Implement Scope Functions

fn() {
    /* body */
}

Signed-off-by: Bob Weinand <bobwei9@hotmail.com>

Author: bwoebi

Zend/Optimizer/block_pass.c

@@ -942,6 +942,8 @@ static void zend_optimize_block(zend_basic_block *block, zend_op_array *op_array
 						src->opcode != ZEND_ADD_ARRAY_ELEMENT &&
 						src->opcode != ZEND_ADD_ARRAY_UNPACK &&
 						(src->opcode != ZEND_DECLARE_LAMBDA_FUNCTION ||
+						 src == opline -1) &&
+						(src->opcode != ZEND_DECLARE_SCOPE_FUNC ||
 						 src == opline -1)) {
 						src->result.var = opline->result.var;
 						VAR_SOURCE(opline->op1) = NULL;

Zend/Optimizer/compact_literals.c

@@ -35,12 +35,18 @@
 
 typedef struct _literal_info {
 	uint8_t num_related;
+	bool pinned;
 } literal_info;
 
 #define LITERAL_INFO(n, related) do { \
 		info[n].num_related = (related); \
 	} while (0)
 
+#define LITERAL_INFO_PIN(n) do { \
+		info[n].num_related = 1; \
+		info[n].pinned = true; \
+	} while (0)
+
 static uint32_t add_static_slot(HashTable     *hash,
                                 zend_op_array *op_array,
                                 uint32_t       op1,
@@ -239,6 +245,14 @@ void zend_optimizer_compact_literals(zend_op_array *op_array, zend_optimizer_ctx
 						}
 					}
 					break;
+				case ZEND_ENTER_SCOPE_FUNC: {
+					/* We must preserve the precise ordering of the scope fn literals mapping to CV offsets. */
+					uint32_t num_params = opline->op1.num;
+					for (uint32_t k = 0; k < num_params; k++) {
+						LITERAL_INFO_PIN(k);
+					}
+					break;
+				}
 				default:
 					if (opline->op1_type == IS_CONST) {
 						LITERAL_INFO(opline->op1.constant, 1);
@@ -315,7 +329,14 @@ void zend_optimizer_compact_literals(zend_op_array *op_array, zend_optimizer_ctx
 					map[i] = l_true;
 					break;
 				case IS_LONG:
-					if (info[i].num_related == 1) {
+					if (info[i].pinned) {
+						map[i] = j;
+						if (i != j) {
+							op_array->literals[j] = op_array->literals[i];
+							info[j] = info[i];
+						}
+						j++;
+					} else if (info[i].num_related == 1) {
 						if ((pos = zend_hash_index_find(&hash, Z_LVAL(op_array->literals[i]))) != NULL) {
 							map[i] = Z_LVAL_P(pos);
 						} else {

Zend/Optimizer/compact_vars.c

@@ -18,11 +18,58 @@
 #include "zend_bitset.h"
 #include "zend_observer.h"
 
