Sean Silva | ee47edf | 2012-12-05 00:26:32 +0000 | [diff] [blame^] | 1 | =========================================== |
| 2 | Kaleidoscope: Implementing a Parser and AST |
| 3 | =========================================== |
| 4 | |
| 5 | .. contents:: |
| 6 | :local: |
| 7 | |
| 8 | Written by `Chris Lattner <mailto:sabre@nondot.org>`_ |
| 9 | |
| 10 | Chapter 2 Introduction |
| 11 | ====================== |
| 12 | |
| 13 | Welcome to Chapter 2 of the "`Implementing a language with |
| 14 | LLVM <index.html>`_" tutorial. This chapter shows you how to use the |
| 15 | lexer, built in `Chapter 1 <LangImpl1.html>`_, to build a full |
| 16 | `parser <http://en.wikipedia.org/wiki/Parsing>`_ for our Kaleidoscope |
| 17 | language. Once we have a parser, we'll define and build an `Abstract |
| 18 | Syntax Tree <http://en.wikipedia.org/wiki/Abstract_syntax_tree>`_ (AST). |
| 19 | |
| 20 | The parser we will build uses a combination of `Recursive Descent |
| 21 | Parsing <http://en.wikipedia.org/wiki/Recursive_descent_parser>`_ and |
| 22 | `Operator-Precedence |
| 23 | Parsing <http://en.wikipedia.org/wiki/Operator-precedence_parser>`_ to |
| 24 | parse the Kaleidoscope language (the latter for binary expressions and |
| 25 | the former for everything else). Before we get to parsing though, lets |
| 26 | talk about the output of the parser: the Abstract Syntax Tree. |
| 27 | |
| 28 | The Abstract Syntax Tree (AST) |
| 29 | ============================== |
| 30 | |
| 31 | The AST for a program captures its behavior in such a way that it is |
| 32 | easy for later stages of the compiler (e.g. code generation) to |
| 33 | interpret. We basically want one object for each construct in the |
| 34 | language, and the AST should closely model the language. In |
| 35 | Kaleidoscope, we have expressions, a prototype, and a function object. |
| 36 | We'll start with expressions first: |
| 37 | |
| 38 | .. code-block:: c++ |
| 39 | |
| 40 | /// ExprAST - Base class for all expression nodes. |
| 41 | class ExprAST { |
| 42 | public: |
| 43 | virtual ~ExprAST() {} |
| 44 | }; |
| 45 | |
| 46 | /// NumberExprAST - Expression class for numeric literals like "1.0". |
| 47 | class NumberExprAST : public ExprAST { |
| 48 | double Val; |
| 49 | public: |
| 50 | NumberExprAST(double val) : Val(val) {} |
| 51 | }; |
| 52 | |
| 53 | The code above shows the definition of the base ExprAST class and one |
| 54 | subclass which we use for numeric literals. The important thing to note |
| 55 | about this code is that the NumberExprAST class captures the numeric |
| 56 | value of the literal as an instance variable. This allows later phases |
| 57 | of the compiler to know what the stored numeric value is. |
| 58 | |
| 59 | Right now we only create the AST, so there are no useful accessor |
| 60 | methods on them. It would be very easy to add a virtual method to pretty |
| 61 | print the code, for example. Here are the other expression AST node |
| 62 | definitions that we'll use in the basic form of the Kaleidoscope |
| 63 | language: |
| 64 | |
| 65 | .. code-block:: c++ |
| 66 | |
| 67 | /// VariableExprAST - Expression class for referencing a variable, like "a". |
| 68 | class VariableExprAST : public ExprAST { |
| 69 | std::string Name; |
| 70 | public: |
| 71 | VariableExprAST(const std::string &name) : Name(name) {} |
| 72 | }; |
| 73 | |
| 74 | /// BinaryExprAST - Expression class for a binary operator. |
| 75 | class BinaryExprAST : public ExprAST { |
| 76 | char Op; |
| 77 | ExprAST *LHS, *RHS; |
| 78 | public: |
| 79 | BinaryExprAST(char op, ExprAST *lhs, ExprAST *rhs) |
| 80 | : Op(op), LHS(lhs), RHS(rhs) {} |
| 81 | }; |
| 82 | |
| 83 | /// CallExprAST - Expression class for function calls. |
| 84 | class CallExprAST : public ExprAST { |
| 85 | std::string Callee; |
| 86 | std::vector<ExprAST*> Args; |
| 87 | public: |
| 88 | CallExprAST(const std::string &callee, std::vector<ExprAST*> &args) |
| 89 | : Callee(callee), Args(args) {} |
| 90 | }; |
| 91 | |
| 92 | This is all (intentionally) rather straight-forward: variables capture |
| 93 | the variable name, binary operators capture their opcode (e.g. '+'), and |
| 94 | calls capture a function name as well as a list of any argument |
| 95 | expressions. One thing that is nice about our AST is that it captures |
| 96 | the language features without talking about the syntax of the language. |
| 97 | Note that there is no discussion about precedence of binary operators, |
| 98 | lexical structure, etc. |
| 99 | |
| 100 | For our basic language, these are all of the expression nodes we'll |
| 101 | define. Because it doesn't have conditional control flow, it isn't |
| 102 | Turing-complete; we'll fix that in a later installment. The two things |
| 103 | we need next are a way to talk about the interface to a function, and a |
| 104 | way to talk about functions themselves: |
| 105 | |
| 106 | .. code-block:: c++ |
| 107 | |
| 108 | /// PrototypeAST - This class represents the "prototype" for a function, |
| 109 | /// which captures its name, and its argument names (thus implicitly the number |
| 110 | /// of arguments the function takes). |
| 111 | class PrototypeAST { |
| 112 | std::string Name; |
| 113 | std::vector<std::string> Args; |
| 114 | public: |
| 115 | PrototypeAST(const std::string &name, const std::vector<std::string> &args) |
| 116 | : Name(name), Args(args) {} |
| 117 | }; |
| 118 | |
| 119 | /// FunctionAST - This class represents a function definition itself. |
| 120 | class FunctionAST { |
| 121 | PrototypeAST *Proto; |
| 122 | ExprAST *Body; |
| 123 | public: |
| 124 | FunctionAST(PrototypeAST *proto, ExprAST *body) |
