Safe navigation operator

* compile.c (iseq_peephole_optimize): peephole optimization for
  branchnil jumps.
* compile.c (iseq_compile_each): generate save navigation operator
  code.
* insns.def (branchnil): new opcode to pop the tos and branch if
  it is nil.
* parse.y (NEW_QCALL, call_op, parser_yylex): parse token '.?'.
  [Feature #11537]

git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@52214 b2dd03c8-39d4-4d8f-98ff-823fe69b080e

ChangeLog

@@ -1,3 +1,17 @@
+Thu Oct 22 15:30:08 2015  Nobuyoshi Nakada  <nobu@ruby-lang.org>
+
+	* compile.c (iseq_peephole_optimize): peephole optimization for
+	  branchnil jumps.
+
+	* compile.c (iseq_compile_each): generate save navigation operator
+	  code.
+
+	* insns.def (branchnil): new opcode to pop the tos and branch if
+	  it is nil.
+
+	* parse.y (NEW_QCALL, call_op, parser_yylex): parse token '.?'.
+	  [Feature #11537]
+
 Thu Oct 22 13:16:19 2015  Guilherme Reis Campos  <guilhermekbsa@gmail.com>
 
 	* dir.c (ruby_brace_expand): glob brace expansion edge case fix.

NEWS

@@ -18,6 +18,17 @@ with all sufficient information, see the ChangeLog file.
   * besides, --enable/--disable=frozen-string-literal options also have
     been introduced.
 
+* safe navigation operator:
+
+  * new method call syntax, `object.?foo', method #foo is called on
+    `object' if it is not nil.
+    this is similar to `try!' in ActiveSupport, except for:
+    * method name is syntactically required
+          obj.try! {} # valid
+          obj.? {}    # syntax error
+    * attribute assignment is valid
+          obj.?attr += 1
+
 === Core classes updates (outstanding ones only)
 
 * ARGF

compile.c

@@ -1942,6 +1942,7 @@ iseq_peephole_optimize(rb_iseq_t *iseq, LINK_ELEMENT *list, const int do_tailcal
     }
 
     if (iobj->insn_id == BIN(branchif) ||
+	iobj->insn_id == BIN(branchnil) ||
 	iobj->insn_id == BIN(branchunless)) {
 	/*
 	 *   if L1
@@ -1955,6 +1956,31 @@ iseq_peephole_optimize(rb_iseq_t *iseq, LINK_ELEMENT *list, const int do_tailcal
 	if (nobj->insn_id == BIN(jump)) {
 	    OPERAND_AT(iobj, 0) = OPERAND_AT(nobj, 0);
 	}
+
+	if (nobj->insn_id == BIN(dup)) {
+	    /*
+	     *   dup
+	     *   if L1
+	     *   ...
+	     * L1:
+	     *   dup
+	     *   if L2
+	     * =>
+	     *   dup
+	     *   if L2
+	     *   ...
+	     * L1:
+	     *   dup
+	     *   if L2
+	     */
+	    INSN *pobj = (INSN *)iobj->link.prev;
+	    nobj = (INSN *)nobj->link.next;
+	    /* basic blocks, with no labels in the middle */
+	    if ((pobj && pobj->insn_id == BIN(dup)) &&
+		(nobj && nobj->insn_id == iobj->insn_id)) {
+		OPERAND_AT(iobj, 0) = OPERAND_AT(nobj, 0);
+	    }
+	}
     }
 
     if (do_tailcallopt && iobj->insn_id == BIN(leave)) {
@@ -4319,6 +4345,7 @@ iseq_compile_each(rb_iseq_t *iseq, LINK_ANCHOR *ret, NODE * node, int poped)
 	VALUE asgnflag;
 	LABEL *lfin = NEW_LABEL(line);
 	LABEL *lcfin = NEW_LABEL(line);
+	LABEL *lskip = 0;
 	/*
 	  class C; attr_accessor :c; end
 	  r = C.new
@@ -4362,6 +4389,11 @@ iseq_compile_each(rb_iseq_t *iseq, LINK_ANCHOR *ret, NODE * node, int poped)
 	*/
 
 	asgnflag = COMPILE_RECV(ret, "NODE_OP_ASGN2#recv", node);
+	if (node->nd_next->nd_aid) {
+	    lskip = NEW_LABEL(line);
+	    ADD_INSN(ret, line, dup);
+	    ADD_INSNL(ret, line, branchnil, lskip);
+	}
 	ADD_INSN(ret, line, dup);
 	ADD_SEND(ret, line, vid, INT2FIX(0));
 
@@ -4385,21 +4417,26 @@ iseq_compile_each(rb_iseq_t *iseq, LINK_ANCHOR *ret, NODE * node, int poped)
 
