Announcements • PA1 – Due today at midnight Error Handling – README, test case Syntax-Directed Translation – Your name(s)! Recursive Descent Parsing • WA1 – Due today at 5pm Lecture 6 • PA2 – Assigned today • WA2 – Assigned Tuesday Prof. Aiken CS 143 Lecture 6 1 Prof. Aiken CS 143 Lecture 6 2 Outline Error Handling • Extensions of CFG for parsing • Purpose of the compiler is – Precedence declarations – To detect non-valid programs – Error handling – To translate the valid ones – Semantic actions • Many kinds of possible errors (e.g. in C) Error kind Example Detected by … • Constructing a parse tree Lexical … $ … Lexer Syntax … x *% … Parser • Recursive descent Semantic … int x; y = x(3); … Type checker Correctness your favorite program Tester/User Prof. Aiken CS 143 Lecture 6 3 Prof. Aiken CS 143 Lecture 6 4 Syntax Error Handling Approaches to Syntax Error Recovery • Error handler should • From simple to complex – Report errors accurately and clearly – Panic mode – Recover from an error quickly – Error productions – Not slow down compilation of valid code – Automatic local or global correction • Good error handling is not easy to achieve • Not all are supported by all parser generators Prof. Aiken CS 143 Lecture 6 5 Prof. Aiken CS 143 Lecture 6 6 1 Error Recovery: Panic Mode Syntax Error Recovery: Panic Mode (Cont.) • Simplest, most popular method • Consider the erroneous expression (1 + + 2) + 3 • When an error is detected: • Panic-mode recovery: – Discard tokens until one with a clear role is found – Skip ahead to next integer and then continue – Continue from there • Bison: use the special terminal error to • Such tokens are called synchronizing tokens describe how much input to skip – Typically the statement or expression terminators E int | E + E | ( E ) | error int | ( error ) Prof. Aiken CS 143 Lecture 6 7 Prof. Aiken CS 143 Lecture 6 8 Syntax Error Recovery: Error Productions Error Recovery: Local and Global Correction • Idea: specify in the grammar known common mistakes • Idea: find a correct “nearby” program – Try token insertions and deletions • Essentially promotes common errors to alternative – Exhaustive search syntax • Example: • Disadvantages: – Write 5 x instead of 5 * x – Hard to implement – Add the production E … | E E – Slows down parsing of correct programs • Disadvantage – “Nearby” is not necessarily “the intended” program – Complicates the grammar – Not all tools support it Prof. Aiken CS 143 Lecture 6 9 Prof. Aiken CS 143 Lecture 6 10 Syntax Error Recovery: Past and Present Abstract Syntax Trees • Past • So far a parser traces the derivation of a – Slow recompilation cycle (even once a day) sequence of tokens – Find as many errors in one cycle as possible – Researchers could not let go of the topic • The rest of the compiler needs a structural • Present representation of the program – Quick recompilation cycle – Users tend to correct one error/cycle • Abstract syntax trees – Complex error recovery is less compelling – Like parse trees but ignore some details – Panic-mode seems enough – Abbreviated as AST Prof. Aiken CS 143 Lecture 6 11 Prof. Aiken CS 143 Lecture 6 12 2 Abstract Syntax Tree. (Cont.) Example of Parse Tree • Consider the grammar E E int | ( E ) | E + E • Traces the operation E + E of the parser • And the string 5 + (2 + 3) • Does capture the int5 ( E ) nesting structure • After lexical analysis (a list of tokens) • But too much info E + int5 ‘+’ ‘(‘ int2 ‘+’ int3 ‘)’ E – Parentheses – Single-successor int nodes • During parsing we build a parse tree … 2 int3 Prof. Aiken CS 143 Lecture 6 13 Prof. Aiken CS 143 Lecture 6 14 Example of Abstract Syntax Tree Semantic Actions PLUS • This is what we’ll use to construct ASTs PLUS • Each grammar symbol may have attributes – For terminal symbols (lexical tokens) attributes can 5 2 3 be calculated by the lexer • Also captures the nesting structure • But abstracts from the concrete syntax • Each production may have an action – Written as: X Y … Y { action } => more compact and easier to use 1 n – That can refer to or compute symbol attributes • An important data structure in a compiler Prof. Aiken CS 143 Lecture 6 15 Prof. Aiken CS 143 Lecture 6 16 Semantic Actions: An Example Semantic Actions: An Example (Cont.) • Consider the grammar • String: 5 + (2 + 3) E int | E + E | ( E ) • Tokens: int5 ‘+’ ‘(‘ int2 ‘+’ int3 ‘)’ • For each symbol X define an attribute X.val – For terminals, val is the associated lexeme Productions Equations – For non-terminals, val is the expression’s value (and is computed from values of subexpressions) E E1 + E2 E.val = E1.val + E2.val E1 int5 E1.val = int5.val = 5 • We annotate