| 125 | : Proto(proto), Body(body) {} |
| 126 | }; |
| 127 | |
| 128 | In Kaleidoscope, functions are typed with just a count of their |
| 129 | arguments. Since all values are double precision floating point, the |
| 130 | type of each argument doesn't need to be stored anywhere. In a more |
| 131 | aggressive and realistic language, the "ExprAST" class would probably |
| 132 | have a type field. |
| 133 | |
| 134 | With this scaffolding, we can now talk about parsing expressions and |
| 135 | function bodies in Kaleidoscope. |
| 136 | |
| 137 | Parser Basics |
| 138 | ============= |
| 139 | |
| 140 | Now that we have an AST to build, we need to define the parser code to |
| 141 | build it. The idea here is that we want to parse something like "x+y" |
| 142 | (which is returned as three tokens by the lexer) into an AST that could |
| 143 | be generated with calls like this: |
| 144 | |
| 145 | .. code-block:: c++ |
| 146 | |
| 147 | ExprAST *X = new VariableExprAST("x"); |
| 148 | ExprAST *Y = new VariableExprAST("y"); |
| 149 | ExprAST *Result = new BinaryExprAST('+', X, Y); |
| 150 | |
| 151 | In order to do this, we'll start by defining some basic helper routines: |
| 152 | |
| 153 | .. code-block:: c++ |
| 154 | |
| 155 | /// CurTok/getNextToken - Provide a simple token buffer. CurTok is the current |
| 156 | /// token the parser is looking at. getNextToken reads another token from the |
| 157 | /// lexer and updates CurTok with its results. |
| 158 | static int CurTok; |
| 159 | static int getNextToken() { |
| 160 | return CurTok = gettok(); |
| 161 | } |
| 162 | |
| 163 | This implements a simple token buffer around the lexer. This allows us |
| 164 | to look one token ahead at what the lexer is returning. Every function |
| 165 | in our parser will assume that CurTok is the current token that needs to |
| 166 | be parsed. |
| 167 | |
| 168 | .. code-block:: c++ |
| 169 | |
| 170 | |
| 171 | /// Error* - These are little helper functions for error handling. |
| 172 | ExprAST *Error(const char *Str) { fprintf(stderr, "Error: %s\n", Str);return 0;} |
| 173 | PrototypeAST *ErrorP(const char *Str) { Error(Str); return 0; } |
| 174 | FunctionAST *ErrorF(const char *Str) { Error(Str); return 0; } |
| 175 | |
| 176 | The ``Error`` routines are simple helper routines that our parser will |
| 177 | use to handle errors. The error recovery in our parser will not be the |
| 178 | best and is not particular user-friendly, but it will be enough for our |
| 179 | tutorial. These routines make it easier to handle errors in routines |
| 180 | that have various return types: they always return null. |
| 181 | |
| 182 | With these basic helper functions, we can implement the first piece of |
| 183 | our grammar: numeric literals. |
| 184 | |
| 185 | Basic Expression Parsing |
| 186 | ======================== |
| 187 | |
| 188 | We start with numeric literals, because they are the simplest to |
| 189 | process. For each production in our grammar, we'll define a function |
| 190 | which parses that production. For numeric literals, we have: |
| 191 | |
| 192 | .. code-block:: c++ |
| 193 | |
| 194 | /// numberexpr ::= number |
| 195 | static ExprAST *ParseNumberExpr() { |
| 196 | ExprAST *Result = new NumberExprAST(NumVal); |
| 197 | getNextToken(); // consume the number |
| 198 | return Result; |
| 199 | } |
| 200 | |
| 201 | This routine is very simple: it expects to be called when the current |
| 202 | token is a ``tok_number`` token. It takes the current number value, |
| 203 | creates a ``NumberExprAST`` node, advances the lexer to the next token, |
| 204 | and finally returns. |
| 205 | |
| 206 | There are some interesting aspects to this. The most important one is |
| 207 | that this routine eats all of the tokens that correspond to the |
| 208 | production and returns the lexer buffer with the next token (which is |
| 209 | not part of the grammar production) ready to go. This is a fairly |
| 210 | standard way to go for recursive descent parsers. For a better example, |
| 211 | the parenthesis operator is defined like this: |
| 212 | |
| 213 | .. code-block:: c++ |
| 214 | |
| 215 | /// parenexpr ::= '(' expression ')' |
| 216 | static ExprAST *ParseParenExpr() { |
| 217 | getNextToken(); // eat (. |
| 218 | ExprAST *V = ParseExpression(); |
| 219 | if (!V) return 0; |
| 220 | |
| 221 | if (CurTok != ')') |
| 222 | return Error("expected ')'"); |
| 223 | getNextToken(); // eat ). |
| 224 | return V; |
| 225 | } |
| 226 | |
| 227 | This function illustrates a number of interesting things about the |
| 228 | parser: |
| 229 | |
| 230 | 1) It shows how we use the Error routines. When called, this function |
| 231 | expects that the current token is a '(' token, but after parsing the |
| 232 | subexpression, it is possible that there is no ')' waiting. For example, |
| 233 | if the user types in "(4 x" instead of "(4)", the parser should emit an |
| 234 | error. Because errors can occur, the parser needs a way to indicate that |
| 235 | they happened: in our parser, we return null on an error. |
| 236 | |
| 237 | 2) Another interesting aspect of this function is that it uses recursion |
| 238 | by calling ``ParseExpression`` (we will soon see that |
| 239 | ``ParseExpression`` can call ``ParseParenExpr``). This is powerful |
| 240 | because it allows us to handle recursive grammars, and keeps each |
| 241 | production very simple. Note that parentheses do not cause construction |
| 242 | of AST nodes themselves. While we could do it this way, the most |
| 243 | important role of parentheses are to guide the parser and provide |
