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));
 
@@ -4385,6 +4417,9 @@ 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);
@@ -4392,14 +4427,16 @@ iseq_compile_each(rb_iseq_t *iseq, LINK_ANCHOR *ret, NODE * node, int poped)
 	}
 	else {
 	    COMPILE(ret, "NODE_OP_ASGN2 val", node->nd_value);
-	    ADD_SEND(ret, line, node->nd_next->nd_mid,
+	    ADD_SEND(ret, line, atype, INT2FIX(1));
-		     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_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) NEW_NODE(NODE_OP_ASGN2,i,o,0)
+#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)

parse.y

@@ -371,6 +371,9 @@ static int parser_yyerror(struct parser_params*, const char*);
 #define ruby_coverage		(parser->coverage)
 #endif
 
+#define NODE_CALL_Q(q) (((q) == tDOTQ) ? NODE_QCALL : NODE_CALL)
+#define NEW_QCALL(q,r,m,a) NEW_NODE(NODE_CALL_Q(q),r,m,a)
+
 static int yylex(YYSTYPE*, struct parser_params*);
 
 #ifndef RIPPER
@@ -457,8 +460,8 @@ static NODE *node_assign_gen(struct parser_params*,NODE*,NODE*);
 #define node_assign(node1, node2) node_assign_gen(parser, (node1), (node2))
 
 static NODE *new_op_assign_gen(struct parser_params *parser, NODE *lhs, ID op, NODE *rhs);
-static NODE *new_attr_op_assign_gen(struct parser_params *parser, NODE *lhs, ID attr, ID op, NODE *rhs);
+static NODE *new_attr_op_assign_gen(struct parser_params *parser, NODE *lhs, ID atype, ID attr, ID op, NODE *rhs);
-#define new_attr_op_assign(lhs, type, attr, op, rhs) new_attr_op_assign_gen(parser, (lhs), (attr), (op), (rhs))
+#define new_attr_op_assign(lhs, type, attr, op, rhs) new_attr_op_assign_gen(parser, (lhs), (type), (attr), (op), (rhs))
 static NODE *new_const_op_assign_gen(struct parser_params *parser, NODE *lhs, ID op, NODE *rhs);
 #define new_const_op_assign(lhs, op, rhs) new_const_op_assign_gen(parser, (lhs), (op), (rhs))
 