+/* Recursively include scope fns for shared CVs; apply changes with vars_map, otherwise fill used_vars */
+static void scope_fn_visit_parent_cvs(zend_op_array *scope_op, zend_bitset used_vars, const uint32_t *vars_map)
+{
+#define VISIT_CV(slot) if (slot##_type == IS_CV) { \
+	uint32_t old = VAR_NUM(slot.var); \
+	if (vars_map) slot.var = NUM_VAR(vars_map[old]); \
+	else zend_bitset_incl(used_vars, old); \
+}
+	bool in_body = false;
+	for (uint32_t i = 0; i < scope_op->last; i++) {
+		zend_op *opline = &scope_op->opcodes[i];
+		if (opline->opcode == ZEND_ENTER_SCOPE_FUNC) {
+			in_body = true;
+			/* Mapping is pinned at literals[0..num_params). */
+			uint32_t num_params = opline->op1.num;
+			for (uint32_t j = 0; j < num_params; j++) {
+				zval *zv = &scope_op->literals[j];
+				uint32_t off = (uint32_t)Z_LVAL_P(zv);
+				if (vars_map) {
+					ZVAL_LONG(zv, NUM_VAR(vars_map[VAR_NUM(off)]));
+				} else {
+					zend_bitset_incl(used_vars, VAR_NUM(off));
+				}
+			}
+			continue;
+		}
+		if (!in_body) {
+			continue;
+		}
+		VISIT_CV(opline->op1);
+		VISIT_CV(opline->op2);
+		VISIT_CV(opline->result);
+	}
+	for (uint32_t i = 0; i < scope_op->num_dynamic_func_defs; i++) {
+		zend_op_array *nested = scope_op->dynamic_func_defs[i];
+		if (nested->fn_flags2 & ZEND_ACC2_SCOPE_FUNC) {
+			scope_fn_visit_parent_cvs(nested, used_vars, vars_map);
+		}
+	}
+#undef VISIT_CV
+}
+
 /* This pass removes all CVs and temporaries that are completely unused. It does *not* merge any CVs or TMPs.
  * This pass does not operate on SSA form anymore. */
 void zend_optimizer_compact_vars(zend_op_array *op_array) {
 	int i;
 
+	/* Scope fn CVs are handled by the ancestor. Just skip this here. */
+	if (op_array->fn_flags2 & ZEND_ACC2_SCOPE_FUNC) {
+		return;
+	}
+
 	ALLOCA_FLAG(use_heap1);
 	ALLOCA_FLAG(use_heap2);
 	uint32_t used_vars_len = zend_bitset_len(op_array->last_var + op_array->T);
@@ -52,6 +99,13 @@ void zend_optimizer_compact_vars(zend_op_array *op_array) {
 		}
 	}
 
+	for (i = 0; i < (int)op_array->num_dynamic_func_defs; i++) {
+		zend_op_array *child = op_array->dynamic_func_defs[i];
+		if (child->fn_flags2 & ZEND_ACC2_SCOPE_FUNC) {
+			scope_fn_visit_parent_cvs(child, used_vars, NULL);
+		}
+	}
+
 	num_cvs = 0;
 	for (i = 0; i < op_array->last_var; i++) {
 		if (zend_bitset_in(used_vars, i)) {
@@ -93,6 +147,13 @@ void zend_optimizer_compact_vars(zend_op_array *op_array) {
 		}
 	}
 
+	for (i = 0; i < (int)op_array->num_dynamic_func_defs; i++) {
+		zend_op_array *child = op_array->dynamic_func_defs[i];
+		if (child->fn_flags2 & ZEND_ACC2_SCOPE_FUNC) {
+			scope_fn_visit_parent_cvs(child, NULL, vars_map);
+		}
+	}
+
 	/* Update CV name table */
 	if (num_cvs != op_array->last_var) {
 		if (num_cvs) {

Zend/Optimizer/optimize_temp_vars_5.c

@@ -122,6 +122,11 @@ void zend_optimize_temporary_variables(zend_op_array *op_array, zend_optimizer_c
 							}
 						}
 					}
+					/* DECLARE_SCOPE_FUNC op1 is never consumed, just used for itself.
+					 * It has to simply exist for the whole function, so it gets its own TMP. */
+					if (opline->opcode == ZEND_DECLARE_SCOPE_FUNC) {
+						use_new_var = 1;
+					}
 					if (use_new_var) {
 						i = ++max;
 						zend_bitset_incl(taken_T, i);