| 244 | grouping. Once the parser constructs the AST, parentheses are not |
| 245 | needed. |
| 246 | |
| 247 | The next simple production is for handling variable references and |
| 248 | function calls: |
| 249 | |
| 250 | .. code-block:: c++ |
| 251 | |
| 252 | /// identifierexpr |
| 253 | /// ::= identifier |
| 254 | /// ::= identifier '(' expression* ')' |
| 255 | static ExprAST *ParseIdentifierExpr() { |
| 256 | std::string IdName = IdentifierStr; |
| 257 | |
| 258 | getNextToken(); // eat identifier. |
| 259 | |
| 260 | if (CurTok != '(') // Simple variable ref. |
| 261 | return new VariableExprAST(IdName); |
| 262 | |
| 263 | // Call. |
| 264 | getNextToken(); // eat ( |
| 265 | std::vector<ExprAST*> Args; |
| 266 | if (CurTok != ')') { |
| 267 | while (1) { |
| 268 | ExprAST *Arg = ParseExpression(); |
| 269 | if (!Arg) return 0; |
| 270 | Args.push_back(Arg); |
| 271 | |
| 272 | if (CurTok == ')') break; |
| 273 | |
| 274 | if (CurTok != ',') |
| 275 | return Error("Expected ')' or ',' in argument list"); |
| 276 | getNextToken(); |
| 277 | } |
| 278 | } |
| 279 | |
| 280 | // Eat the ')'. |
| 281 | getNextToken(); |
| 282 | |
| 283 | return new CallExprAST(IdName, Args); |
| 284 | } |
| 285 | |
| 286 | This routine follows the same style as the other routines. (It expects |
| 287 | to be called if the current token is a ``tok_identifier`` token). It |
| 288 | also has recursion and error handling. One interesting aspect of this is |
| 289 | that it uses *look-ahead* to determine if the current identifier is a |
| 290 | stand alone variable reference or if it is a function call expression. |
| 291 | It handles this by checking to see if the token after the identifier is |
| 292 | a '(' token, constructing either a ``VariableExprAST`` or |
| 293 | ``CallExprAST`` node as appropriate. |
| 294 | |
| 295 | Now that we have all of our simple expression-parsing logic in place, we |
| 296 | can define a helper function to wrap it together into one entry point. |
| 297 | We call this class of expressions "primary" expressions, for reasons |
| 298 | that will become more clear `later in the |
| 299 | tutorial <LangImpl6.html#unary>`_. In order to parse an arbitrary |
| 300 | primary expression, we need to determine what sort of expression it is: |
| 301 | |
| 302 | .. code-block:: c++ |
| 303 | |
| 304 | /// primary |
| 305 | /// ::= identifierexpr |
| 306 | /// ::= numberexpr |
| 307 | /// ::= parenexpr |
| 308 | static ExprAST *ParsePrimary() { |
| 309 | switch (CurTok) { |
| 310 | default: return Error("unknown token when expecting an expression"); |
| 311 | case tok_identifier: return ParseIdentifierExpr(); |
| 312 | case tok_number: return ParseNumberExpr(); |
| 313 | case '(': return ParseParenExpr(); |
| 314 | } |
| 315 | } |
| 316 | |
| 317 | Now that you see the definition of this function, it is more obvious why |
| 318 | we can assume the state of CurTok in the various functions. This uses |
| 319 | look-ahead to determine which sort of expression is being inspected, and |
| 320 | then parses it with a function call. |
| 321 | |
| 322 | Now that basic expressions are handled, we need to handle binary |
| 323 | expressions. They are a bit more complex. |
| 324 | |
| 325 | Binary Expression Parsing |
| 326 | ========================= |
| 327 | |
| 328 | Binary expressions are significantly harder to parse because they are |
| 329 | often ambiguous. For example, when given the string "x+y\*z", the parser |
| 330 | can choose to parse it as either "(x+y)\*z" or "x+(y\*z)". With common |
| 331 | definitions from mathematics, we expect the later parse, because "\*" |
| 332 | (multiplication) has higher *precedence* than "+" (addition). |
| 333 | |
| 334 | There are many ways to handle this, but an elegant and efficient way is |
| 335 | to use `Operator-Precedence |
| 336 | Parsing <http://en.wikipedia.org/wiki/Operator-precedence_parser>`_. |
| 337 | This parsing technique uses the precedence of binary operators to guide |
| 338 | recursion. To start with, we need a table of precedences: |
| 339 | |
| 340 | .. code-block:: c++ |
| 341 | |
| 342 | /// BinopPrecedence - This holds the precedence for each binary operator that is |
| 343 | /// defined. |
| 344 | static std::map<char, int> BinopPrecedence; |
| 345 | |
| 346 | /// GetTokPrecedence - Get the precedence of the pending binary operator token. |
| 347 | static int GetTokPrecedence() { |
| 348 | if (!isascii(CurTok)) |
| 349 | return -1; |
| 350 | |
| 351 | // Make sure it's a declared binop. |
| 352 | int TokPrec = BinopPrecedence[CurTok]; |
| 353 | if (TokPrec <= 0) return -1; |
| 354 | return TokPrec; |
| 355 | } |
| 356 | |
| 357 | int main() { |
| 358 | // Install standard binary operators. |
| 359 | // 1 is lowest precedence. |
| 360 | BinopPrecedence['<'] = 10; |
| 361 | BinopPrecedence['+'] = 20; |
| 362 | BinopPrecedence['-'] = 20; |
| 363 | BinopPrecedence['*'] = 40; // highest. |
| 364 | ... |
| 365 | } |
| 366 | |
| 367 | For the basic form of Kaleidoscope, we will only support 4 binary |
| 368 | operators (this can obviously be extended by you, our brave and intrepid |
| 369 | reader). The ``GetTokPrecedence`` function returns the precedence for |
| 370 | the current token, or -1 if the token is not a binary operator. Having a |
| 371 | map makes it easy to add new operators and makes it clear that the |
| 372 | algorithm doesn't depend on the specific operators involved, but it |
| 373 | would be easy enough to eliminate the map and do the comparisons in the |
| 374 | ``GetTokPrecedence`` function. (Or just use a fixed-size array). |
| 375 | |
