Parsing Tools

Parsing Tools Introduction to Bison and Flex

Scanning/parsing tools lex - original UNIX lexics generator (Lesk, 1975) • create a C function that will parse input according to a set of regular expressions yacc - "yet another compiler compiler" UNIX parser (Johnson, 1975) • generate a C program for a parser from BNF rules bison and flex ("fast lex") - more powerful, free versions of yacc and lex, from GNU Software Fnd'n. Jflex - generates Java code for a scanner CUP - generates Java code for a parser

Bison Overview Purpose: automatically write a parser program for a grammar written in BNF. Usage: you write a bison source file containing rules that look like BNF. Bison creates a C program that parses according to the rules term : term '*' factor { $$ = $1 * $3; } | term '/' factor { $$ = $1 / $3; } | factor { $$ = $1; } ; factor : ID { $$ = valueof($1); } | NUMBER { $$ = $1; } ;

Bison Overview (2) The programmer puts BNF rules and token rules for the parser he wants in a bison source file myparser.y myparser.y BNF rules and actions for your grammar. run bison to create a C program (*.tab.c) containing a parser function. The programmer must also supply a tokenizer named yylex( ) > bison myparser.y myparser.tab.c parser source code yylex.c tokenizer function in C > gcc -o myprog myparser.tab.c yylex.c myprog executable program

Bison Overview (3) In operation: your main program calls yyparse( ). yyparse( ) calls yylex when it wants a token. yylex returns the type of the token. yylex puts the value of the token in a global variable named yylval input file to be parsed. yylex( ) tokenizer returns the type of the next token yylval yyparse( ) parser created by bison parse tree or other result

Bison source file The file has 3 sections, separated by "%%" lines. /* declarations go here*/ %% /* grammar rules go here */ %% /* additional C code goes here */ Note: format for "yacc" is the same as for bison.

Bison source file with C declarations You usually include C code in the declarations section. %{ /* C declarations and #define statements go here */ #include <stdio.h> #define YYSTYPE double %} /* bison declarations go here*/ %% /* grammar rules go here */ %% /* additional C code goes here */ Declare that yylval will be a "double" type.

Bison/Yacc file for example (1) Structure of Bison or Yacc input: %{ /* C declarations and #DEFINE statements go here */ #include <stdio.h> #define YYSTYPE double %} /* Bison/Yacc declarations go here*/ %token NUMBER /* define token type NUMBER */ %left '+' '-' /* + and - are left associative */ %left '*' '/' /* * and / are left associative */ %% /* grammar rules go here */ %% /* additional C code goes here */

Bison/Yacc example (2) %% /* Bison grammar rules */ input : /* empty production to allow an empty input */ | input line ; line : expr '\n' { printf("Result is %f\n", $1); } expr : expr '+' term { $$ = $1 + $3; } | expr '-' term { $$ = $1 - $3; } | term { $$ = $1; } ; term : term '*' factor { $$ = $1 * $3; } | term '/' factor { $$ = $1 / $3; } | factor { $$ = $1; } ; factor : '(' expr ')' { $$ = $2; } | NUMBER { $$ = $1; } ;

Bison/Yacc example (3) • $1, $2, ... represent the actual values of tokens or non-terminals (rules) that match the production. • $$ is the result. rule pattern to match action expr : expr '+' term { $$ = $1 + $3; } | expr '-' term { $$ = $1 - $3; } | term { $$ = $1; } ; Example:if the input matches expr + term then set the result ($$) equal to the sum of expr plus term ($1 + $3).

Bison/Yacc example (4) Q: why can we write "$$ = $1 + $3" ? A: because we declared "#define YYSTYPE double", so all tokens and results are double. rule pattern to match action expr : expr '+' term { $$ = $1 + $3; } | expr '-' term { $$ = $1 - $3; } | term { $$ = $1; } ;

Scanner function: yylex( ) • You must supply a scanner named yylex. • yylex returns the token TYPE (not token value). int yylex( void ) { int c = getchar(); /* read from stdin */ if (c < 0) return 0; /* end of input */ if ( c == '+' || c == '-' ) return c; /* for character tokens, TYPE = character itself */ if ( isdigit(c) ) { yylval = c - '0'; /* yylval is a global var */ while( isdigit( c=getchar() ) ) yylval = 10*yylval + (c - '0'); if (c >= 0) ungetc(c,stdin); return NUMBER; /* token type is NUMBER */ } ... }

