Add 3 '*' operators: one prefix, one infix, and one postfix, per #523. (

#582) The presence or absence of whitespace is used to determine which operator is in use, following the rules described in #520. Support for prefix * dereference operator follows #523. Co-authored-by: Geoff Romer <gromer@google.com>
carbon-language · Jun 22, 2021 · 89e2111 · 89e2111
1 parent d6b47ba
commit 89e2111
Show file tree

Hide file tree

Showing 12 changed files with 179 additions and 21 deletions.
diff --git a/executable_semantics/BUILD b/executable_semantics/BUILD
@@ -61,6 +61,7 @@ EXAMPLES = [
     "pattern_init",
     "pattern_variable_fail",
     "record1",
+    "star",
     "struct1",
     "struct2",
     "struct3",

diff --git a/executable_semantics/ast/expression.cpp b/executable_semantics/ast/expression.cpp
@@ -245,15 +245,18 @@ auto Expression::MakeIndex(int line_num, const Expression* exp,
 
 static void PrintOp(Operator op) {
   switch (op) {
-    case Operator::Neg:
-      std::cout << "-";
-      break;
     case Operator::Add:
       std::cout << "+";
       break;
+    case Operator::Neg:
     case Operator::Sub:
       std::cout << "-";
       break;
+    case Operator::Mul:
+    case Operator::Deref:
+    case Operator::Ptr:
+      std::cout << "*";
+      break;
     case Operator::Not:
       std::cout << "not";
       break;

diff --git a/executable_semantics/ast/expression.h b/executable_semantics/ast/expression.h
@@ -42,11 +42,14 @@ enum class ExpressionKind {
 enum class Operator {
   Add,
   And,
+  Deref,
   Eq,
+  Mul,
   Neg,
   Not,
   Or,
   Sub,
+  Ptr,
 };
 
 struct Expression;

diff --git a/executable_semantics/interpreter/interpreter.cpp b/executable_semantics/interpreter/interpreter.cpp
@@ -281,6 +281,9 @@ auto EvalPrim(Operator op, const std::vector<const Value*>& args, int line_num)
     case Operator::Sub:
       return Value::MakeIntVal(ValToInt(args[0], line_num) -
                                ValToInt(args[1], line_num));
+    case Operator::Mul:
+      return Value::MakeIntVal(ValToInt(args[0], line_num) *
+                               ValToInt(args[1], line_num));
     case Operator::Not:
       return Value::MakeBoolVal(!ValToBool(args[0], line_num));
     case Operator::And:
@@ -291,6 +294,11 @@ auto EvalPrim(Operator op, const std::vector<const Value*>& args, int line_num)
                                 ValToBool(args[1], line_num));
     case Operator::Eq:
       return Value::MakeBoolVal(ValueEqual(args[0], args[1], line_num));
+    case Operator::Ptr:
+      return Value::MakePtrTypeVal(args[0]);
+    case Operator::Deref:
+      std::cerr << line_num << ": dereference not implemented yet\n";
+      exit(-1);
   }
 }
 

diff --git a/executable_semantics/interpreter/typecheck.cpp b/executable_semantics/interpreter/typecheck.cpp
@@ -30,6 +30,18 @@ void ExpectType(int line_num, const std::string& context, const Value* expected,
   }
 }
 
+void ExpectPointerType(int line_num, const std::string& context,
+                       const Value* actual) {
+  if (actual->tag != ValKind::PointerTV) {
+    std::cerr << line_num << ": type error in " << context << std::endl;
+    std::cerr << "expected a pointer type\n";
+    std::cerr << "actual: ";
+    PrintValue(actual, std::cerr);
+    std::cerr << std::endl;
+    exit(-1);
+  }
+}
+
 void PrintErrorString(const std::string& s) { std::cerr << s; }
 