| 376 | With the helper above defined, we can now start parsing binary |
| 377 | expressions. The basic idea of operator precedence parsing is to break |
| 378 | down an expression with potentially ambiguous binary operators into |
| 379 | pieces. Consider ,for example, the expression "a+b+(c+d)\*e\*f+g". |
| 380 | Operator precedence parsing considers this as a stream of primary |
| 381 | expressions separated by binary operators. As such, it will first parse |
| 382 | the leading primary expression "a", then it will see the pairs [+, b] |
| 383 | [+, (c+d)] [\*, e] [\*, f] and [+, g]. Note that because parentheses are |
| 384 | primary expressions, the binary expression parser doesn't need to worry |
| 385 | about nested subexpressions like (c+d) at all. |
| 386 | |
| 387 | To start, an expression is a primary expression potentially followed by |
| 388 | a sequence of [binop,primaryexpr] pairs: |
| 389 | |
| 390 | .. code-block:: c++ |
| 391 | |
| 392 | /// expression |
| 393 | /// ::= primary binoprhs |
| 394 | /// |
| 395 | static ExprAST *ParseExpression() { |
| 396 | ExprAST *LHS = ParsePrimary(); |
| 397 | if (!LHS) return 0; |
| 398 | |
| 399 | return ParseBinOpRHS(0, LHS); |
| 400 | } |
| 401 | |
| 402 | ``ParseBinOpRHS`` is the function that parses the sequence of pairs for |
| 403 | us. It takes a precedence and a pointer to an expression for the part |
| 404 | that has been parsed so far. Note that "x" is a perfectly valid |
| 405 | expression: As such, "binoprhs" is allowed to be empty, in which case it |
| 406 | returns the expression that is passed into it. In our example above, the |
| 407 | code passes the expression for "a" into ``ParseBinOpRHS`` and the |
| 408 | current token is "+". |
| 409 | |
| 410 | The precedence value passed into ``ParseBinOpRHS`` indicates the |
| 411 | *minimal operator precedence* that the function is allowed to eat. For |
| 412 | example, if the current pair stream is [+, x] and ``ParseBinOpRHS`` is |
| 413 | passed in a precedence of 40, it will not consume any tokens (because |
| 414 | the precedence of '+' is only 20). With this in mind, ``ParseBinOpRHS`` |
| 415 | starts with: |
| 416 | |
| 417 | .. code-block:: c++ |
| 418 | |
| 419 | /// binoprhs |
| 420 | /// ::= ('+' primary)* |
| 421 | static ExprAST *ParseBinOpRHS(int ExprPrec, ExprAST *LHS) { |
| 422 | // If this is a binop, find its precedence. |
| 423 | while (1) { |
| 424 | int TokPrec = GetTokPrecedence(); |
| 425 | |
| 426 | // If this is a binop that binds at least as tightly as the current binop, |
| 427 | // consume it, otherwise we are done. |
| 428 | if (TokPrec < ExprPrec) |
| 429 | return LHS; |
| 430 | |
| 431 | This code gets the precedence of the current token and checks to see if |
| 432 | if is too low. Because we defined invalid tokens to have a precedence of |
| 433 | -1, this check implicitly knows that the pair-stream ends when the token |
| 434 | stream runs out of binary operators. If this check succeeds, we know |
| 435 | that the token is a binary operator and that it will be included in this |
| 436 | expression: |
| 437 | |
| 438 | .. code-block:: c++ |
| 439 | |
| 440 | // Okay, we know this is a binop. |
| 441 | int BinOp = CurTok; |
| 442 | getNextToken(); // eat binop |
| 443 | |
| 444 | // Parse the primary expression after the binary operator. |
| 445 | ExprAST *RHS = ParsePrimary(); |
| 446 | if (!RHS) return 0; |
| 447 | |
| 448 | As such, this code eats (and remembers) the binary operator and then |
| 449 | parses the primary expression that follows. This builds up the whole |
| 450 | pair, the first of which is [+, b] for the running example. |
| 451 | |
| 452 | Now that we parsed the left-hand side of an expression and one pair of |
| 453 | the RHS sequence, we have to decide which way the expression associates. |
| 454 | In particular, we could have "(a+b) binop unparsed" or "a + (b binop |
| 455 | unparsed)". To determine this, we look ahead at "binop" to determine its |
| 456 | precedence and compare it to BinOp's precedence (which is '+' in this |
| 457 | case): |
| 458 | |
| 459 | .. code-block:: c++ |
| 460 | |
| 461 | // If BinOp binds less tightly with RHS than the operator after RHS, let |
| 462 | // the pending operator take RHS as its LHS. |
| 463 | int NextPrec = GetTokPrecedence(); |
| 464 | if (TokPrec < NextPrec) { |
| 465 | |
| 466 | If the precedence of the binop to the right of "RHS" is lower or equal |
| 467 | to the precedence of our current operator, then we know that the |
| 468 | parentheses associate as "(a+b) binop ...". In our example, the current |
| 469 | operator is "+" and the next operator is "+", we know that they have the |
| 470 | same precedence. In this case we'll create the AST node for "a+b", and |
| 471 | then continue parsing: |
| 472 | |
| 473 | .. code-block:: c++ |
| 474 | |
| 475 | ... if body omitted ... |
| 476 | } |
| 477 | |
| 478 | // Merge LHS/RHS. |
| 479 | LHS = new BinaryExprAST(BinOp, LHS, RHS); |
| 480 | } // loop around to the top of the while loop. |
| 481 | } |
| 482 | |
| 483 | In our example above, this will turn "a+b+" into "(a+b)" and execute the |
| 484 | next iteration of the loop, with "+" as the current token. The code |
| 485 | above will eat, remember, and parse "(c+d)" as the primary expression, |
| 486 | which makes the current pair equal to [+, (c+d)]. It will then evaluate |
| 487 | the 'if' conditional above with "\*" as the binop to the right of the |
| 488 | primary. In this case, the precedence of "\*" is higher than the |
| 489 | precedence of "+" so the if condition will be entered. |
| 490 | |