 	    ADD_LABEL(ret, lfin);
 	    ADD_INSN(ret, line, pop);
+	    if (lskip) {
+		ADD_LABEL(ret, lskip);
+	    }
 	    if (poped) {
 		/* we can apply more optimize */
 		ADD_INSN(ret, line, pop);
 	    }
 	}
 	else {
 	    COMPILE(ret, "NODE_OP_ASGN2 val", node->nd_value);
-	    ADD_SEND(ret, line, node->nd_next->nd_mid,
-		     INT2FIX(1));
+	    ADD_SEND(ret, line, atype, INT2FIX(1));
 	    if (!poped) {
 		ADD_INSN(ret, line, swap);
 		ADD_INSN1(ret, line, topn, INT2FIX(1));
 	    }
 	    ADD_SEND_WITH_FLAG(ret, line, aid, INT2FIX(1), INT2FIX(asgnflag));
 	    ADD_INSN(ret, line, pop);
+	    if (lskip) {
+		ADD_LABEL(ret, lskip);
+	    }
 	}
 	break;
       }
@@ -4548,6 +4585,7 @@ iseq_compile_each(rb_iseq_t *iseq, LINK_ANCHOR *ret, NODE * node, int poped)
 	    }
 	    break;
 	}
+      case NODE_QCALL:
       case NODE_FCALL:
       case NODE_VCALL:{		/* VCALL: variable or call */
 	/*
@@ -4557,6 +4595,7 @@ iseq_compile_each(rb_iseq_t *iseq, LINK_ANCHOR *ret, NODE * node, int poped)
 	*/
 	DECL_ANCHOR(recv);
 	DECL_ANCHOR(args);
+	LABEL *lskip = 0;
 	ID mid = node->nd_mid;
 	VALUE argc;
 	unsigned int flag = 0;
@@ -4631,8 +4670,13 @@ iseq_compile_each(rb_iseq_t *iseq, LINK_ANCHOR *ret, NODE * node, int poped)
 	}
 #endif
 	/* receiver */
-	if (type == NODE_CALL) {
+	if (type == NODE_CALL || type == NODE_QCALL) {
 	    COMPILE(recv, "recv", node->nd_recv);
+	    if (type == NODE_QCALL) {
+		lskip = NEW_LABEL(line);
+		ADD_INSN(recv, line, dup);
+		ADD_INSNL(recv, line, branchnil, lskip);
+	    }
 	}
 	else if (type == NODE_FCALL || type == NODE_VCALL) {
 	    ADD_CALL_RECEIVER(recv, line);
@@ -4662,6 +4706,9 @@ iseq_compile_each(rb_iseq_t *iseq, LINK_ANCHOR *ret, NODE * node, int poped)
 
 	ADD_SEND_R(ret, line, mid, argc, parent_block, INT2FIX(flag), keywords);
 
+	if (lskip) {
+	    ADD_LABEL(ret, lskip);
+	}
 	if (poped) {
 	    ADD_INSN(ret, line, pop);
 	}

doc/syntax/calling_methods.rdoc

@@ -27,6 +27,10 @@ This sends the +my_method+ message to +my_object+.  Any object can be a
 receiver but depending on the method's visibility sending a message may raise a
 NoMethodError.
 
+You may use <code>.?</code> to designate a receiver, then +my_method+ is not
+invoked and the result is +nil+ when the receiver is +nil+.  In that case, the
+argument of +my_method+ are not evaluated.
+
 You may also use <code>::</code> to designate a receiver, but this is rarely
 used due to the potential for confusion with <code>::</code> for namespaces.
 

insns.def

@@ -1132,6 +1132,23 @@ branchunless
     }
 }
 
+/**
+  @c jump
+  @e if val is nil, set PC to (PC + dst).
+  @j もし val が nil ならば、PC を (PC + dst) にする。
+ */
+DEFINE_INSN
+branchnil
+(OFFSET dst)
+(VALUE val)
+()
+{
+    if (NIL_P(val)) {
+	RUBY_VM_CHECK_INTS(th);
+	JUMP(dst);
+    }
+}
+
 
 /**********************************************************/
 /* for optimize                                           */

node.c

@@ -357,7 +357,10 @@ dump_node(VALUE buf, VALUE indent, int comment, NODE *node)
 	ANN("          where [attr]: [nd_next->nd_vid]");
 	ANN("example: struct.field += foo");
 	F_NODE(nd_recv, "receiver");
-	F_ID(nd_next->nd_vid, "attr");
+	F_CUSTOM1(nd_next->nd_vid, "attr") {
+	    if (node->nd_next->nd_aid) A("? ");
+	    A_ID(node->nd_next->nd_vid);
+	}
 	F_CUSTOM1(nd_next->nd_mid, "operator") {
 	    switch (node->nd_next->nd_mid) {
 	      case 0: A("0 (||)"); break;

node.h

@@ -96,6 +96,8 @@ enum node_type {
 #define NODE_FCALL       NODE_FCALL
     NODE_VCALL,
 #define NODE_VCALL       NODE_VCALL
+    NODE_QCALL,
+#define NODE_QCALL       NODE_QCALL
     NODE_SUPER,
 #define NODE_SUPER       NODE_SUPER
     NODE_ZSUPER,
@@ -394,8 +396,8 @@ typedef struct RNode {
 #define NEW_CVASGN(v,val) NEW_NODE(NODE_CVASGN,v,val,0)
 #define NEW_CVDECL(v,val) NEW_NODE(NODE_CVDECL,v,val,0)
 #define NEW_OP_ASGN1(p,id,a) NEW_NODE(NODE_OP_ASGN1,p,id,a)
-#define NEW_OP_ASGN2(r,i,o,val) NEW_NODE(NODE_OP_ASGN2,r,val,NEW_OP_ASGN22(i,o))
-#define NEW_OP_ASGN22(i,o) NEW_NODE(NODE_OP_ASGN2,i,o,0)
+#define NEW_OP_ASGN2(r,t,i,o,val) NEW_NODE(NODE_OP_ASGN2,r,val,NEW_OP_ASGN22(i,o,t))
+#define NEW_OP_ASGN22(i,o,t) NEW_NODE(NODE_OP_ASGN2,i,o,t)
 #define NEW_OP_ASGN_OR(i,val) NEW_NODE(NODE_OP_ASGN_OR,i,val,0)
 #define NEW_OP_ASGN_AND(i,val) NEW_NODE(NODE_OP_ASGN_AND,i,val,0)
 #define NEW_OP_CDECL(v,op,val) NEW_NODE(NODE_OP_CDECL,v,val,op)