@@ -844,7 +847,7 @@ static void token_info_pop(struct parser_params*, const char *token, size_t len)
 %type <node> mlhs mlhs_head mlhs_basic mlhs_item mlhs_node mlhs_post mlhs_inner
 %type <id>   fsym keyword_variable user_variable sym symbol operation operation2 operation3
 %type <id>   cname fname op f_rest_arg f_block_arg opt_f_block_arg f_norm_arg f_bad_arg
-%type <id>   f_kwrest f_label f_arg_asgn
+%type <id>   f_kwrest f_label f_arg_asgn call_op
 /*%%%*/
 /*%
 %type <val> program reswords then do dot_or_colon
@@ -869,6 +872,7 @@ static void token_info_pop(struct parser_params*, const char *token, size_t len)
 %token tASET		RUBY_TOKEN(ASET)   "[]="
 %token tLSHFT		RUBY_TOKEN(LSHFT)  "<<"
 %token tRSHFT		RUBY_TOKEN(RSHFT)  ">>"
+%token tDOTQ		RUBY_TOKEN(DOTQ)   ".?"
 %token tCOLON2		"::"
 %token tCOLON3		":: at EXPR_BEG"
 %token <id> tOP_ASGN	/* +=, -=  etc. */
@@ -1260,15 +1264,15 @@ stmt		: keyword_alias fitem {lex_state = EXPR_FNAME;} fitem
 			$$ = dispatch3(opassign, $$, $5, $6);
 		    %*/
 		    }
-		| primary_value '.' tIDENTIFIER tOP_ASGN command_call
+		| primary_value call_op tIDENTIFIER tOP_ASGN command_call
 		    {
 			value_expr($5);
-			$$ = new_attr_op_assign($1, '.', $3, $4, $5);
+			$$ = new_attr_op_assign($1, $2, $3, $4, $5);
 		    }
-		| primary_value '.' tCONSTANT tOP_ASGN command_call
+		| primary_value call_op tCONSTANT tOP_ASGN command_call
 		    {
 			value_expr($5);
-			$$ = new_attr_op_assign($1, '.', $3, $4, $5);
+			$$ = new_attr_op_assign($1, $2, $3, $4, $5);
 		    }
 		| primary_value tCOLON2 tCONSTANT tOP_ASGN command_call
 		    {
@@ -1456,24 +1460,24 @@ command		: fcall command_args       %prec tLOWEST
 			$$ = method_add_block($$, $3);
 		    %*/
 		    }
-		| primary_value '.' operation2 command_args	%prec tLOWEST
+		| primary_value call_op operation2 command_args	%prec tLOWEST
 		    {
 		    /*%%%*/
-			$$ = NEW_CALL($1, $3, $4);
+			$$ = NEW_QCALL($2, $1, $3, $4);
 			fixpos($$, $1);
 		    /*%
-			$$ = dispatch4(command_call, $1, ripper_id2sym('.'), $3, $4);
+			$$ = dispatch4(command_call, $1, ripper_id2sym($2), $3, $4);
 		    %*/
 		    }
-		| primary_value '.' operation2 command_args cmd_brace_block
+		| primary_value call_op operation2 command_args cmd_brace_block
 		    {
 		    /*%%%*/
 			block_dup_check($4,$5);
-		        $5->nd_iter = NEW_CALL($1, $3, $4);
+		        $5->nd_iter = NEW_QCALL($2, $1, $3, $4);
 			$$ = $5;
 			fixpos($$, $1);
 		    /*%
-			$$ = dispatch4(command_call, $1, ripper_id2sym('.'), $3, $4);
+			$$ = dispatch4(command_call, $1, ripper_id2sym($2), $3, $4);
 			$$ = method_add_block($$, $5);
 		    %*/
 		   }
@@ -1713,12 +1717,12 @@ mlhs_node	: user_variable
 			$$ = dispatch2(aref_field, $1, escape_Qundef($3));
 		    %*/
 		    }
-		| primary_value '.' tIDENTIFIER
+		| primary_value call_op tIDENTIFIER
 		    {
 		    /*%%%*/
 			$$ = attrset($1, $3);
 		    /*%
-			$$ = dispatch3(field, $1, ripper_id2sym('.'), $3);
+			$$ = dispatch3(field, $1, ripper_id2sym($2), $3);
 		    %*/
 		    }
 		| primary_value tCOLON2 tIDENTIFIER
@@ -1729,12 +1733,12 @@ mlhs_node	: user_variable
 			$$ = dispatch2(const_path_field, $1, $3);
 		    %*/
 		    }
-		| primary_value '.' tCONSTANT
+		| primary_value call_op tCONSTANT
 		    {
 		    /*%%%*/
 			$$ = attrset($1, $3);
 		    /*%
-			$$ = dispatch3(field, $1, ripper_id2sym('.'), $3);
+			$$ = dispatch3(field, $1, ripper_id2sym($2), $3);
 		    %*/
 		    }
 		| primary_value tCOLON2 tCONSTANT
@@ -1804,12 +1808,12 @@ lhs		: user_variable
 			$$ = dispatch2(aref_field, $1, escape_Qundef($3));
 		    %*/
 		    }
-		| primary_value '.' tIDENTIFIER
+		| primary_value call_op tIDENTIFIER
 		    {
 		    /*%%%*/
 			$$ = attrset($1, $3);
 		    /*%
-			$$ = dispatch3(field, $1, ripper_id2sym('.'), $3);
+			$$ = dispatch3(field, $1, ripper_id2sym($2), $3);
 		    %*/
 		    }
 		| primary_value tCOLON2 tIDENTIFIER
@@ -1820,12 +1824,12 @@ lhs		: user_variable
 			$$ = dispatch3(field, $1, ID2SYM(idCOLON2), $3);
 		    %*/
 		    }
-		| primary_value '.' tCONSTANT
+		| primary_value call_op tCONSTANT
 		    {
 		    /*%%%*/
 			$$ = attrset($1, $3);
 		    /*%
-			$$ = dispatch3(field, $1, ripper_id2sym('.'), $3);
+			$$ = dispatch3(field, $1, ripper_id2sym($2), $3);
 		    %*/
 		    }
 		| primary_value tCOLON2 tCONSTANT
@@ -2064,15 +2068,15 @@ arg		: lhs '=' arg
 			$$ = dispatch3(opassign, $1, $5, $6);
 		    %*/
 		    }
-		| primary_value '.' tIDENTIFIER tOP_ASGN arg
+		| primary_value call_op tIDENTIFIER tOP_ASGN arg
 		    {
 			value_expr($5);
-			$$ = new_attr_op_assign($1, '.', $3, $4, $5);
+			$$ = new_attr_op_assign($1, $2, $3, $4, $5);
 		    }
-		| primary_value '.' tCONSTANT tOP_ASGN arg
+		| primary_value call_op tCONSTANT tOP_ASGN arg
 		    {
 			value_expr($5);
-			$$ = new_attr_op_assign($1, '.', $3, $4, $5);
+			$$ = new_attr_op_assign($1, $2, $3, $4, $5);
 		    }
 		| primary_value tCOLON2 tIDENTIFIER tOP_ASGN arg
 		    {
@@ -3648,7 +3652,7 @@ method_call	: fcall paren_args
 			$$ = method_arg(dispatch1(fcall, $1), $2);
 		    %*/
 		    }
-		| primary_value '.' operation2
+		| primary_value call_op operation2
 		    {
 		    /*%%%*/
 			$<num>$ = ruby_sourceline;
@@ -3657,10 +3661,10 @@ method_call	: fcall paren_args
 		  opt_paren_args
 		    {
 		    /*%%%*/
-			$$ = NEW_CALL($1, $3, $5);
+			$$ = NEW_QCALL($2, $1, $3, $5);
 			nd_set_line($$, $<num>4);
 		    /*%
-			$$ = dispatch3(call, $1, ripper_id2sym('.'), $3);
+			$$ = dispatch3(call, $1, ripper_id2sym($2), $3);
 			$$ = method_optarg($$, $5);
 		    %*/
 		    }
@@ -3688,7 +3692,7 @@ method_call	: fcall paren_args
 			$$ = dispatch3(call, $1, ID2SYM(idCOLON2), $3);
 		    %*/
 		    }
-		| primary_value '.'
+		| primary_value call_op
 		    {
 		    /*%%%*/
 			$<num>$ = ruby_sourceline;
@@ -3697,10 +3701,10 @@ method_call	: fcall paren_args
 		  paren_args
 		    {
 		    /*%%%*/
-			$$ = NEW_CALL($1, idCall, $4);
+			$$ = NEW_QCALL($2, $1, idCall, $4);
 			nd_set_line($$, $<num>3);
 		    /*%
-			$$ = dispatch3(call, $1, ripper_id2sym('.'),
+			$$ = dispatch3(call, $1, ripper_id2sym($2),
 				       ID2SYM(idCall));
 			$$ = method_optarg($$, $4);
 		    %*/
@@ -5103,7 +5107,7 @@ operation3	: tIDENTIFIER
 		| op
 		;
 