| 491 | The critical question left here is "how can the if condition parse the |
| 492 | right hand side in full"? In particular, to build the AST correctly for |
| 493 | our example, it needs to get all of "(c+d)\*e\*f" as the RHS expression |
| 494 | variable. The code to do this is surprisingly simple (code from the |
| 495 | above two blocks duplicated for context): |
| 496 | |
| 497 | .. code-block:: c++ |
| 498 | |
| 499 | // If BinOp binds less tightly with RHS than the operator after RHS, let |
| 500 | // the pending operator take RHS as its LHS. |
| 501 | int NextPrec = GetTokPrecedence(); |
| 502 | if (TokPrec < NextPrec) { |
| 503 | RHS = ParseBinOpRHS(TokPrec+1, RHS); |
| 504 | if (RHS == 0) return 0; |
| 505 | } |
| 506 | // Merge LHS/RHS. |
| 507 | LHS = new BinaryExprAST(BinOp, LHS, RHS); |
| 508 | } // loop around to the top of the while loop. |
| 509 | } |
| 510 | |
| 511 | At this point, we know that the binary operator to the RHS of our |
| 512 | primary has higher precedence than the binop we are currently parsing. |
| 513 | As such, we know that any sequence of pairs whose operators are all |
| 514 | higher precedence than "+" should be parsed together and returned as |
| 515 | "RHS". To do this, we recursively invoke the ``ParseBinOpRHS`` function |
| 516 | specifying "TokPrec+1" as the minimum precedence required for it to |
| 517 | continue. In our example above, this will cause it to return the AST |
| 518 | node for "(c+d)\*e\*f" as RHS, which is then set as the RHS of the '+' |
| 519 | expression. |
| 520 | |
| 521 | Finally, on the next iteration of the while loop, the "+g" piece is |
| 522 | parsed and added to the AST. With this little bit of code (14 |
| 523 | non-trivial lines), we correctly handle fully general binary expression |
| 524 | parsing in a very elegant way. This was a whirlwind tour of this code, |
| 525 | and it is somewhat subtle. I recommend running through it with a few |
| 526 | tough examples to see how it works. |
| 527 | |
| 528 | This wraps up handling of expressions. At this point, we can point the |
| 529 | parser at an arbitrary token stream and build an expression from it, |
| 530 | stopping at the first token that is not part of the expression. Next up |
| 531 | we need to handle function definitions, etc. |
| 532 | |
| 533 | Parsing the Rest |
| 534 | ================ |
| 535 | |
| 536 | The next thing missing is handling of function prototypes. In |
| 537 | Kaleidoscope, these are used both for 'extern' function declarations as |
| 538 | well as function body definitions. The code to do this is |
| 539 | straight-forward and not very interesting (once you've survived |
| 540 | expressions): |
| 541 | |
| 542 | .. code-block:: c++ |
| 543 | |
| 544 | /// prototype |
| 545 | /// ::= id '(' id* ')' |
| 546 | static PrototypeAST *ParsePrototype() { |
| 547 | if (CurTok != tok_identifier) |
| 548 | return ErrorP("Expected function name in prototype"); |
| 549 | |
| 550 | std::string FnName = IdentifierStr; |
| 551 | getNextToken(); |
| 552 | |
| 553 | if (CurTok != '(') |
| 554 | return ErrorP("Expected '(' in prototype"); |
| 555 | |
| 556 | // Read the list of argument names. |
| 557 | std::vector<std::string> ArgNames; |
| 558 | while (getNextToken() == tok_identifier) |
| 559 | ArgNames.push_back(IdentifierStr); |
| 560 | if (CurTok != ')') |
| 561 | return ErrorP("Expected ')' in prototype"); |
| 562 | |
| 563 | // success. |
| 564 | getNextToken(); // eat ')'. |
| 565 | |
| 566 | return new PrototypeAST(FnName, ArgNames); |
| 567 | } |
| 568 | |
| 569 | Given this, a function definition is very simple, just a prototype plus |
| 570 | an expression to implement the body: |
| 571 | |
| 572 | .. code-block:: c++ |
| 573 | |
| 574 | /// definition ::= 'def' prototype expression |
| 575 | static FunctionAST *ParseDefinition() { |
| 576 | getNextToken(); // eat def. |
| 577 | PrototypeAST *Proto = ParsePrototype(); |
| 578 | if (Proto == 0) return 0; |
| 579 | |
| 580 | if (ExprAST *E = ParseExpression()) |
| 581 | return new FunctionAST(Proto, E); |
| 582 | return 0; |
| 583 | } |
| 584 | |
| 585 | In addition, we support 'extern' to declare functions like 'sin' and |
| 586 | 'cos' as well as to support forward declaration of user functions. These |
| 587 | 'extern's are just prototypes with no body: |
| 588 | |
| 589 | .. code-block:: c++ |
| 590 | |
| 591 | /// external ::= 'extern' prototype |
| 592 | static PrototypeAST *ParseExtern() { |
| 593 | getNextToken(); // eat extern. |
| 594 | return ParsePrototype(); |
| 595 | } |
| 596 | |
| 597 | Finally, we'll also let the user type in arbitrary top-level expressions |
| 598 | and evaluate them on the fly. We will handle this by defining anonymous |
| 599 | nullary (zero argument) functions for them: |
| 600 | |
| 601 | .. code-block:: c++ |
| 602 | |
| 603 | /// toplevelexpr ::= expression |
| 604 | static FunctionAST *ParseTopLevelExpr() { |
| 605 | if (ExprAST *E = ParseExpression()) { |
| 606 | // Make an anonymous proto. |
| 607 | PrototypeAST *Proto = new PrototypeAST("", std::vector<std::string>()); |
| 608 | return new FunctionAST(Proto, E); |
| 609 | } |
| 610 | return 0; |
| 611 | } |
| 612 | |
| 613 | Now that we have all the pieces, let's build a little driver that will |
| 614 | let us actually *execute* this code we've built! |
| 615 | |
| 616 | The Driver |
| 617 | ========== |
| 618 | |
| 619 | The driver for this simply invokes all of the parsing pieces with a |
| 620 | top-level dispatch loop. There isn't much interesting here, so I'll just |
| 621 | include the top-level loop. See `below <#code>`_ for full code in the |
