Parsing Tools Aslan Askarov aslan@cs.au.dk Today: using parsing - - PowerPoint PPT Presentation

Compilation 2016

Parsing Tools

Aslan Askarov aslan@cs.au.dk

Today: using parsing tools to parse 3 simple languages

• 1. Language of arithmetic expressions
• 2. Straight-line commands
• 3. Straight-line commands with alternative syntax

Language of arithmetic expressions

expr -> id | num | expr op expr | ( expr )

• p -> plus | minus | times | div

datatype aexp = Id of string | Number of int | Op of binop * aexp * aexp and binop = Plus | Minus | Times | Div

Example: 1 + (2*x- 3) CFG: ML code:

?

Using ml-yacc

ML declarations to be copied verbatim into the generated parser %% parser declarations (nonterminals, terminals, precedence, etc) %% grammar rules .grm file format parser generator .grm.sml file that implements an LR parser .grm.desc file that helps debug grammar conflicts make sure to use %verbose declaration lexer the rest of the compiler

>_

Operator associativity

• Consider expression 7 - 5 - 3
• What should be the default interpretation?

(7 - 5) - 3 7 - (5 - 3)

Say that operator MINUS has left associativity … right associativity

error

also a possibility

• For arithmetic MINUS the correct choices are either left

associativity or error, but not right associativity

• The desired choice is left associativity

Examples of associativity

Operators Associativity Arithmetic minus, division Left Arithmetic addition, multiplication Left/Right a+b+c = (a+b)+c = a+(b+c) Assignment in C: x = y = 5 Right y is assigned the value of 5; x is assigned the updated value of y Arithmetic exponentiation 2^1^3 Right 2^1^3 = 2^ (1^3) = 2
 not the same as (2^1)^3 = 8 Comparison operators 2 < 3 < 4 Non-associative parse error

What can parse trees tell about associativity?

• 7

5 3

left-associative parse tree

• 7
• 5

3

7 - 5 - 3

right-associative parse tree

?

Grammar conflicts

• Shift/reduce conflicts
• Typical fixes
• try specifying associativity/precedence
• otherwise rewrite grammar
• Default action: shift
• Reduce/reduce conflicts
• Rarely a good sign: must rewrite grammar

Scenario 1

Example shift/reduce conflict

exp : exp . MINUS exp exp : exp MINUS exp . (reduce by rule 4)

stack: input string exp MINUS exp exp MINUS

Scenario 1

Example shift/reduce conflict

exp : exp . MINUS exp exp : exp MINUS exp . (reduce by rule 4)

stack: input string exp MINUS exp exp MINUS shift action Scenario 2 stack: input string exp MINUS exp exp MINUS

Scenario 1

Example shift/reduce conflict

exp : exp . MINUS exp exp : exp MINUS exp . (reduce by rule 4)

stack: input string exp MINUS exp exp MINUS shift action Scenario 2 stack: input string exp exp MINUS reduce action

Example shift/reduce conflict

• Possible fix: specifying associativity using%left or

%right directive

• %right favors shifting
• %left favors reducing
• Q: What should we do for MINUS?
• Other fixes: precedence is given by the order of

parser directives in the .grm file

• If not enough: rework the grammar!

Language of straight-line programs

… cmd -> id := aexp | if aexp then cmd | if aexp then cmd else cmd | ( cmds ) cmds -> cmd 
 | cmd ; cmds … and cmd = Assign of id * aexp | If of aexp * cmd * cmd option | Cmds of cmd list

CFG: ML code:

Example: y:=1 + (2*x- 3); 
 if x then if y then z := 20 else y := 2

Syntactic variation of SLP

… cmd -> id := aexp | if aexp then cmd fi | if aexp then cmd else cmd fi | ( cmds ) cmds -> cmd 
 | cmd ; cmds … and cmd = Assign of id * aexp | If of aexp * cmd * cmd option | Cmds of cmd list

CFG: ML code:

Example: if x then if y then z := 20 else y := 2 fi fi

extra keyword to mark the end of IfThenElse same AST new grammar

Syntactic variation of SLP

• Pros of explicit syntax
• Avoids some grammar conflicts
• Handy at prototyping stages when designing

your own PL and the language features are not yet stable (unlike this course)

• Cons
• Not very elegant; your programmers may be

grumpy

Example: if x then if y then z := 20 else y := 2 fi fi

Reporting syntax errors

• Simplest strategy: fail on the first syntax error
• OK, but not very helpful
• Local error recovery:
• adjust parse stack/input at the point of error
• used in many variants of Yacc (but not ML-Yacc)
• Global error recovery:
• adjust input stream before the point of error
• see Burke-Fisher error repair
• essentially tries all possible edits
• used in ML-Yacc
• see %change and %value directives

Summary

• Use %left, %right parser declarations to control

associativity

• See ML-Yacc manual on Precedence
• Conflict reports in .desc file
• Develop your grammar slowly to understand the

source of conflicts

• Exploring design space in developing surface

syntax

• ML-Yacc uses global error recovery