 void PrintTypeEnv(TypeEnv types, std::ostream& out) {
@@ -71,6 +83,9 @@ auto ReifyType(const Value* t, int line_num) -> const Expression* {
       return Expression::MakeVar(0, *t->GetStructType().name);
     case ValKind::ChoiceTV:
       return Expression::MakeVar(0, *t->GetChoiceType().name);
+    case ValKind::PointerTV:
+      return Expression::MakeUnOp(
+          0, Operator::Ptr, ReifyType(t->GetPointerType().type, line_num));
     default:
       std::cerr << line_num << ": expected a type, not ";
       PrintValue(t, std::cerr);
@@ -317,10 +332,21 @@ auto TypeCheckExp(const Expression* e, TypeEnv types, Env values,
           ExpectType(e->line_num, "negation", Value::MakeIntTypeVal(), ts[0]);
           return TCResult(new_e, Value::MakeIntTypeVal(), new_types);
         case Operator::Add:
+          ExpectType(e->line_num, "addition(1)", Value::MakeIntTypeVal(),
+                     ts[0]);
+          ExpectType(e->line_num, "addition(2)", Value::MakeIntTypeVal(),
+                     ts[1]);
+          return TCResult(new_e, Value::MakeIntTypeVal(), new_types);
         case Operator::Sub:
           ExpectType(e->line_num, "subtraction(1)", Value::MakeIntTypeVal(),
                      ts[0]);
-          ExpectType(e->line_num, "substration(2)", Value::MakeIntTypeVal(),
+          ExpectType(e->line_num, "subtraction(2)", Value::MakeIntTypeVal(),
+                     ts[1]);
+          return TCResult(new_e, Value::MakeIntTypeVal(), new_types);
+        case Operator::Mul:
+          ExpectType(e->line_num, "multiplication(1)", Value::MakeIntTypeVal(),
+                     ts[0]);
+          ExpectType(e->line_num, "multiplication(2)", Value::MakeIntTypeVal(),
                      ts[1]);
           return TCResult(new_e, Value::MakeIntTypeVal(), new_types);
         case Operator::And:
@@ -337,6 +363,12 @@ auto TypeCheckExp(const Expression* e, TypeEnv types, Env values,
         case Operator::Eq:
           ExpectType(e->line_num, "==", ts[0], ts[1]);
           return TCResult(new_e, Value::MakeBoolTypeVal(), new_types);
+        case Operator::Deref:
+          ExpectPointerType(e->line_num, "*", ts[0]);
+          return TCResult(new_e, ts[0]->GetPointerType().type, new_types);
+        case Operator::Ptr:
+          ExpectType(e->line_num, "*", Value::MakeTypeTypeVal(), ts[0]);
+          return TCResult(new_e, Value::MakeTypeTypeVal(), new_types);
       }
       break;
     }

diff --git a/executable_semantics/interpreter/value.cpp b/executable_semantics/interpreter/value.cpp
@@ -357,9 +357,8 @@ auto PrintValue(const Value* val, std::ostream& out) -> void {
       out << "Continuation";
       break;
     case ValKind::PointerTV:
-      out << "Ptr(";
       PrintValue(val->GetPointerType().type, out);
-      out << ")";
+      out << "*";
       break;
     case ValKind::FunctionTV:
       out << "fn ";

diff --git a/executable_semantics/syntax/lexer.lpp b/executable_semantics/syntax/lexer.lpp
@@ -18,6 +18,13 @@ SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 */
 %option yylineno
 
+/* Lexing a token immediately after consuming some whitespace. */
+%s AFTER_WHITESPACE
+/* Lexing a token immediately after consuming an operand-ending token:
+ * a closing bracket, identifier, or literal.
+ */
+%s AFTER_OPERAND
+
 AND               "and"
 ARROW             "->"
 AUTO              "auto"
@@ -53,13 +60,21 @@ AWAIT             "__await"
 identifier    [A-Za-z_][A-Za-z0-9_]*
 integer_literal   [0-9]+
 horizontal_whitespace [ \t\r]
+whitespace [ \t\r\n]
+operand_start [(A-Za-z0-9_"]
 