Where is the token value? • The value of the token is stored in a global variable named yylval. int yylex( void ) { int c = getchar(); /* read from stdin */ if (c < 0) return 0; /* end of input */ if ( c == '+' || c == '-' ) return c; /* read number and store as yylval */ if ( isdigit(c) ) { yylval = c - '0'; /* get each digit */ while( isdigit( c=getchar() ) ) yylval = 10*yylval + (c - '0'); /* push next character back into input stream */ if (c >= 0) ungetc(c,stdin); return NUMBER; /* token type is NUMBER */ } ... }

Useful C functions for yylex int c = getchar( ); read next char from stdin returns -1 at end of input ungetc( c, stdin) put character back in input #include <ctype.h> isdigit(c) true if c contains a digit isalpha(c) true if c contains a letter isalnum(c) isdigit(c) || isalnum(c) islower(c) true if c is lowercase isupper(c) guess? isspace(c) space tab newline formfeed scanf("%d", &num) read an integer from input scanf("%lf", &dnum) read a double from input

Scanner function for double values • Suppose we want the data type of all tokens to be double. • In Bison input: #defineYYSTYPE double. • This changes yylval to be a double! %{ /* C declarations and #DEFINE statements go here */ #include <stdio.h> #define YYSTYPE double %} %% /* bison definitions go here */

Scanner function for double • Now yylex must know that yylval is "extern double". • Here is example of using scanf to parse numbers. int yylex( void ) { int c = getchar(); /* read from stdin */ if (c < 0) return 0; /* end of the input*/ while ( c == ' ' || c == '\t' ) c = getchar( ); if ( isdigit(c) || c == '.' ) { ungetc(c, stdin); /* put c back into input */ scanf ("%lf", &yylval); /* get value using scanf */ return NUMBER; /* return the token type */ } return c; /* anything else... return char itself */ }

Other C functions: yyerror • Bison requires an error routine named yyerror. • yyerror is called by the parser when there is an error. • You can include yyerror( ) in your Bison source file. /* display error message */ int yyerror( char *errmsg ) { printf("%s\n", errmsg); }

Other C functions: main • you need a write a main() function that starts the parser. • For a simple parser, main() calls yyparse(). /* main method to run the program */ int main( ) { printf("Type some input. Enter ? for help.\n"); yyparse( ); }

Running Bison • Compile the example file simple.y CMD>bison simple.y • Output is "simple.tab.c" it is C code for the parser. • To compile the parser by itself using gcc: CMD> gcc -o simple.exe simple.tab.c • To run the program at the command prompt: CMD> simple.exe

Simple example: definitions /* The Bison declarations section */ %{ /* C declarations and #DEFINE statements go here */ #include <math.h> #define YYSTYPE double %} %token NUMBER /* define token type for numbers */ %token '+' '-' /* + and - are left associative */ File:simple.y No left/right associativity specified

Simple example: grammar rules %% /* Simple grammar rules */ input : /* allow empty input */ | input line ; line : expr '\n' { printf("answer: %d\n", $1); } expr : expr '+' term { $$ = $1 + $3; } | expr '-' term { $$ = $1 - $3; } | term { $$ = $1; } ; term : NUMBER { $$ = $1; } ;

yyerror and main %% /* extra C code */ /* display error message */ int yyerror( char *errmsg ) { printf("%s\n", errmsg); } /* main */ int main() { printf("type an expression:\n"); yyparse( ); }

Simple example: does it work? cmd> bison simple.y cmd> gcc -o calc.exe simple.tab.c cmd> calc Test the grammar rules: 3 + 5 3 - 4 - 5 + 6 How about this: -3

Simple example: exploring BNF %% /* Simple grammar rules */ input : /* empty input */ | input line ; line : expr '\n' { printf("answer: %d\n", $1); } expr : expr '+' expr { $$ = $1 + $3; } | expr '-' expr { $$ = $1 - $3; } | term { $$ = $1; } ; term : NUMBER { $$ = $1; } | '-' NUMBER { $$ = -$2; } ; %token NUMBER %right '+' '-'

Exercise Expand the grammar to include these operations: 4 * 5 multiplication 2 / 3 division 10 + 3 * 4 – 1 / 2 correct order of operations 2 * ( 3 + 4 ) grouping

Complete example: definitions /* The Bison declarations section */ %{ /* C declarations and #DEFINE statements go here */ #include <math.h> #define YYSTYPE double %} %token NUMBER /* define token type for numbers */ %left '+' '-' /* + and - are left associative */ %left '*' '/' /* * and / are left associative */ File:simple.y

Complete example: grammar rules %% /* Bison grammar rules */ input : /* allow empty input */ | input line ; line : expr '\n' { printf("Result is %f\n", $1); } expr : expr '+' term { $$ = $1 + $3; } | expr '-' term { $$ = $1 - $3; } | term { $$ = $1; } ; term : term '*' factor { $$ = $1 * $3; } | term '/' factor { $$ = $1 / $3; } | factor { $$ = $1; } ; factor : '(' expr ')' { $$ = $2; } | NUMBER { $$ = $1; } | '-' NUMBER { $$ = -$2; } ;

