CSE 12

Abstract Syntax Trees

Compilers and Interpreters
Parse Trees and Abstract Syntax Trees (AST's)
Creating and Evaluating AST's
The Table ADT and Symbol Tables

16

Using Algorithms and Data

Structures

Digital computers can do anything that can conceivably be

done with information: they are general purpose information
processing machines
You get a computer to do something by programming it to run
algorithms that manipulate data structures
Picking what algorithms and data structures are suitable for a
particular application is something every good programmer
needs to know
One important application area is: writing programs that
understand programs
2

Compilers and interpreters

Distinguish two kinds of program-understanding programs:

Compilers

take a program written in a source language as input and

translate it into a machine language, for later execution

Interpreters

take a source language program as input, and execute

("interpret") it immediately, line by line

(You can think of a computer as an interpreter of machine

language instructions; the Java Virtual Machine is an
interpreter of Java bytecode instructions, which are generated
by the Java compiler)
Lets look at how some ideas from CSE 12 (like recursion, and
trees) can be used to implement an interpreter for a language
3

Towards an interpreter

An interpreter reads a line or statement at a time of a source

language text, and then does what that line or statement says
to do
The source language should have clearly defined syntax and
semantics rules

the syntax rules say what strings legally belong to the

source language, and what their structure is
the semantics rules say what each legal string in the
language means, based on its structure

Without clearly defined syntax and semantics, writing an

interpreter isnt really possible!

Tasks for an interpreter

A language interpreter does two things with each line or

statement in the input: syntactic analysis and semantic evaluation
Syntactic analysis is done in accordance with the syntax rules
for the language

Semantic evaluation is done in accordance with the semantic

rules

the output from syntactic analysis is a tree data structure

corresponds to "compile time" in a compiled language

Semantic evaluation takes as input the tree created in the

syntactic analysis phase
corresponds to "run time" for compiled code

The result of all this is: doing what the statement says! (For
example, computing a value and returning it or printing it out)
5

Parts of an interpreter
In designing an interpreter, follow the decomposition into two
tasks, and design it to have two parts:
1. A syntactic analysis engine, which takes as input a string, and
outputs an appropriate tree structure
2. A semantic evaluation engine, which takes as input that tree,
and does what the original input string said to do

program
text

Syntax
analyzer

Semantic
evaluator

appropriate
action

Syntax rules and BNF

The syntax rules for the language can be written down in

Backus-Naur Form (BNF) or a similar notation

A BNF grammar is a list of syntax rules

Each rule defines one "nonterminal symbol", which appears at
the left of a " := " sign in the rule
Alternative definitions of the nonterminal symbol appear to the
right of the " := " sign, separated by " | " signs
Often the definition of a nonterminal in a BNF grammar is
recursive: it defines the nonterminal in terms of itself
7

Syntax rules and BNF

The nonterminal symbol defined in rule listed first in the

grammar is called the "start" symbol of the grammar
A symbol not defined by a rule in the grammar is a "terminal
symbol", and is usually taken literally
If a string satisfies the definition of the "start" symbol, it is in the
language defined by the BNF grammar; otherwise not

The process of using the grammar to check to see if a string is

in the language is called parsing the string

Parsing and derivations

The process of using the grammar to check to see if a string is in

the language defined by the grammar is called parsing the string

One way to parse a string is to try to write down a derivation of the

target string from the grammars start symbol:

1) Write down the start symbol. This is the first step in the derivation.
2) If the current step in the derivation consists of only terminal
symbols, and is equal to the target string, you have parsed the
string; done.

3) Else replace one of the nonterminal symbols in the current step of

the derivation with one of its definitions, to produce the next step
4) Go to 2
9

A BNF grammar for a language

Here is a BNF grammar for a simple formal language

The "start symbol for this grammar is <A>
If a string satisfies the definition of <A>, it is in the language
defined by this grammar; otherwise not

<A> := <B> | <C>

<B> := <ident> = <A>
<C> := <C> + <D> | <C> - <D> | <D>
<D> := <D> * <M> | <D> / <M> | <M>
<M> := <ident> | <const> | (<A>)
<ident> := w | x | y | z
<const> := 0 | 1 | 2 | 3 | 4
10

Example derivations
A derivation starts with the start symbol, and at each step, replaces
one nonterminal symbol by one of the definitions of that nonterminal.
Here are derivations of the strings 2 and 2 + w + 3 based on

that grammar:
<A> => <C> => <D> => <M> => <const> => 2

<A> => <C> => <C> + <D> => <C> + <M> =>
<C> + <const> => <C> + 3 =>
<C> + <D> + 3 => <D> + <D> + 3 =>
<M> + <D> + 3 => <M> + <M> + 3 =>
<const> + <M> + 3 => <const> + <ident> + 3 =>
2 + <ident> + 3 => 2 + w + 3
11