 %{
-  // This macro is expanded to run each time a token is recognized.
+  // This macro is expanded immediately before each action specified below.
   //
   // Advances the current token position by yyleng columns without changing
-  // the line number.
-  # define YY_USER_ACTION context.current_token_position.columns(yyleng);
+  // the line number, and takes us out of the after-whitespace / after-operand
+  // state.
+  # define YY_USER_ACTION \
+      context.current_token_position.columns(yyleng); \
+      if (YY_START == AFTER_WHITESPACE || \
+          YY_START == AFTER_OPERAND) { \
+        BEGIN(INITIAL); \
+      }
 %}
 
 %%
@@ -105,27 +120,66 @@ horizontal_whitespace [ \t\r]
 "=" return yy::parser::make_EQUAL(context.current_token_position);
 "-" return yy::parser::make_MINUS(context.current_token_position);
 "+" return yy::parser::make_PLUS(context.current_token_position);
-"*" return yy::parser::make_STAR(context.current_token_position);
 "/" return yy::parser::make_SLASH(context.current_token_position);
 "(" return yy::parser::make_LEFT_PARENTHESIS(context.current_token_position);
-")" return yy::parser::make_RIGHT_PARENTHESIS(context.current_token_position);
+")" { BEGIN(AFTER_OPERAND); return yy::parser::make_RIGHT_PARENTHESIS(context.current_token_position); }
 "{" return yy::parser::make_LEFT_CURLY_BRACE(context.current_token_position);
-"}" return yy::parser::make_RIGHT_CURLY_BRACE(context.current_token_position);
+"}" { BEGIN(AFTER_OPERAND); return yy::parser::make_RIGHT_CURLY_BRACE(context.current_token_position); }
 "[" return yy::parser::make_LEFT_SQUARE_BRACKET(context.current_token_position);
-"]" return yy::parser::make_RIGHT_SQUARE_BRACKET(context.current_token_position);
+"]" { BEGIN(AFTER_OPERAND); return yy::parser::make_RIGHT_SQUARE_BRACKET(context.current_token_position); }
 "." return yy::parser::make_PERIOD(context.current_token_position);
 "," return yy::parser::make_COMMA(context.current_token_position);
 ";" return yy::parser::make_SEMICOLON(context.current_token_position);
 ":" return yy::parser::make_COLON(context.current_token_position);
 
+ /*
+For a `*` operator, we look at whitespace and local context to determine the
+arity and fixity. There are two ways to write a binary operator:
+
+ 1) Whitespace on both sides.
+ 2) Whitespace on neither side, and the previous token is considered to be
+    the end of an operand, and the next token is considered to be the start
+    of an operand.
+
+Otherwise, the operator is unary, but we also check for whitespace to help
+the parser enforce the rule that whitespace is not permitted between the
+operator and its operand, leading to three more cases:
+
+ 3) Whitespace before (but implicitly not after, because that would give a
+    longer match and hit case 1): this can only be a prefix operator.
+ 4) Whitespace after and not before: this can only be a postfix operator.
+ 5) No whitespace on either side (otherwise the longest match would take us
+    to case 4): this is a unary operator and could be either prefix or
+    postfix.
+*/
+<AFTER_WHITESPACE>"*"{whitespace}+ /*case 1*/ {
+  BEGIN(AFTER_WHITESPACE);
+  return yy::parser::make_BINARY_STAR(context.current_token_position);
+}
+<AFTER_OPERAND>"*"/{operand_start} /*case 2*/ {
+  return yy::parser::make_BINARY_STAR(context.current_token_position);
+}
+<AFTER_WHITESPACE>"*" /*case 3*/ {
+  return yy::parser::make_PREFIX_STAR(context.current_token_position);
+}
+<INITIAL,AFTER_OPERAND>"*"{whitespace}+ /*case 4*/ {
+  BEGIN(AFTER_WHITESPACE);
+  return yy::parser::make_POSTFIX_STAR(context.current_token_position);
+}
+<INITIAL,AFTER_OPERAND>"*" /*case 5*/ {
+  return yy::parser::make_UNARY_STAR(context.current_token_position);
+}
+
 {identifier} {
+  BEGIN(AFTER_OPERAND);
   int n = strlen(yytext);
   auto r = reinterpret_cast<char*>(malloc((n + 1) * sizeof(char)));
   strncpy(r, yytext, n + 1);
   return yy::parser::make_identifier(r, context.current_token_position);
 }
 