| 622 | "Top-Level Parsing" section. |
| 623 | |
| 624 | .. code-block:: c++ |
| 625 | |
| 626 | /// top ::= definition | external | expression | ';' |
| 627 | static void MainLoop() { |
| 628 | while (1) { |
| 629 | fprintf(stderr, "ready> "); |
| 630 | switch (CurTok) { |
| 631 | case tok_eof: return; |
| 632 | case ';': getNextToken(); break; // ignore top-level semicolons. |
| 633 | case tok_def: HandleDefinition(); break; |
| 634 | case tok_extern: HandleExtern(); break; |
| 635 | default: HandleTopLevelExpression(); break; |
| 636 | } |
| 637 | } |
| 638 | } |
| 639 | |
| 640 | The most interesting part of this is that we ignore top-level |
| 641 | semicolons. Why is this, you ask? The basic reason is that if you type |
| 642 | "4 + 5" at the command line, the parser doesn't know whether that is the |
| 643 | end of what you will type or not. For example, on the next line you |
| 644 | could type "def foo..." in which case 4+5 is the end of a top-level |
| 645 | expression. Alternatively you could type "\* 6", which would continue |
| 646 | the expression. Having top-level semicolons allows you to type "4+5;", |
| 647 | and the parser will know you are done. |
| 648 | |
| 649 | Conclusions |
| 650 | =========== |
| 651 | |
| 652 | With just under 400 lines of commented code (240 lines of non-comment, |
| 653 | non-blank code), we fully defined our minimal language, including a |
| 654 | lexer, parser, and AST builder. With this done, the executable will |
| 655 | validate Kaleidoscope code and tell us if it is grammatically invalid. |
| 656 | For example, here is a sample interaction: |
| 657 | |
| 658 | .. code-block:: bash |
| 659 | |
| 660 | $ ./a.out |
| 661 | ready> def foo(x y) x+foo(y, 4.0); |
| 662 | Parsed a function definition. |
| 663 | ready> def foo(x y) x+y y; |
| 664 | Parsed a function definition. |
| 665 | Parsed a top-level expr |
| 666 | ready> def foo(x y) x+y ); |
| 667 | Parsed a function definition. |
| 668 | Error: unknown token when expecting an expression |
| 669 | ready> extern sin(a); |
| 670 | ready> Parsed an extern |
| 671 | ready> ^D |
| 672 | $ |
| 673 | |
| 674 | There is a lot of room for extension here. You can define new AST nodes, |
| 675 | extend the language in many ways, etc. In the `next |
| 676 | installment <LangImpl3.html>`_, we will describe how to generate LLVM |
| 677 | Intermediate Representation (IR) from the AST. |
| 678 | |
| 679 | Full Code Listing |
| 680 | ================= |
| 681 | |
| 682 | Here is the complete code listing for this and the previous chapter. |
| 683 | Note that it is fully self-contained: you don't need LLVM or any |
| 684 | external libraries at all for this. (Besides the C and C++ standard |
| 685 | libraries, of course.) To build this, just compile with: |
| 686 | |
| 687 | .. code-block:: bash |
| 688 | |
| 689 | # Compile |
| 690 | clang++ -g -O3 toy.cpp |
| 691 | # Run |
| 692 | ./a.out |
| 693 | |
| 694 | Here is the code: |
| 695 | |
| 696 | .. code-block:: c++ |
| 697 | |
| 698 | #include <cstdio> |
| 699 | #include <cstdlib> |
| 700 | #include <string> |
| 701 | #include <map> |
| 702 | #include <vector> |
| 703 | |
| 704 | //===----------------------------------------------------------------------===// |
| 705 | // Lexer |
| 706 | //===----------------------------------------------------------------------===// |
| 707 | |
| 708 | // The lexer returns tokens [0-255] if it is an unknown character, otherwise one |
| 709 | // of these for known things. |
| 710 | enum Token { |
| 711 | tok_eof = -1, |
| 712 | |
| 713 | // commands |
| 714 | tok_def = -2, tok_extern = -3, |
| 715 | |
| 716 | // primary |
| 717 | tok_identifier = -4, tok_number = -5 |
| 718 | }; |
| 719 | |
| 720 | static std::string IdentifierStr; // Filled in if tok_identifier |
| 721 | static double NumVal; // Filled in if tok_number |
| 722 | |
| 723 | /// gettok - Return the next token from standard input. |
| 724 | static int gettok() { |
| 725 | static int LastChar = ' '; |
| 726 | |
| 727 | // Skip any whitespace. |
| 728 | while (isspace(LastChar)) |
| 729 | LastChar = getchar(); |
| 730 | |
| 731 | if (isalpha(LastChar)) { // identifier: [a-zA-Z][a-zA-Z0-9]* |
| 732 | IdentifierStr = LastChar; |
| 733 | while (isalnum((LastChar = getchar()))) |
| 734 | IdentifierStr += LastChar; |
| 735 | |
| 736 | if (IdentifierStr == "def") return tok_def; |
| 737 | if (IdentifierStr == "extern") return tok_extern; |
| 738 | return tok_identifier; |
| 739 | } |
| 740 | |
| 741 | if (isdigit(LastChar) || LastChar == '.') { // Number: [0-9.]+ |
| 742 | std::string NumStr; |
| 743 | do { |
| 744 | NumStr += LastChar; |
| 745 | LastChar = getchar(); |
| 746 | } while (isdigit(LastChar) || LastChar == '.'); |
| 747 | |
| 748 | NumVal = strtod(NumStr.c_str(), 0); |
| 749 | return tok_number; |
| 750 | } |
| 751 | |
| 752 | if (LastChar == '#') { |
| 753 | // Comment until end of line. |
| 754 | do LastChar = getchar(); |
| 755 | while (LastChar != EOF && LastChar != '\n' && LastChar != '\r'); |
| 756 | |
| 757 | if (LastChar != EOF) |
| 758 | return gettok(); |
| 759 | } |
| 760 | |
| 761 | // Check for end of file. Don't eat the EOF. |
| 762 | if (LastChar == EOF) |
| 763 | return tok_eof; |
| 764 | |
| 765 | // Otherwise, just return the character as its ascii value. |
| 766 | int ThisChar = LastChar; |
| 767 | LastChar = getchar(); |
| 768 | return ThisChar; |
| 769 | } |
| 770 | |
| 771 | //===----------------------------------------------------------------------===// |
| 772 | // Abstract Syntax Tree (aka Parse Tree) |
| 773 | //===----------------------------------------------------------------------===// |
