRAY_NUMERIC_HASH)
 #define MAY_BE_ARRAY_KEY_STRING     MAY_BE_ARRAY_STRING_HASH
-#define MAY_BE_ARRAY_KEY_ANY        (MAY_BE_ARRAY_KEY_LONG | MAY_BE_ARRAY_KEY_STRING)
+#define MAY_BE_ARRAY_KEY_ANY        (MAY_BE_ARRAY_KEY_LONG | MAY_BE_ARRAY_KEY_STRING | MAY_BE_ARRAY_EMPTY)
 
 #define MAY_BE_PACKED(t)            ((t) & MAY_BE_ARRAY_PACKED)
 #define MAY_BE_HASH(t)              ((t) & (MAY_BE_ARRAY_NUMERIC_HASH | MAY_BE_ARRAY_KEY_STRING))
-#define MAY_BE_PACKED_ONLY(t)       (MAY_BE_PACKED(t) && !MAY_BE_HASH(t))
-#define MAY_BE_HASH_ONLY(t)         (MAY_BE_HASH(t) && !MAY_BE_PACKED(t))
+#define MAY_BE_PACKED_ONLY(t)       (((t) & MAY_BE_ARRAY_KEY_ANY) == MAY_BE_ARRAY_PACKED)
+#define MAY_BE_HASH_ONLY(t)         (MAY_BE_HASH(t) && !((t) & (MAY_BE_ARRAY_PACKED|MAY_BE_ARRAY_EMPTY)))
+#define MAY_BE_EMPTY_ONLY(t)        (((t) & MAY_BE_ARRAY_KEY_ANY) == MAY_BE_ARRAY_EMPTY)
 
 #define MAY_BE_CLASS                (1<<24)
 #define MAY_BE_INDIRECT             (1<<25)

ext/opcache/tests/jit/gh12527.phpt

@@ -0,0 +1,28 @@
+--TEST--
+GH-12527: Incorrect hash/packed inference
+--INI--
+opcache.enable=1
+opcache.enable_cli=1
+--FILE--
+<?php
+function foo(array $a1) {
+    $a2 = [];
+    foreach ($a1 as $key => $val) {
+        if (!$val) {
+            $a2["bad"][] = $key;
+        } else {
+            $a2[0][] = $key;
+        }
+    }
+    foreach ($a2 as $key => $val) {
+    	var_dump($key);
+    }
+}
+foo([1, 2, 3]);
+foo([1, 2, 3]);
+foo([0, 0]);
+?>
+--EXPECT--
+int(0)
+int(0)
+string(3) "bad"