 {integer_literal} {
+  BEGIN(AFTER_OPERAND);
   auto r = atof(yytext);
   return yy::parser::make_integer_literal(r, context.current_token_position);
 }
@@ -140,13 +194,15 @@ horizontal_whitespace [ \t\r]
 {horizontal_whitespace}+ {
   // Make the span empty by setting start to end.
   context.current_token_position.step();
+  BEGIN(AFTER_WHITESPACE);
 }
 
 \n+ {
   // Advance end by yyleng lines, resetting the column to zero.
   context.current_token_position.lines(yyleng);
   // Make the span empty by setting start to end.
   context.current_token_position.step();
+  BEGIN(AFTER_WHITESPACE);
 }
 
 . {

diff --git a/executable_semantics/syntax/parser.ypp b/executable_semantics/syntax/parser.ypp
@@ -123,7 +123,8 @@ void yy::parser::error(
 %token TYPE
 %token FN
 %token FNTY
-%token ARROW
+%token ARROW "->"
+%token FNARROW "-> in return type"
 %token VAR
 %token EQUAL_EQUAL
 %token IF
@@ -142,14 +143,20 @@ void yy::parser::error(
 %token CHOICE
 %token MATCH
 %token CASE
-%token DBLARROW
+%token DBLARROW "=>"
 %token DEFAULT
 %token AUTO
 %token
   EQUAL  "="
   MINUS  "-"
   PLUS   "+"
-  STAR   "*"
+  // The lexer determines the arity and fixity of each `*` based on whitespace
+  // and adjacent tokens. UNARY_STAR indicates that the operator is unary but
+  // could be either prefix or postfix.
+  UNARY_STAR "unary *"
+  PREFIX_STAR "prefix *"
+  POSTFIX_STAR "postfix *"
+  BINARY_STAR "binary *"
   SLASH  "/"
   LEFT_PARENTHESIS "("
   RIGHT_PARENTHESIS ")"
@@ -163,12 +170,22 @@ void yy::parser::error(
   COLON ":"
 ;
 
+%precedence FNARROW
 %precedence "{" "}"
 %precedence ":" "," DBLARROW
 %left OR AND
 %nonassoc EQUAL_EQUAL
 %left "+" "-"
-%precedence NOT UNARY_MINUS
+%left BINARY_STAR
+%precedence NOT UNARY_MINUS PREFIX_STAR
+// We need to give the `UNARY_STAR` token a precedence, rather than overriding
+// the precedence of the `expression UNARY_STAR` rule below, because bison
+// compares the precedence of the final token (for a shift) to the precedence
+// of the other rule (for a reduce) when attempting to resolve a shift-reduce
+// conflict. See https://stackoverflow.com/a/26188429/1041090. When UNARY_STAR
+// is the final token of a rule, it must be a postfix usage, so we give it the
+// same precedence as POSTFIX_STAR.
+%precedence POSTFIX_STAR UNARY_STAR
 %left "." ARROW
 %precedence "(" ")" "[" "]"
 