Parsing strings
Using that grammar, parse these strings of terminal
symbols that is, show a derivation of each of them from
the start symbol <A>:
w
2 + w
2 + w * 3
(2 + w) * 3

12

Parsing and parse trees

For every derivation, there is a corresponding tree: a parse tree

Each node in the parse tree corresponds to one symbol in the

BNF grammar
Leaves in the parse tree correspond to terminal symbols;
internal nodes correspond to nonterminal symbols
The root of the parse tree is the "start" symbol
The children of an internal node in the parse tree correspond to
the symbols in a definition of the nonterminal symbol
corresponding to their parent node
Reading the leaves of the parse tree left to right gives you the
string that has been parsed
13

Parse trees and derivations

How to build a parse tree?

Given a derivation of a string, you could build a parse tree top

down:

The start symbol is the root of the tree.

Children of the root are symbols in the definition of the start

symbol used to create the next step in the derivation.
In general, children of a node in the parse tree are symbols
in the definition of a nonterminal that was used to create a
next step in the derivation.

14

Parse tree facts

Note that a BNF parse must have a tree structure, because:

in a tree, each node except the root has exactly one parent
in a BNF grammar, each rule has exactly one nonterminal
symbol on the left hand side of the rule (this makes BNF
grammars what are called "context free" grammars)

A BNF grammar may permit several different derivations of the

same string, but if the grammar is unambiguous, a string will
have only one parse tree
A parse tree can also be built bottom up, creating leaves first,
and the root last. The algorithm we will consider later does that
15

Parse tree from a derivation

A derivation:

<A> => <C> => <D> => <M> => <const> => 2

Parse tree corresponding to that derivation:

<A>
<C>
<D>
<M>
<const>
2

16

Another parse tree from a derivation

Another derivation:
<A> => <C> => <C> - <D> => <C> - <M> => <C> - <const> =>
<C> - 3 => <C> + <D> - 3 => <D> + <D> - 3 =>
<M> + <D> - 3 => <M> + <M> - 3 => <const> + <M> - 3 =>
<const> + <ident> - 3 => 2 + <ident> - 3 => 2 + w - 3

Parse tree corresponding to that derivation:

<A>
<C>
<C>

<C>

- <D>
<D>
<M>

<D>

<M>

<const>

<M>

<ident>

<const>

2
17

Constructing parse trees

Using the example grammar, construct parse trees for

these strings:

2+w*3

(2 + w) * 3

w=2

18

Abstract syntax trees

A parse tree includes all the information in a derivation of a

string from the start symbol

In a parse tree, the children of a nonterminal symbol node

include nodes for all of the symbols in the definition of that
nonterminal

But this may be more information than is required for some

applications of the tree
For example: do not need to include nodes for parentheses just
used for grouping; information in nodes corresponding to
expression operators can be moved into their parent nodes;
etc.
A parse tree with nonessential nodes left out is called an
Abstract Syntax Tree (AST)
19

Parse tree and AST

Parse tree and an AST for the string

1 ( 2 + 3 )

<A>

<A>

<C>
<C>

<D>

<M>

<D>
(

<M>
<const>

<C>

<A>

<C>

<C>

<D>

<D>

<M>

<M>

<const>

<const>

<C>
-

<D>

<D>

<M>

<M>

<A>

<const>
1

<C>
+

<C>

<D>

<D>

<M>

<M>

<const>
3

<const>
2

20

Syntactic analysis == building an AST

The input to the syntactic analysis phase of an interpreter

is a string; the output is an AST
So the process of syntactic analysis is the process of
constructing an AST
There are two main issues to address before
implementing syntactic analysis in this way:

the design of the AST data structure: what type of

objects to use for the nodes, etc.

what algorithm to use to build the AST

21

AST data structure design issues

You know how to implement binary trees...
... but ASTs raise some additional issues:

In general, a node in an AST can have more than 2 children

(because more than 2 symbols can appear on the right hand
side of a rule in the grammar)
Intuitively, the nodes of an AST are of somewhat different
types... a <const> node has a different meaning from a <ident>
node, for example

At the same time, the nodes of an AST are of similar type: they
all correspond to part of an input string at some level of
analysis

22

Applying OO design to AST design

In an object-oriented approach, you can easily deal with these
design issues:

Define a separate class corresponding to each different

nonterminal symbol in the grammar: nodes in the AST will be
instances of these classes

lets you make different types of nodes have different

possible numbers of children (different number of
private instance variables), and different behavior
(different methods)

Make all these classes derived from one base class, or make
them all implement one interface

permits polymorphism
23

Abstract syntax tree algorithm design

The easiest way to implement AST construction is to closely
follow the BNF grammar definitions

Each class in the OO design should have a method that

"knows how" to parse a string according to the definition of the
nonterminal symbol corresponding to that class

"Knowing how" to parse a string will rely on parse methods

in other classes, that know how to parse strings according
to the definitions of their corresponding symbols!