-dot_or_colon	: '.'
+dot_or_colon	: call_op
 		    /*%c%*/
 		    /*%c
 		    { $$ = $<val>1; }
@@ -5115,6 +5119,10 @@ dot_or_colon	: '.'
 		    %*/
 		;
 
+call_op 	: '.' {$$ = '.';}
+		| tDOTQ {$$ = tDOTQ;}
+		;
+
 opt_terms	: /* none */
 		| terms
 		;
@@ -8356,6 +8364,10 @@ parser_yylex(struct parser_params *parser)
 	    pushback(c);
 	    return tDOT2;
 	}
+	if (c == '?') {
+	    lex_state = EXPR_DOT;
+	    return tDOTQ;
+	}
 	pushback(c);
 	if (c != -1 && ISDIGIT(c)) {
 	    yyerror("no .<digit> floating literal anymore; put 0 before dot");
@@ -10036,7 +10048,8 @@ new_op_assign_gen(struct parser_params *parser, NODE *lhs, ID op, NODE *rhs)
 }
 
 static NODE *
-new_attr_op_assign_gen(struct parser_params *parser, NODE *lhs, ID attr, ID op, NODE *rhs)
+new_attr_op_assign_gen(struct parser_params *parser, NODE *lhs,
+		       ID atype, ID attr, ID op, NODE *rhs)
 {
     NODE *asgn;
 
@@ -10046,7 +10059,7 @@ new_attr_op_assign_gen(struct parser_params *parser, NODE *lhs, ID attr, ID op,
     else if (op == tANDOP) {
 	op = 1;
     }
-    asgn = NEW_OP_ASGN2(lhs, attr, op, rhs);
+    asgn = NEW_OP_ASGN2(lhs, (atype == tDOTQ), attr, op, rhs);
     fixpos(asgn, lhs);
     return asgn;
 }

template/id.h.tmpl

@@ -18,7 +18,7 @@ op_id_offset = 128
 token_op_ids = %w[
   tDOT2 tDOT3 tUPLUS tUMINUS tPOW tDSTAR tCMP tLSHFT tRSHFT
   tLEQ tGEQ tEQ tEQQ tNEQ tMATCH tNMATCH tAREF tASET
-  tCOLON2 tCOLON3 tANDOP tOROP
+  tCOLON2 tCOLON3 tANDOP tOROP tDOTQ
 ]
 
 defs = File.join(File.dirname(File.dirname(erb.filename)), "defs/id.def")

test/ruby/test_call.rb

@@ -31,4 +31,16 @@ class TestCall < Test::Unit::TestCase
     assert_nothing_raised(ArgumentError) {o.foo}
     assert_raise_with_message(ArgumentError, e.message, bug9622) {o.foo(100)}
   end
+
+  def test_safe_call
+    s = Struct.new(:x, :y)
+    o = s.new("x")
+    assert_equal("X", o.x.?upcase)
+    assert_nil(o.y.?upcase)
+    assert_equal("x", o.x)
+    o.?x = 6
+    assert_equal(6, o.x)
+    o.?x *= 7
+    assert_equal(42, o.x)
+  end
 end