Zend/Optimizer/zend_call_graph.c

@@ -25,12 +25,18 @@
 static void zend_op_array_calc(zend_op_array *op_array, void *context)
 {
 	zend_call_graph *call_graph = context;
+	if (op_array->fn_flags2 & ZEND_ACC2_SCOPE_FUNC) {
+		return;
+	}
 	call_graph->op_arrays_count++;
 }
 
 static void zend_op_array_collect(zend_op_array *op_array, void *context)
 {
 	zend_call_graph *call_graph = context;
+	if (op_array->fn_flags2 & ZEND_ACC2_SCOPE_FUNC) {
+		return;
+	}
 	zend_func_info *func_info = call_graph->func_infos + call_graph->op_arrays_count;
 
 	ZEND_SET_FUNC_INFO(op_array, func_info);

Zend/Optimizer/zend_inference.c

@@ -4010,9 +4010,12 @@ static zend_always_inline zend_result _zend_update_type_info(
 			UPDATE_SSA_TYPE(MAY_BE_BOOL, ssa_op->result_def);
 			break;
 		case ZEND_DECLARE_LAMBDA_FUNCTION:
+		case ZEND_DECLARE_SCOPE_FUNC:
 			UPDATE_SSA_TYPE(MAY_BE_OBJECT|MAY_BE_RC1|MAY_BE_RCN, ssa_op->result_def);
 			UPDATE_SSA_OBJ_TYPE(zend_ce_closure, /* is_instanceof */ false, ssa_op->result_def);
 			break;
+		case ZEND_ENTER_SCOPE_FUNC:
+			break;
 		case ZEND_PRE_DEC_STATIC_PROP:
 		case ZEND_PRE_INC_STATIC_PROP:
 		case ZEND_POST_DEC_STATIC_PROP:

Zend/Optimizer/zend_optimizer.c

@@ -1048,13 +1048,19 @@ zend_op *zend_optimizer_get_loop_var_def(const zend_op_array *op_array, zend_op
 	return NULL;
 }
 
+static zend_always_inline bool zend_op_array_is_scope_fn(const zend_op_array *op_array) {
+	return (op_array->fn_flags2 & ZEND_ACC2_SCOPE_FUNC) != 0;
+}
+
 static void zend_optimize(zend_op_array      *op_array,
                           zend_optimizer_ctx *ctx)
 {
 	if (op_array->type == ZEND_EVAL_CODE) {
 		return;
 	}
 
+	bool is_scope_fn = zend_op_array_is_scope_fn(op_array);
+
 	if (ctx->debug_level & ZEND_DUMP_BEFORE_OPTIMIZER) {
 		zend_dump_op_array(op_array, ZEND_DUMP_LIVE_RANGES, "before optimizer", NULL);
 	}
@@ -1106,9 +1112,13 @@ static void zend_optimize(zend_op_array      *op_array,
 
 	/* pass 6:
 	 * - DFA optimization
-	 */
+	 *
+	 * Skipped for scope-fn op_arrays: DCE on a body's CV write would
+	 * eliminate a store the parent later reads, but single-op_array SSA
+	 * cannot see that. Re-enabling needs cross-op-array DCE protection
+	 * (treat every body CV write as live-out). */
 	if ((ZEND_OPTIMIZER_PASS_6 & ctx->optimization_level) &&
-	    !(ZEND_OPTIMIZER_PASS_7 & ctx->optimization_level)) {
+	    !(ZEND_OPTIMIZER_PASS_7 & ctx->optimization_level) && !is_scope_fn) {
 		zend_optimize_dfa(op_array, ctx);
 		if (ctx->debug_level & ZEND_DUMP_AFTER_PASS_6) {
 			zend_dump_op_array(op_array, 0, "after pass 6", NULL);
@@ -1141,7 +1151,8 @@ static void zend_optimize(zend_op_array      *op_array,
 	 */
 	if ((ZEND_OPTIMIZER_PASS_11 & ctx->optimization_level) &&
 	    (!(ZEND_OPTIMIZER_PASS_6 & ctx->optimization_level) ||
-	     !(ZEND_OPTIMIZER_PASS_7 & ctx->optimization_level))) {
+	     !(ZEND_OPTIMIZER_PASS_7 & ctx->optimization_level) ||
+	     is_scope_fn /* normally passes 6/7 would handle it, but scope fns are exempt */)) {
 		zend_optimizer_compact_literals(op_array, ctx);
 		if (ctx->debug_level & ZEND_DUMP_AFTER_PASS_11) {
 			zend_dump_op_array(op_array, 0, "after pass 11", NULL);
@@ -1172,6 +1183,8 @@ static void zend_revert_pass_two(zend_op_array *op_array)
 