Common Errors in Bison • Forgetting to quote literals: term '+' term • Not skipping space where space is allowed • Ambiguity in grammar rules • Forgetting ';' at the end of rule %token NUMBER %token + - %% /* grammar rules */ expr : term + term { $$ = $1 + $3; } | term - term { $$ = $1 - $3; } | term { $$ = $1; } term : NUMBER { $$ = $1; } | - NUMBER { $$ = -$2; }

Shift / Reduce and operator order • Bison uses token look-ahead and a token stack. It shifts tokens onto the stack until it can choose a rule to reduce (replace) tokens with a non-terminal. Example: expr ::= expr + expr | expr - expr | expr * expr | expr / expr | term term ::= NUMBER | ( expr ) • Suppose the input read is 10 - • shift these onto stack because no matching rule yet. • Suppose the next token is 2. What should Bison do?

Shift / Reduce and operator order STACK: 10 - CURRENT TOKEN: 2 • This grammar is ambiguous. Bison could match "expr - expr" or it could shift 2 onto the stack and look at the next token... it maybe * such as: 10 - 2 * 3 • This is called a "shift / reduce conflict". • Unless you specify a disambiguating rule (next slide), Bison chooses "shift" over "reduce". • Meaning: if it's not clear how to resolve conflicts, wait. INPUT ACTION STACK 10 shift 10 - shift 10 - 2 shift 10 - 2 * shift 10 - 2 * INPUT ACTION STACK 3 reduce 10 - 6 reduce 4 done

The declarations section • The declarations section can contain Bison directives and C directives. • We already saw the use of %left, %right, etc. %{ /* semantic value of tokens, as C datatype */ #define YYSTYPE double #include <math.h> /* why do we have to declare these? */ int yylex (void); void yyerror (char const *); %} %token NUMBER %left '+' '-' %left '*' '/'

Specifying operator order • In the declarations section you can write: %noassoc NEG /* unary minus sign: - 3 */ %left '+' '-' %left '*' '/' '%' %right '^' • '+' and '-' are left associative and have the same precedence. • '*', '/', and '%' left associative and same precedence; they have higher precedence (later declarations are higher). • '^' is right associative and higher precedence than + - * / % • 'NEG' has no associativity: "- - 3" is an error

Scanner function main points • Tokens have a TYPE and a VALUE. • The scanner (yylex) returns the TYPE of the next token • For one-character tokens like '+', '=', "(' the character itself can be used as the type. • Define symbolic names for token types in Bison using %token NAME • Return the VALUE of a token using the global variable yylval. • By default, yylval is type "int". Change it using: #defineYYSTYPE double • Rather than parse numbers yourself, use scanf.

Scanner function main points (1) %{ #include <ctype.h> #include <math.h> #define YYSTYPE double %} %token NUMBER %token SQRT %left '+' '-' %left '*' '/' %% line : expr '\n' { printf("%g\n", $$); } ; expr : expr * expr { $$ = $1 + $3; } | SQRT '(' expr ')' { $$ = sqrt( $2 ); } ... /* more rules */ Token values are double Token for sqrt function

Scanner function main points (2) %% int yylex(void) { int c = getchar( ); while ( c == ' ' || c == '\t' ) c = getchar( ); if ( isdigit(c) ) { ungetc(c,stdin); scanf("%f", &yylval); return NUMBER; } if ( isalpha(c) ) { char *word = getword(c); /* get next word */ if ( strcmp(word,"sqrt") ) return SQRT; ... } Token is a number Token is sqrt function For a more general way of handling identifiers, see the Bison User's Guide, "mfcalc" example.

Handling Multiple Data Types • For a more general grammar, the scanner should be able to return different data types as the value (yylval). • In definitions section, define "%union" as union of all data types that yylval (and hence $1, $2, ...) can have. %union { double number; char* string; } %token <number> NUMBER %token <string> IDENT %type <number> expr %type <number> term %left '+' '-' %left '*' '/' Define all the data types that token values can have For each token type, define the data type of its value. For tokens that represent their own value, don't need to define data type.

Handling Multiple Data Types (2) %% int yylex(void) { int c = getchar( ); while ( c == ' ' || c == '\t' ) c = getchar( ); if ( isdigit(c) ) { ungetc(c,stdin); double x; scanf("%f", &x); yylval.number = x; return NUMBER; } if ( isalpha(c) ) { char *word = getword(c); /* get next word */ yylval.string = word; return IDENT; ... } token value is a double token value is a string

Using a Separate File for yylex • You can put the scanner (yylex) in a separate file. • BUT, yylex needs values that are defined by Bison, such as NUMBER, IDENT, yylval. • Solution: add the "%defines" option to your rules • Bison will create a header file named "simple.tab.h". %defines %{ #include <math.h> #define YYSTYPE double %} token NUMBER ...

