240 likes | 435 Views
Attribute Grammars. Prabhaker Mateti ACK: Assembled from many sources. About Attribute Grammars. Attribute grammars (AGs) add semantic info on parse tree nodes Used for semantic checking and other compile-time analyses, e.g., type checking in a compiler
E N D
Attribute Grammars PrabhakerMateti ACK: Assembled from many sources
About Attribute Grammars • Attribute grammars (AGs) add semantic info on parse tree nodes • Used for semantic checking and other compile-time analyses, e.g., type checking in a compiler • Used for translation, e.g., parse tree to assembly code • A traversal of the parse tree and the computation of information.
Attribute Grammars: Definition • An attribute grammar is a context-free grammar G = (S, N, T, P) with the following additions. • For each terminal and non-terminal X, disjoint sets S(X) synthesized attributes and I(X) inherited attributes: • A(X) = S(X) ∪ I(X) • X0 ::= X1 ... Xn • S(X0) = f(I(X0), A(X1), ... , A(Xn)) • I(Xj) = g(A(X0), ... , A(Xn)) • depends on the attribute value at parent and those of siblings. • Each rule has a set of predicates/ conditions to check for attribute consistency: P( A(X0), A(X1), A(X2), …, A(Xn) )
Example-1: Binary Numbers • N ::= 0 • N ::= 1 • N ::= N 0 • N ::= N 1 • CFGs do not provide semantics. • CGGs provide only syntax, that too without context-sensitive details • N.val:= 0 • N.val:= 1 • N.val:= 2*N.rhs.val • N.val:= 2*N.rhs.val+1 • N.val is an attribute associated with node N of the parse tree • N.rhs Node corresponding to the rhs • Synthesized Attributes
Example-2: Type Checking • E ::= n • E ::= x • E ::= E1 + E2 • E ::= E1 * E2 • Semantics we wish to add • n is an int, x is a real. • op + returns an int if both the operands are int, otherwise a real. • E.type:= int • E.type := real • if E1.type = E2.type then E.type := E1.type else E.type:= realfi • Item 3 derived version of the semantics • Static semantics
Example-3: Assignment Arithmetic • stm ::= var := exp | stm; stm • exp ::= var + var | var • var ::= A | B | C • synthesized actual-type for var and exp • inherited expected-type for exp • lookup (var.string) a helper function; gives the actual type of A, B, C • exp ::= var1 + var2 • subscripts added • exp.actual-type:= var1.actual-type • exp.expected-type • from parent in the parse tree • Predicates: • var1.actual-type== var2.actual-type • exp.expected-type== exp.actual-type • var ::= A | B | C • var.actual-type:= lookup (var.string)
Inherited Attributes Example • Declaration and Use{ int i, j, k; i := i + j + j; } • assign ::= var := exp • env: environment • var.env := assign.env • exp.env := assign.env
Information Flow inherited computed available synthesized ... ...
Attribute Value Computation • If all attributes were inherited, the tree could be decorated in top-down order. Inherited Attributes pass information • down the parse tree, or • from left siblings to the right siblings • If all attributes were synthesized, the tree could be decorated in bottom-up order. Synthesized Attributes pass information up the parse tree • In many cases, both kinds of attributes are used, and it is some combination of top-down and bottom-up that must be used. • Initially, there are intrinsic attributes on the leaves • If a condition in a tree evaluates to false, an error occurs.
