Semantic Analysis

CSCI 3136 Principles of Programming Languages Faculty of Computer Science Dalhousie University Winter 2012 Reading: Chapter 4

Motivation
Source program (character stream) Front end Scanner (lexical analysis) Token stream Parser (syntactic analysis) Parse tree Symbol table Semantic analysis and code generation Machine-independent code improvement Back end Modied intermediate form Target code generation Target language (e.g., assembly) Abstract syntax tree or other intermediate form

Modied target language

Machine-specic code improvement

Semantic Analysis CSCI 3136: Principles of Programming Languages

Syntax and Semantics

Syntax Describes form of a valid program Can be described by a context-free grammar Semantics Describes meaning of a program Cannot be be described by a context-free grammar Some constraints that may appear syntactic are enforced by semantic analysis: E.g., use of identier only after its declaration

Semantic Analysis CSCI 3136: Principles of Programming Languages

Semantic Analysis

Enforces semantic rules Builds intermediate representation (e.g., abstract syntax tree) Fills symbol table Passes results to intermediate code generator Two approaches Interleaved with syntactic analysis As a separate phase Formal mechanism: Attributes grammars

Semantic Analysis CSCI 3136: Principles of Programming Languages

Enforcing Semantic Rules

Static semantic rules Enforced by compiler at compile time Example: Do not use undeclared variable Dynamic semantic rules Compiler generates code for enforcement at run time Examples: division by zero, array index out of bounds Some compilers allow these checks to be disabled

Semantic Analysis CSCI 3136: Principles of Programming Languages

Attribute Grammars

An attribute grammar is an augmented context-free grammar: Each symbol in a production has a number of attributes. Each production is augmented with semantic rules that Copy attribute values between symbols, Evaluate attribute values using semantic functions, Enforce constraints on attribute values.

Semantic Analysis CSCI 3136: Principles of Programming Languages

Attribute Grammars: Example

EE+T EE-T ET TT *F TT /F TF F -F F ( E ) F const

Semantic Analysis CSCI 3136: Principles of Programming Languages

Attribute Grammars: Example

EE+T EE-T ET TT *F TT /F TF F -F F ( E ) F const

E1 E2 + T E1 E2 - T ET T1 T2 * F T1 T2 / F TF F1 - F2 F ( E ) F const

E1 .val := sum(E2 .val, T.val) E1 .val := difference(E2 .val, T.val) E.val := T.val T1 .val := product(T2 .val, F.val) T1 .val := quotient(T2 .val, F.val) T.val := F.val F1 .val := negate(F2 .val) F.val := E.val F.val := const.val

Semantic Analysis CSCI 3136: Principles of Programming Languages

Synthesized Attributes (Bottom-Up)

The language {an bn cn | n 1} = {abc, aabbcc, aaabbbccc, . . . } is not context-free but can be recognized by an attribute grammar. Synthesized attributes: Attributes of LHS are computed from attributes of RHS. S ABC Aa A1 A2 a Bb B1 B2 b C c C1 C2 c Condition: A.count = B.count = C.count A.count := 1 A1 .count := A2 .count + 1 B.count := 1 B1 .count := B2 .count + 1 C.count := 1 C1 .count := C2 .count + 1
Semantic Analysis CSCI 3136: Principles of Programming Languages

Synthesized Attributes: Parse Tree Decoration (1)

Input: aaabbbccc

Parse tree A A A

S B C

a a
B

b b
C

c c

Semantic Analysis CSCI 3136: Principles of Programming Languages

Synthesized Attributes: Parse Tree Decoration (1)

Input: aaabbbccc

Parse tree 3 A A 2 A 1

S 3=3=3 B 3 C 3 C 2 C 1

B 2 B 1

Semantic Analysis CSCI 3136: Principles of Programming Languages

Synthesized Attributes: Parse Tree Decoration (2)

Input: aaabbccc

Parse tree A A A

S B C

a a

b
C

c c

Semantic Analysis CSCI 3136: Principles of Programming Languages

Synthesized Attributes: Parse Tree Decoration (2)

Input: aaabbccc

Parse tree 3 A A 2 A 1

S 3=2=3 B 2 C 3 C 2 C 1

B 1

Semantic Analysis CSCI 3136: Principles of Programming Languages

Inherited Attributes (Top-Down)

Inherited attributes: Attributes ow from left to right (from LHS to RHS and from symbols of RHS to symbols of RHS further to the right). S ABC Aa A1 A2 a Bb B1 B2 b C c C1 C2 c B.inhCount := A.count; C.inhCount := A.count A.count := 1 A1 .count := A2 .count + 1 Condition: B.inhCount = 1 B2 .inhCount := B1 .inhCount 1 Condition: C.inhCount := 1 C2 .inhCount := C1 .inhCount 1

Semantic Analysis CSCI 3136: Principles of Programming Languages

Inherited Attributes: Parse Tree Decoration (1)

Input: aaabbbccc

Parse tree A A A

S B C

a a
B

b b
C

c c

Semantic Analysis CSCI 3136: Principles of Programming Languages

Inherited Attributes: Parse Tree Decoration (1)

Input: aaabbbccc

Parse tree 3 A A 2 A 1

S B 3 C 3 C 2 C 1=1

B 2 B 1=1

b b

c c

Semantic Analysis CSCI 3136: Principles of Programming Languages

Inherited Attributes: Parse Tree Decoration (2)

Input: aaabbccc

Parse tree A A A

S B C

a a

b
C

c c

Semantic Analysis CSCI 3136: Principles of Programming Languages

Inherited Attributes: Parse Tree Decoration (2)

Input: aaabbccc

Parse tree 3 A A 2 A 1

S B 3 C 3

B 2=1

C 2 C 1=1

c c

Semantic Analysis CSCI 3136: Principles of Programming Languages

Attribute Flow

Annotation or decoration of parse tree Process of evaluating attributes Synthesized attributes Attributes of LHS of a production are computed from attributes of the symbols in its RHS. Attributes ow bottom-up in the parse tree. Inherited attributes Attributes for symbols in RHS are computed from attributes of LHS and symbols in RHS preceding them. Attributes ow top-down in the parse tree.