Using a Makefile • make is an automatic build facility. Just type "make" • Reads a Makefile of "rules" for how to make things # Makefile for simple calculator calc.exe: simple.tab.c yylex.o gcc -o calc.exe simple.tab.c yylex.o -lm # how to make simple.tab.c from simple.y simple.tab.c: simple.y bison simple.y # yylex.o (obj. file) depends on simple.tab.h yylex.o: yylex.c simple.tab.h gcc –c yylex.c

Running Make C:/calculator> make bison -d simple.y gcc -c simple.tab.c gcc -o simple.exe simple.tab.o yylex.c -lm

Advantages of Make • Makefile record all the dependencies in a project • make only builds the parts that are missing or out of date • can manage complex projects with multiple Makefiles in subdirectories

Introduction to flex • Flex is a program that automatically creates a scanner in C, using rules for tokens as regular expressions. • Format of the input file is like Bison. %{ /* C definitions for scanner */ %} flex definitions %% rules %% user code (extra C code)

Flex Example • Read console input and describes each token read. %{ inline int yywrap(void) { return 1; }; %} /* flex definitions */ DIGIT [0-9] LETTER [A-Za-z] %% {LETTER}+ printf("Word: %s\n", yytext); return 1; -?{DIGIT}+ printf("Number: %s\n", yytext); return 2; [[:punct:]] printf("Punct: %s\n", yytext); return 3; \n printf("End of line\n"); return 0; %% /* main method calls yylex to tokenize input */ int main( ) { printf("Type something: "); while( yylex() > 0 ) { }; }

Flex Example (2) • Run flex to create yylex.c (on Linux: lex.yy.c). cmd> flex mylex.fl • Compile lex.yy.c and create myscanner.exe cmd> gcc -o myscanner.exe yylex.c • Run the program: cmd> myscanner Type something: hello, it's 9:00 o'clock. word: hello punct: , word: it punct: '

Flex Explained: definitions %{ /* C definitions and includes */ #include "myparser.tab.h" /* this fixes an unknown symbol "yywrap" */ #ifdef yywrap #undef yywrap #endif inline int yywrap(void) { return 0; }; %} /* flex definitions */ DIGIT [0-9] LETTER [A-Za-z] %%

Flex Explained: Parsing Rules (1) • pattern is a regular expression or a literal value. • action is zero or more C statements to execute. • The current matched input is (char *) yytext • The length of matched input is yyleng {LETTER}+ printf("Found a word: %s", yytext); -?[0-9]+ printf("Found an int: %s", yytext); pattern action yytext is a string containing the current token.

Flex Explained: Parsing Rules (2) • Keep matching input until a return statement. • Flex chooses the rule that produces longest match. • If there is a tie, choose the first rulethat matches. • Use yylval (global variable) for value of token. %{ /* Bison header file defines INT, IDENT...*/ #include "myparser.tab.h" %} %% [0-9]+ yylval.itype = atoi(yytext); return INT; {LETTER}+ yylval.ctype = yytext; return IDENT; "+"|"-" yylval.ctype = yytext; return OP; "=" return ASSIGN;

Flex yywrap Flex can read input from multiple input files. • when flex reaches the end of a file, it calls yywrap() • if yywrap() returns 0, it means that there is another input file to read. • if yywrap() returns 1, it means that there are no more input files. %{ inline int yywrap(void) { return 1; }; %} /* same thing */ %option noyywrap

Using flex with bison • Run "bison -d file.y" to create file.tab.h • #include "file.tab.h" in your flex source. • Set yylval to the token value. • Run flex. • Compile and link lex.yy.c and file.tab.c %{ #include <math.h> #include "file.tab.h" %} LETTER [A-Za-z] /* same as [[:alpha:]] */ %% -?[0-9]+ yylval.itype = atoi(yytext); return INT; {LETTER}+ yylval.ctype = yytext; return IDENT; "+"|"-" yylval.ctype = yytext; return OP; "=" return ASSIGN;

Parsing Tools

Parsing Tools

Presentation Transcript

Parsing Tools

Parsing

Parsing

Parsing

Parsing

Parsing

Parsing

Parsing

Parsing

Parsing

Parsing

Parsing

Parsing

XML Apache Tools Parsing and Transformation

Parsing

Parsing

Parsing Tools

Parsing