 	ZEND_ASSERT((op_array->fn_flags & ZEND_ACC_DONE_PASS_TWO) != 0);
 
+	zend_pass_two_revert_scope_fn_reservations(op_array);
+
 	opline = op_array->opcodes;
 	const zend_op *end = opline + op_array->last;
 	while (opline < end) {
@@ -1304,6 +1317,7 @@ static void zend_redo_pass_two(zend_op_array *op_array)
 	}
 
 	op_array->fn_flags |= ZEND_ACC_DONE_PASS_TWO;
+	zend_pass_two_install_scope_fn_reservations(op_array);
 }
 
 static void zend_redo_pass_two_ex(zend_op_array *op_array, const zend_ssa *ssa)
@@ -1446,23 +1460,35 @@ static void zend_redo_pass_two_ex(zend_op_array *op_array, const zend_ssa *ssa)
 	}
 
 	op_array->fn_flags |= ZEND_ACC_DONE_PASS_TWO;
+	zend_pass_two_install_scope_fn_reservations(op_array);
 }
 
 static void zend_optimize_op_array(zend_op_array      *op_array,
                                    zend_optimizer_ctx *ctx)
 {
+	bool is_scope_fn = zend_op_array_is_scope_fn(op_array);
+
+	/* Make scope fns individually optimizeable first, before running optimizer. */
+	uint32_t saved_scope_T = is_scope_fn ? op_array->T : 0;
+	uint32_t saved_scope_ex_offset = is_scope_fn ? zend_unfixup_scope_func_self(op_array, saved_scope_T) : 0;
+
 	/* Revert pass_two() */
 	zend_revert_pass_two(op_array);
 
 	/* Do actual optimizations */
 	zend_optimize(op_array, ctx);
 
-	/* Redo pass_two() */
-	zend_redo_pass_two(op_array);
-
+	/* Live ranges must be recalculated before scope fn fixup. */
 	if (op_array->live_range) {
 		zend_recalc_live_ranges(op_array, NULL);
 	}
+
+	if (is_scope_fn) {
+		zend_refixup_scope_func_self(op_array, saved_scope_ex_offset, saved_scope_T);
+	}
+
+	/* Redo pass_two() */
+	zend_redo_pass_two(op_array);
 }
 
 static void zend_adjust_fcall_stack_size(const zend_op_array *op_array, const zend_optimizer_ctx *ctx)
@@ -1576,6 +1602,12 @@ static void step_optimize_op_array(zend_op_array *op_array, void *context) {
 	zend_optimize_op_array(op_array, (zend_optimizer_ctx *) context);
 }
 
+static void step_optimize_scope_fn_op_array(zend_op_array *op_array, void *context) {
+	if (zend_op_array_is_scope_fn(op_array)) {
+		zend_optimize_op_array(op_array, (zend_optimizer_ctx *) context);
+	}
+}
+
 static void step_adjust_fcall_stack_size(zend_op_array *op_array, void *context) {
 	zend_adjust_fcall_stack_size(op_array, (zend_optimizer_ctx *) context);
 }
@@ -1634,6 +1666,12 @@ ZEND_API void zend_optimize_script(zend_script *script, zend_long optimization_l
 		}
 