@@ -214,6 +231,8 @@ expression:
     { $$ = Carbon::Expression::MakeBinOp(yylineno, Carbon::Operator::Add, $1, $3); }
 | expression "-" expression
     { $$ = Carbon::Expression::MakeBinOp(yylineno, Carbon::Operator::Sub, $1, $3); }
+| expression BINARY_STAR expression
+    { $$ = Carbon::Expression::MakeBinOp(yylineno, Carbon::Operator::Mul, $1, $3); }
 | expression AND expression
     { $$ = Carbon::Expression::MakeBinOp(yylineno, Carbon::Operator::And, $1, $3); }
 | expression OR expression
@@ -222,8 +241,16 @@ expression:
     { $$ = Carbon::Expression::MakeUnOp(yylineno, Carbon::Operator::Not, $2); }
 | "-" expression %prec UNARY_MINUS
     { $$ = Carbon::Expression::MakeUnOp(yylineno, Carbon::Operator::Neg, $2); }
+| PREFIX_STAR expression
+    { $$ = Carbon::Expression::MakeUnOp(yylineno, Carbon::Operator::Deref, $2); }
+| UNARY_STAR expression %prec PREFIX_STAR
+    { $$ = Carbon::Expression::MakeUnOp(yylineno, Carbon::Operator::Deref, $2); }
 | expression tuple
     { $$ = Carbon::Expression::MakeCall(yylineno, $1, $2); }
+| expression POSTFIX_STAR
+    { $$ = Carbon::Expression::MakeUnOp(yylineno, Carbon::Operator::Ptr, $1); }
+| expression UNARY_STAR
+    { $$ = Carbon::Expression::MakeUnOp(yylineno, Carbon::Operator::Ptr, $1); }
 | FNTY tuple return_type
     { $$ = Carbon::Expression::MakeFunType(yylineno, $2, $3); }
 ;
@@ -324,7 +351,7 @@ statement_list:
 return_type:
   // Empty
     { $$ = Carbon::Expression::MakeUnit(yylineno); }
-| ARROW expression
+| ARROW expression %prec FNARROW
     { $$ = $2; }
 ;
 function_definition:

diff --git a/executable_semantics/testdata/star.carbon b/executable_semantics/testdata/star.carbon
@@ -0,0 +1,16 @@
+// Part of the Carbon Language project, under the Apache License v2.0 with LLVM
+// Exceptions. See /LICENSE for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+
+// Never actually called, so this only tests typechecking and semantic analysis.
+fn F(n: Int, p: Int*, q: Int***) -> Int* {
+  var a: Int = n * *p;
+  var b: Int = a*n;
+  *p = b*(*p);
+  **q = p;
+  return **q;
+}
+
+fn main() -> Int {
+  return 0;
+}
diff --git a/executable_semantics/testdata/star.golden b/executable_semantics/testdata/star.golden
@@ -0,0 +1 @@
+result: 0
diff --git a/toolchain/parser/parse_tree_test.cpp b/toolchain/parser/parse_tree_test.cpp
@@ -537,6 +537,7 @@ TEST_F(ParseTreeTest, OperatorFixity) {
       "  var t: Type = Int*;\n"
       "  t = t**;\n"
       "  n = n * n;\n"
+      "  n = n * *p;\n"
       "  n = n*n;\n"
       "  G(Int*, n * n);\n"
       "}");
@@ -577,6 +578,11 @@ TEST_F(ParseTreeTest, OperatorFixity) {
                        MatchNameReference("n"), "=",
                        MatchInfixOperator(MatchNameReference("n"), "*",
                                           MatchNameReference("n")))),
+                   MatchExpressionStatement(MatchInfixOperator(
+                       MatchNameReference("n"), "=",
+                       MatchInfixOperator(
+                           MatchNameReference("n"), "*",
+                           MatchPrefixOperator("*", MatchNameReference("p"))))),
                    MatchExpressionStatement(MatchInfixOperator(
                        MatchNameReference("n"), "=",
                        MatchInfixOperator(MatchNameReference("n"), "*",
-Original file line number
+Diff line change
@@ Expand Up / @@ -42,11 +42,14 @@ enum class ExpressionKind { @@
     enum class Operator {
       Add,
       And,
+      Deref,
       Eq,
+      Mul,
       Neg,
       Not,
       Or,
       Sub,
+      Ptr,
     };
     struct Expression;
@@ Expand Down @@