Prolog for Example-2 • type(n, int). • type(x, real). • type(+(E, F), T) :- type(E, T), type(F, T). • type(+(E, F), real) :- type(E, T1), type(F, T2), T1 \= T2. • Type Checking ?- type(+(n, x), real). • Type Inference ?- type(+(n, x), T). • (Definite Clause Grammars)
F ::= . N N ::= 0 N ::= 1 N ::= 0 N N ::= 1 N Synthesized: val, value Inherited: pow, the number of bits between left of a non-terminal and the binary point Nr: N-right, Nl: N-left 1: F.val:= N.val; N.pow:= 1 2: N.val := 0 3: N.val := (1/2^N.pow) 4: Nl.val := Nr.val 4: Nr.pow := 1 + Nl.pow 5: Nl.Val := Nr.val+(1/2^N.pow) 5: Nr.pow := 1 + Nl.pow Example-4: Fractions in Binary
Binary Fractions, same grammar as before: F ::= . N N ::= 0 N ::= 1 N ::= 0 N N ::= 1 N Alternate computation based on synthesized attribute val only 1: F.val := N.val / 2 2: N.val := 0 3: N.val := 1 4: N.val := N.val / 2 5: N.val := N.val / 2 + 1 Ex4: Synthesized Attributes Only
Example-5: Distinct Identifiers • Compute the number of distinct identifiers in a straight-line program. • Semantics specified in terms of sets of identifiers. • Attributes • var id • exp ids • stm ids num
Example-5: Distinct Identifiers • exp::= var • exp.ids = { var.id } • exp::= exp1 + exp2 • exp.ids = exp1.ids U exp2.ids • stm::= var:= exp • stm.ids = { var.id } U exp.ids • stm.num = | stm.ids | • stm::= stm1;stm2 • stm.ids = stm1.ids U stm2.ids • stm.num = | stm.ids |
Example-5: Using Lists • Attributes • envi: list of vars in preceding context • envo: list of vars for following context • dnum: number of new variables • exp ::= var exp.envo = if member(var.id, exp.envi) then exp.envi else cons(var.id, exp.envi) fi
Example-5: Using Lists • exp ::= exp1 + exp2 • envienvienvi • envoenvoenvo • dnumdnumdnum • exp1.envi := exp.envi • exp2.envi := exp1.envo • exp.envo := exp2.envo • exp.dnum := length(exp.envo) • exp.envo = append-sans-duplicates(exp1.envo, exp2.envo )
Complete Evaluation Rules • Synthesized attribute associated with N: • Each alternative in “N ::= …” should contain a rule for evaluating the Synthesized attribute. • Inherited attribute associated with N: • For every occurrence of N in “… ::= … N …” there must be a rule for evaluating the Inherited attribute. • Whenever you create an attribute grammar (in home work/ exams), make sure it satisfies these requirements.
One Pass Attribute Computation • To enable one-pass top-down left-to-right computation of the attributes: • each inherited attribute of the right-hand side symbol can depend on all the attributes associated with preceding right-hand side symbols and the inherited attribute of the left-hand side non-terminal. • Similarly, the synthesized attribute of the left-hand side non-terminal can depend on all the attributes associated with all the right-hand side symbols and the inherited attribute of the left-hand side non-terminal.
More than Context-Free Power • LABC = { a^nb^nc^n | n > 0 } • Unlike LAB = { a^nb^n | n > 0 }, here we need explicit counting of a’s, b’s and c’s • LWCW = { wcw | w ∈{a, b}* } • The “flavor” of checking whether identifiers are declared before their uses • LABC, LWCW cannot be defined with a context-free grammar • Syntax analysis (i.e., parser based on CFGs) cannot handle semantic properties
LABC = { a^nb^nc^n | n > 0 } • ls ::= as bscs • ExpNb(bs) := Na(as); ExpNc(cs) := Na(as) • as ::= a | a as1 • Na(as) := 1; Na(as) := Na(as1) + 1 • bs ::= b | b bs1 • cond(ExpNb(bs) = 1); ExpNb(bs1) := ExpNb(bs) - 1 • cs ::= c | c cs1 • Cond(ExpNc(cs) = 1); ExpNc(cs1) := ExpNc(cs) – 1 • Na: synthesized by as • ExpNb: inherited from bs • ExpNc: inherited from cs
Uses of Attribute Grammars • Compiler Generation • Top-down Parsers (LL(1)) • FIRST sets, FOLLOW sets, etc • Code Generation Computations • Type, Storage determination, etc. • Databases • Optimizing Bottom-up Query Evaluation (Magic Sets) • Programming and Definitions
Uses of Inherited Attributes • ex: 5.0 + 2 • need to generate code to coerce int 2 to real 2.0 • Determination of un-initialized variables • Determination of reachable non-terminals • Evaluation of an expression containing variables
Use of Attribute Grammars • Useful for expressing arbitrary cycle-free computational walks over CFG derivation trees • Synthesized and inherited attributes • Conditions to reject invalid parse trees • Evaluation order depends on attribute dependencies • Realistic applications: • type checking • code generation • “Global” data structures must be passed around as attributes • Any container data structure (sets, etc.) can be used • The evaluation rules can call auxiliary/helper functions but the functions cannot have side effects
References • T. K. Prasad, Attribute Grammars and their Applications, In: Encyclopedia of Information Science and Technology, pp. 268-273, 2008. Attribute-Grammars.pdf • PL TextBookSections • Pagan: 2.1, 2.2, 2.3, 3.2 • Stansifer: 2.2, 2.3 • Slonneger and Kurtz: 3.1, 3.2