| 774 | |
| 775 | /// ExprAST - Base class for all expression nodes. |
| 776 | class ExprAST { |
| 777 | public: |
| 778 | virtual ~ExprAST() {} |
| 779 | }; |
| 780 | |
| 781 | /// NumberExprAST - Expression class for numeric literals like "1.0". |
| 782 | class NumberExprAST : public ExprAST { |
| 783 | double Val; |
| 784 | public: |
| 785 | NumberExprAST(double val) : Val(val) {} |
| 786 | }; |
| 787 | |
| 788 | /// VariableExprAST - Expression class for referencing a variable, like "a". |
| 789 | class VariableExprAST : public ExprAST { |
| 790 | std::string Name; |
| 791 | public: |
| 792 | VariableExprAST(const std::string &name) : Name(name) {} |
| 793 | }; |
| 794 | |
| 795 | /// BinaryExprAST - Expression class for a binary operator. |
| 796 | class BinaryExprAST : public ExprAST { |
| 797 | char Op; |
| 798 | ExprAST *LHS, *RHS; |
| 799 | public: |
| 800 | BinaryExprAST(char op, ExprAST *lhs, ExprAST *rhs) |
| 801 | : Op(op), LHS(lhs), RHS(rhs) {} |
| 802 | }; |
| 803 | |
| 804 | /// CallExprAST - Expression class for function calls. |
| 805 | class CallExprAST : public ExprAST { |
| 806 | std::string Callee; |
| 807 | std::vector<ExprAST*> Args; |
| 808 | public: |
| 809 | CallExprAST(const std::string &callee, std::vector<ExprAST*> &args) |
| 810 | : Callee(callee), Args(args) {} |
| 811 | }; |
| 812 | |
| 813 | /// PrototypeAST - This class represents the "prototype" for a function, |
| 814 | /// which captures its name, and its argument names (thus implicitly the number |
| 815 | /// of arguments the function takes). |
| 816 | class PrototypeAST { |
| 817 | std::string Name; |
| 818 | std::vector<std::string> Args; |
| 819 | public: |
| 820 | PrototypeAST(const std::string &name, const std::vector<std::string> &args) |
| 821 | : Name(name), Args(args) {} |
| 822 | |
| 823 | }; |
| 824 | |
| 825 | /// FunctionAST - This class represents a function definition itself. |
| 826 | class FunctionAST { |
| 827 | PrototypeAST *Proto; |
| 828 | ExprAST *Body; |
| 829 | public: |
| 830 | FunctionAST(PrototypeAST *proto, ExprAST *body) |
| 831 | : Proto(proto), Body(body) {} |
| 832 | |
| 833 | }; |
| 834 | |
| 835 | //===----------------------------------------------------------------------===// |
| 836 | // Parser |
| 837 | //===----------------------------------------------------------------------===// |
| 838 | |
| 839 | /// CurTok/getNextToken - Provide a simple token buffer. CurTok is the current |
| 840 | /// token the parser is looking at. getNextToken reads another token from the |
| 841 | /// lexer and updates CurTok with its results. |
| 842 | static int CurTok; |
| 843 | static int getNextToken() { |
| 844 | return CurTok = gettok(); |
| 845 | } |
| 846 | |
| 847 | /// BinopPrecedence - This holds the precedence for each binary operator that is |
| 848 | /// defined. |
| 849 | static std::map<char, int> BinopPrecedence; |
| 850 | |
| 851 | /// GetTokPrecedence - Get the precedence of the pending binary operator token. |
| 852 | static int GetTokPrecedence() { |
| 853 | if (!isascii(CurTok)) |
| 854 | return -1; |
| 855 | |
| 856 | // Make sure it's a declared binop. |
| 857 | int TokPrec = BinopPrecedence[CurTok]; |
| 858 | if (TokPrec <= 0) return -1; |
| 859 | return TokPrec; |
| 860 | } |
| 861 | |
| 862 | /// Error* - These are little helper functions for error handling. |
| 863 | ExprAST *Error(const char *Str) { fprintf(stderr, "Error: %s\n", Str);return 0;} |
| 864 | PrototypeAST *ErrorP(const char *Str) { Error(Str); return 0; } |
| 865 | FunctionAST *ErrorF(const char *Str) { Error(Str); return 0; } |
| 866 | |
| 867 | static ExprAST *ParseExpression(); |
| 868 | |
| 869 | /// identifierexpr |
| 870 | /// ::= identifier |
| 871 | /// ::= identifier '(' expression* ')' |
| 872 | static ExprAST *ParseIdentifierExpr() { |
| 873 | std::string IdName = IdentifierStr; |
| 874 | |
| 875 | getNextToken(); // eat identifier. |
| 876 | |
| 877 | if (CurTok != '(') // Simple variable ref. |
| 878 | return new VariableExprAST(IdName); |
| 879 | |
| 880 | // Call. |
| 881 | getNextToken(); // eat ( |
| 882 | std::vector<ExprAST*> Args; |
| 883 | if (CurTok != ')') { |
| 884 | while (1) { |
| 885 | ExprAST *Arg = ParseExpression(); |
| 886 | if (!Arg) return 0; |
| 887 | Args.push_back(Arg); |
| 888 | |
| 889 | if (CurTok == ')') break; |
| 890 | |
| 891 | if (CurTok != ',') |
| 892 | return Error("Expected ')' or ',' in argument list"); |
| 893 | getNextToken(); |
| 894 | } |
| 895 | } |
| 896 | |
| 897 | // Eat the ')'. |
| 898 | getNextToken(); |
| 899 | |
| 900 | return new CallExprAST(IdName, Args); |
| 901 | } |
| 902 | |
| 903 | /// numberexpr ::= number |
| 904 | static ExprAST *ParseNumberExpr() { |
| 905 | ExprAST *Result = new NumberExprAST(NumVal); |
| 906 | getNextToken(); // consume the number |
| 907 | return Result; |
| 908 | } |
| 909 | |
| 910 | /// parenexpr ::= '(' expression ')' |
| 911 | static ExprAST *ParseParenExpr() { |
| 912 | getNextToken(); // eat (. |
| 913 | ExprAST *V = ParseExpression(); |
| 914 | if (!V) return 0; |
| 915 | |
| 916 | if (CurTok != ')') |
| 917 | return Error("expected ')'"); |
| 918 | getNextToken(); // eat ). |
| 919 | return V; |
| 920 | } |
| 921 | |
| 922 | /// primary |
| 923 | /// ::= identifierexpr |
| 924 | /// ::= numberexpr |
| 925 | /// ::= parenexpr |
| 926 | static ExprAST *ParsePrimary() { |
| 927 | switch (CurTok) { |
| 928 | default: return Error("unknown token when expecting an expression"); |
| 929 | case tok_identifier: return ParseIdentifierExpr(); |
| 930 | case tok_number: return ParseNumberExpr(); |
| 931 | case '(': return ParseParenExpr(); |