the grammar with actions: E2 ( E3) E2.val = E3.val E int { E.val = int.val } E3 E4 + E5 E3.val = E4.val + E5.val | E1 + E2 { E.val = E1.val + E2.val } E4 int2 E4.val = int2.val = 2 | ( E1 ) { E.val = E1.val } E5 int3 E5.val = int3.val = 3 Prof. Aiken CS 143 Lecture 6 17 Prof. Aiken CS 143 Lecture 6 18 3 Semantic Actions: Notes Dependency Graph • Semantic actions specify a system of E + • Each node labeled E equations has one slot for the val – Order of resolution is not specified E1 + E2 attribute • Note the dependencies • Example: int 5 5 ( E3 + ) E3.val = E4.val + E5.val – Must compute E4.val and E5.val before E3.val E + – We say that E3.val depends on E4.val and E5.val 4 E5 int 2 • The parser must find the order of evaluation 2 int3 3 Prof. Aiken CS 143 Lecture 6 19 Prof. Aiken CS 143 Lecture 6 20 Evaluating Attributes Dependency Graph • An attribute must be computed after all its E 10 successors in the dependency graph have been 5 5 computed E1 + E2 – In previous example attributes can be computed bottom-up int 5 5 ( E3 5 ) • Such an order exists when there are no cycles E 2 + 4 E5 3 – Cyclically defined attributes are not legal int 2 2 int3 3 Prof. Aiken CS 143 Lecture 6 21 Prof. Aiken CS 143 Lecture 6 22 Semantic Actions: Notes (Cont.) Inherited Attributes • Synthesized attributes • Another kind of attribute – Calculated from attributes of descendents in the parse tree • Calculated from attributes of parent and/or – E.val is a synthesized attribute siblings in the parse tree – Can always be calculated in a bottom-up order • Grammars with only synthesized attributes • Example: a line calculator are called S-attributed grammars – Most common case Prof. Aiken CS 143 Lecture 6 23 Prof. Aiken CS 143 Lecture 6 24 4 A Line Calculator Attributes for the Line Calculator • Each line contains an expression • Each E has a synthesized attribute val E int | E + E – Calculated as before • Each line is terminated with the = sign • Each L has an attribute val L E = | + E = L E = { L.val = E.val } | + E = { L.val = E.val + L.prev } • In second form the value of previous line is used as starting value • We need the value of the previous line • A program is a sequence of lines • We use an inherited attribute L.prev P | P L Prof. Aiken CS 143 Lecture 6 25 Prof. Aiken CS 143 Lecture 6 26 Attributes for the Line Calculator (Cont.) Example of Inherited Attributes • Each P has a synthesized attribute val P • val synthesized – The value of its last line + P { P.val = 0 } P L | P1 L { P.val = L.val; • prev inherited = L.prev = P1.val } + + E3 – Each L has an inherited attribute prev 0 – L.prev is inherited from sibling P .val • All can be 1 E + 4 E5 computed in depth-first • Example … int 2 order 2 int3 3 Prof. Aiken CS 143 Lecture 6 27 Prof. Aiken CS 143 Lecture 6 28 Example of Inherited Attributes Semantic Actions: Notes (Cont.) P 5 • val synthesized • Semantic actions can be used to build ASTs P 0 0 L 5 • And many other things as well • prev inherited – Also used for type checking, code generation, … + E 5 = 0 3 • All can be • Process is called syntax-directed translation E 2 + 4 E5 3 computed in – Substantial generalization over CFGs depth-first int 2 order 2 int3 3 Prof. Aiken CS 143 Lecture 6 29 Prof. Aiken CS 143 Lecture 6 30 5 Constructing An AST Constructing a Parse Tree • We first define the AST data type • We define a synthesized attribute ast – Supplied by us for the project – Values of ast values are ASTs • Consider an abstract tree type with two – We assume that int.lexval is the value of the constructors: integer lexeme – Computed using semantic actions mkleaf(n) = n E int E.ast = mkleaf(int.lexval) PLUS mkplus( , ) = | E1 + E2 E.ast = mkplus(E1.ast, E2.ast) | ( E1 ) E.ast = E1.ast T1 T2 T1 T2 Prof. Aiken CS 143 Lecture 6 31 Prof. Aiken CS 143 Lecture 6 32 Parse Tree Example Summary • Consider the string int5 ‘+’ ‘(‘ int2 ‘+’ int3 ‘)’ • We can specify language syntax using CFG • A bottom-up evaluation of the ast attribute: E.ast = mkplus(mkleaf(5), • A parser will answer whether s L(G) mkplus(mkleaf(2), mkleaf(3)) – … and will build a parse tree PLUS – … which we convert to an AST – … and pass on to the rest of the compiler PLUS 5 2 3 Prof. Aiken CS 143 Lecture 6 33 Prof. Aiken CS 143 Lecture 6 34 Intro to Top-Down Parsing: The Idea Recursive Descent Parsing • Consider the grammar • The parse tree is constructed 1 E T |T + E – From the top T int | int * T | ( E ) – From left to right t2 3 t9 • Token stream is: ( int5 ) • Terminals are seen in order of 4 7 appearance in the token • Start with top-level non-terminal E stream: t5 t6 t8 – Try the rules for E in order t2 t5 t6 t8 t9 Prof.