Semantic Analysis CSCI 3136: Principles of Programming Languages

S-Attributed and L-Attributed Grammars

S-attributed grammar All attributes are synthesized. Attributes ow bottom-up. L-attributed grammar For each production X Y1 Y2 . . . Yk , X .syn depends on X .inh and Y1 .inh, Y1 .syn, Y2 .inh, Y2 .syn, . . . , Yk .inh, Yk .syn. For all 1 i k, Yi .inh depends on X .inh and Y1 .inh, Y1 .syn, Y2 .inh, Y2 .syn, . . . , Yi1 .inh, Yi1 .syn. S-attributed grammars are a special case of L-attributed grammars.

Semantic Analysis CSCI 3136: Principles of Programming Languages

LL(1) Parsing, Left-Associativity, and L-Attributed Grammars

A simple grammar for arithmetic expressions using only addition and subtraction 10 3 + 5 = (10 3) + 5 = 12 10 3 + 5 = 10 (3 + 5) = 2

Semantic Analysis CSCI 3136: Principles of Programming Languages

LL(1) Parsing, Left-Associativity, and L-Attributed Grammars

A simple grammar for arithmetic expressions using only addition and subtraction ET E EAT T n A+ AE T E A T E A T 10 3 + 5 = (10 3) + 5 = 12 10 3 + 5 = 10 (3 + 5) = 2

10

Semantic Analysis CSCI 3136: Principles of Programming Languages

LL(1) Parsing, Left-Associativity, and L-Attributed Grammars

A simple grammar for arithmetic expressions using only addition and subtraction ET E EAT T n A+ AE T E A T E A T 10 3 + 5 = (10 3) + 5 = 12 10 3 + 5 = 10 (3 + 5) = 2

10
This grammar captures left-associativity.

Semantic Analysis CSCI 3136: Principles of Programming Languages

LL(1) Parsing, Left-Associativity, and L-Attributed Grammars

A simple grammar for arithmetic expressions using only addition and subtraction ET E EAT T n A+ AE 10 T 10 E 7 A sub T 3 E 12 A add T 5 10 3 + 5 = (10 3) + 5 = 12 10 3 + 5 = 10 (3 + 5) = 2

5 5

3 3

10 10
This grammar captures left-associativity.

Semantic Analysis CSCI 3136: Principles of Programming Languages

LL(1) Parsing, Left-Associativity, and L-Attributed Grammars

A simple grammar for arithmetic expressions using only addition and subtraction ET E EAT T n A+ AE 10 T 10 E 7 A sub T 3 E 12 A add T 5 Rule R ET E EAT T n A+ APREDICT(R) {n} {n} {n} {+} {-} 10 3 + 5 = (10 3) + 5 = 12 10 3 + 5 = 10 (3 + 5) = 2

5 5

3 3

10 10
This grammar captures left-associativity. It is not LL(1).

Semantic Analysis CSCI 3136: Principles of Programming Languages

LL(1) Parsing, Left-Associativity, and L-Attributed Grammars

An LL(1) grammar for the same language E TE $ E E1 AT E2 T n A+ ARule R PREDICT(R)

{n} E TE $ E2 .op := A.fun(E1E T.val); E1 .val{$} E2 .val .op, := {+, -} E AT E {n} T n {+} A+ {-} A-

Semantic Analysis CSCI 3136: Principles of Programming Languages

LL(1) Parsing, Left-Associativity, and L-Attributed Grammars

An LL(1) grammar for the same language E TE $ E E1 AT E2 T n A+ ARule R PREDICT(R)

{n} E TE $ E2 .op := A.fun(E1E T.val); E1 .val{$} E2 .val .op, := {+, -} E AT E {n} T n {+} A+ {-} AE

E A T E A T

10

5
Semantic Analysis CSCI 3136: Principles of Programming Languages

LL(1) Parsing, Left-Associativity, and L-Attributed Grammars

An LL(1) grammar for the same language E TE $ E E1 AT E2 T n A+ AE .op := T.val; E.val := E .val E .val := E .op E2 .op := A.fun(E1 .op, T.val); E1 .val := E2 .val T.val := n.val A.fun := add A.fun := sub E 12 T 10

10 E 12
A sub T 3

10 10

7 E 12
A add T 5

3 3

12 E 12

5 5
Semantic Analysis CSCI 3136: Principles of Programming Languages

Action Routines

Action routines are instructions for ad-hoc translation interleaved with parsing. Parser generators allow programmers to specify action routines in the grammar. Action routines can appear anywhere in a rule (as long as the grammar is LL(1)). Example E1 A T { E2 .op = A.fun(E1 .op, T.val) } E2 { E1 .val = E2 .val } Action routines are supported, for example, in yacc and bison.

Semantic Analysis CSCI 3136: Principles of Programming Languages