| 932 | } |
| 933 | } |
| 934 | |
| 935 | /// binoprhs |
| 936 | /// ::= ('+' primary)* |
| 937 | static ExprAST *ParseBinOpRHS(int ExprPrec, ExprAST *LHS) { |
| 938 | // If this is a binop, find its precedence. |
| 939 | while (1) { |
| 940 | int TokPrec = GetTokPrecedence(); |
| 941 | |
| 942 | // If this is a binop that binds at least as tightly as the current binop, |
| 943 | // consume it, otherwise we are done. |
| 944 | if (TokPrec < ExprPrec) |
| 945 | return LHS; |
| 946 | |
| 947 | // Okay, we know this is a binop. |
| 948 | int BinOp = CurTok; |
| 949 | getNextToken(); // eat binop |
| 950 | |
| 951 | // Parse the primary expression after the binary operator. |
| 952 | ExprAST *RHS = ParsePrimary(); |
| 953 | if (!RHS) return 0; |
| 954 | |
| 955 | // If BinOp binds less tightly with RHS than the operator after RHS, let |
| 956 | // the pending operator take RHS as its LHS. |
| 957 | int NextPrec = GetTokPrecedence(); |
| 958 | if (TokPrec < NextPrec) { |
| 959 | RHS = ParseBinOpRHS(TokPrec+1, RHS); |
| 960 | if (RHS == 0) return 0; |
| 961 | } |
| 962 | |
| 963 | // Merge LHS/RHS. |
| 964 | LHS = new BinaryExprAST(BinOp, LHS, RHS); |
| 965 | } |
| 966 | } |
| 967 | |
| 968 | /// expression |
| 969 | /// ::= primary binoprhs |
| 970 | /// |
| 971 | static ExprAST *ParseExpression() { |
| 972 | ExprAST *LHS = ParsePrimary(); |
| 973 | if (!LHS) return 0; |
| 974 | |
| 975 | return ParseBinOpRHS(0, LHS); |
| 976 | } |
| 977 | |
| 978 | /// prototype |
| 979 | /// ::= id '(' id* ')' |
| 980 | static PrototypeAST *ParsePrototype() { |
| 981 | if (CurTok != tok_identifier) |
| 982 | return ErrorP("Expected function name in prototype"); |
| 983 | |
| 984 | std::string FnName = IdentifierStr; |
| 985 | getNextToken(); |
| 986 | |
| 987 | if (CurTok != '(') |
| 988 | return ErrorP("Expected '(' in prototype"); |
| 989 | |
| 990 | std::vector<std::string> ArgNames; |
| 991 | while (getNextToken() == tok_identifier) |
| 992 | ArgNames.push_back(IdentifierStr); |
| 993 | if (CurTok != ')') |
| 994 | return ErrorP("Expected ')' in prototype"); |
| 995 | |
| 996 | // success. |
| 997 | getNextToken(); // eat ')'. |
| 998 | |
| 999 | return new PrototypeAST(FnName, ArgNames); |
| 1000 | } |
| 1001 | |
| 1002 | /// definition ::= 'def' prototype expression |
| 1003 | static FunctionAST *ParseDefinition() { |
| 1004 | getNextToken(); // eat def. |
| 1005 | PrototypeAST *Proto = ParsePrototype(); |
| 1006 | if (Proto == 0) return 0; |
| 1007 | |
| 1008 | if (ExprAST *E = ParseExpression()) |
| 1009 | return new FunctionAST(Proto, E); |
| 1010 | return 0; |
| 1011 | } |
| 1012 | |
| 1013 | /// toplevelexpr ::= expression |
| 1014 | static FunctionAST *ParseTopLevelExpr() { |
| 1015 | if (ExprAST *E = ParseExpression()) { |
| 1016 | // Make an anonymous proto. |
| 1017 | PrototypeAST *Proto = new PrototypeAST("", std::vector<std::string>()); |
| 1018 | return new FunctionAST(Proto, E); |
| 1019 | } |
| 1020 | return 0; |
| 1021 | } |
| 1022 | |
| 1023 | /// external ::= 'extern' prototype |
| 1024 | static PrototypeAST *ParseExtern() { |
| 1025 | getNextToken(); // eat extern. |
| 1026 | return ParsePrototype(); |
| 1027 | } |
| 1028 | |
| 1029 | //===----------------------------------------------------------------------===// |
| 1030 | // Top-Level parsing |
| 1031 | //===----------------------------------------------------------------------===// |
| 1032 | |
| 1033 | static void HandleDefinition() { |
| 1034 | if (ParseDefinition()) { |
| 1035 | fprintf(stderr, "Parsed a function definition.\n"); |
| 1036 | } else { |
| 1037 | // Skip token for error recovery. |
| 1038 | getNextToken(); |
| 1039 | } |
| 1040 | } |
| 1041 | |
| 1042 | static void HandleExtern() { |
| 1043 | if (ParseExtern()) { |
| 1044 | fprintf(stderr, "Parsed an extern\n"); |
| 1045 | } else { |
| 1046 | // Skip token for error recovery. |
| 1047 | getNextToken(); |
| 1048 | } |
| 1049 | } |
| 1050 | |
| 1051 | static void HandleTopLevelExpression() { |
| 1052 | // Evaluate a top-level expression into an anonymous function. |
| 1053 | if (ParseTopLevelExpr()) { |
| 1054 | fprintf(stderr, "Parsed a top-level expr\n"); |
| 1055 | } else { |
| 1056 | // Skip token for error recovery. |
| 1057 | getNextToken(); |
| 1058 | } |
| 1059 | } |
| 1060 | |
| 1061 | /// top ::= definition | external | expression | ';' |
| 1062 | static void MainLoop() { |
| 1063 | while (1) { |
| 1064 | fprintf(stderr, "ready> "); |
| 1065 | switch (CurTok) { |
| 1066 | case tok_eof: return; |
| 1067 | case ';': getNextToken(); break; // ignore top-level semicolons. |
| 1068 | case tok_def: HandleDefinition(); break; |
| 1069 | case tok_extern: HandleExtern(); break; |
| 1070 | default: HandleTopLevelExpression(); break; |
| 1071 | } |
| 1072 | } |
| 1073 | } |
| 1074 | |
| 1075 | //===----------------------------------------------------------------------===// |
| 1076 | // Main driver code. |
| 1077 | //===----------------------------------------------------------------------===// |
| 1078 | |
| 1079 | int main() { |
| 1080 | // Install standard binary operators. |
| 1081 | // 1 is lowest precedence. |
| 1082 | BinopPrecedence['<'] = 10; |
| 1083 | BinopPrecedence['+'] = 20; |
| 1084 | BinopPrecedence['-'] = 20; |
| 1085 | BinopPrecedence['*'] = 40; // highest. |
| 1086 | |
| 1087 | // Prime the first token. |
| 1088 | fprintf(stderr, "ready> "); |
| 1089 | getNextToken(); |
| 1090 | |
| 1091 | // Run the main "interpreter loop" now. |
| 1092 | MainLoop(); |
| 1093 | |
| 1094 | return 0; |
| 1095 | } |
| 1096 | |
| 1097 | `Next: Implementing Code Generation to LLVM IR <LangImpl3.html>`_ |
| 1098 | |