 		for (i = 0; i < call_graph.op_arrays_count; i++) {
+			/* Single-op_array SSA can't see scope-fn children's reads of
+			 * parent CVs. Subsequent DFA-based passes (DCE, SCCP) would
+			 * treat parent CVs only used by a child as dead. */
+			if (call_graph.op_arrays[i]->fn_flags2 & ZEND_ACC2_HAS_TRACKED_TEMPORARIES) {
+				continue;
+			}
 			func_info = ZEND_FUNC_INFO(call_graph.op_arrays[i]);
 			if (func_info) {
 				if (zend_dfa_analyze_op_array(call_graph.op_arrays[i], &ctx, &func_info->ssa) == SUCCESS) {
@@ -1646,6 +1684,9 @@ ZEND_API void zend_optimize_script(zend_script *script, zend_long optimization_l
 
 		//TODO: perform inner-script inference???
 		for (i = 0; i < call_graph.op_arrays_count; i++) {
+			if (call_graph.op_arrays[i]->fn_flags2 & ZEND_ACC2_HAS_TRACKED_TEMPORARIES) {
+				continue;
+			}
 			func_info = ZEND_FUNC_INFO(call_graph.op_arrays[i]);
 			if (func_info) {
 				zend_dfa_optimize_op_array(call_graph.op_arrays[i], &ctx, &func_info->ssa, func_info->call_map);
@@ -1685,14 +1726,9 @@ ZEND_API void zend_optimize_script(zend_script *script, zend_long optimization_l
 			}
 		}
 
-		if (ZEND_OPTIMIZER_PASS_12 & optimization_level) {
-			for (i = 0; i < call_graph.op_arrays_count; i++) {
-				zend_adjust_fcall_stack_size_graph(call_graph.op_arrays[i]);
-			}
-		}
-
 		for (i = 0; i < call_graph.op_arrays_count; i++) {
 			op_array = call_graph.op_arrays[i];
+			zend_op *old_opcodes = op_array->opcodes;
 			func_info = ZEND_FUNC_INFO(op_array);
 			if (func_info && func_info->ssa.var_info) {
 				zend_redo_pass_two_ex(op_array, &func_info->ssa);
@@ -1705,6 +1741,31 @@ ZEND_API void zend_optimize_script(zend_script *script, zend_long optimization_l
 					zend_recalc_live_ranges(op_array, NULL);
 				}
 			}
+			/* Update offsets for pass 12, as zend_redo_pass_two may reallocate */
+			if (func_info && op_array->opcodes != old_opcodes) {
+				ptrdiff_t delta = op_array->opcodes - old_opcodes;
+				zend_call_info *call = func_info->callee_info;
+				while (call) {
+					if (call->caller_init_opline) {
+						call->caller_init_opline += delta;
+					}
+					if (call->caller_call_opline) {
+						call->caller_call_opline += delta;
+					}
+					call = call->next_callee;
+				}
+			}
+		}
+
+		/* zend_build_call_graph ignores scope fns, so explicitely optimize them here. */
+		zend_foreach_op_array(script, step_optimize_scope_fn_op_array, &ctx);
+
+		/* PASS_12 requires knowledge of the final callee->T to compute INIT_FCALL stack sizes;
+		 * for scope fns these change during zend_redo_pass_two, hence we do this late. */
+		if (ZEND_OPTIMIZER_PASS_12 & optimization_level) {
+			for (i = 0; i < call_graph.op_arrays_count; i++) {
+				zend_adjust_fcall_stack_size_graph(call_graph.op_arrays[i]);
+			}
 		}
 
 		for (i = 0; i < call_graph.op_arrays_count; i++) {

Zend/tests/function_arguments/gh20435_2.phpt

@@ -1,5 +1,5 @@
 --TEST--
-GH-20435: ZEND_CALL_HAS_EXTRA_NAMED_PARAMS & magic methods in debug_backtrace_get_args()
+GH-20435: ZEND_CALL_MAYBE_HAS_EXTRA_NAMED_PARAMS & magic methods in debug_backtrace_get_args()
 --FILE--
 <?php