If the method successfully parses the string, it should return a

pointer to an AST node of the appropriate subtype, with its
children initialized in the appropriate way
If the method cannot successfully parse the string, it should
return a null pointer to indicate that fact
24

A piece of an AST construction algorithm

Following this basic design idea, the parse method for nonterminal
symbol <A> defined as <A> := <B> | <C> would be a static factory
method in a class named A , and could have a structure like this:
public static A parse (String s) {
if (B.parse(s) != null) {
// the string satisfies defn of <B>; so,

// make a new A node, with the result of B.parse(s)

// as its child, and return it
} else if (C.parse(s) != null) {
// the string satisfies defn of <C>; so,
// make a new A node, with the result of C.parse(s)
// as its child, and return it
} else // the string does not satisfy defn of <A> !
return null;
}
25

Semantic rules

An interpreter first does syntactic analysis, constructing an AST...

...then it does semantic evaluation, according to the semantic
rules for the language
To implement semantic evaluation correctly, semantic rules must
be clearly stated and followed
Unfortunately there is no universally accepted notation for
specifying semantic rules that is as standard as BNF notation is
for specifying syntax rules
Usually, semantic rules are stated as English or pseudocode,
attached to each BNF rule (or equivalently, to each type of node
in the AST)
26

Semantic rules: some examples

<A> := <B> | <C>
The value of <A> is the value of <B> or <C>, as appropriate.
<B> := <ident> = <A>
The value of <B> is the value of the <A> on the rhs of the =. In addition,
evaluating a <B> has the side effect of assigning the value of <A> to the
variable named by the <ident>.
<C> := <C> + <D>
The value of <C> is the value of <C> plus the value of <D>.
<D> := <D> * <M>
The value of <D> is the value of <D> times the value of <M>.
<ident> := w | x | y | z
The value of <ident> is the value of the variable named by the identifier. If the
variable has not been assigned a value, it is a runtime error.
<const> := 0 | 1 | 2 | 3 | 4
The value of <const> is the value of the matching literal constant.
27

Evaluating an AST
Evaluate these ASTs ( for 1 ( 2 + 3 ) and

<C>

<A>

<A>

<C>
-

<C>
+

<D>

<D>

<M>

<C>

<M>

<A>

<const>
1

<C>
+

<C>

<D>

<D>

<M>

<M>

<const>
3

<const>
2

<C>
-

1 2 + 3

<D>

<D>

<M>

<D>

<M>

<M>

<const>
2

<const>
3

<const>
1

28

Evaluating AST's
Construct ASTs for the following expressions, and then evaluate
them:
2 + 3 * 4

w = 2

2 + w * 3

29

Evaluating AST's

Note that the semantic evaluation phase basically does a postorder

traversal of the syntax tree: evaluate the roots children left to right,
then return the value of the root
Note also that the precedence and associativity rules for operators in
this language never had to be explicitly stated: they are implicit in the
way this grammar is defined

left-recursive rule gives left-to-right associativity, rightrecursive rule gives right-to-left associativity
symbols farther from the root (the start symbol) are
evaluated first, giving higher precedence

Overall, semantic evaluation is very straightforward, once the AST

has been constructed

One additional issue: How to keep track of the values of variables?

... answer: use a symbol table, an application of the Map ADT

30

Map

The ADTs we have talked about so far have been container

structures intended to hold data, sometimes called keys

An insert operation inserts a single key value in the

structure; a find operation says whether a key value is in the
structure; delete removes a key; etc.

The Map ADT has a slightly different emphasis

31

Map ADT

A Map ADT is intended to hold pairs : each pair consists of a

key, together with a related data value
An insert operation inserts a key-value pair in the table
A find operation takes a key and returns the value in the keyvalue pair with that key
A delete operation takes a key and removes the key-value pair
with that key
The Map ADT is also sometimes called a Table" or
"Dictionary" ADT, or an "associative memory"

in the JCF, see the Map<K,V> interface

32

Map ADT

Domain:

a collection of pairs; each pair consists of a key, and

related data value

Operations (typical):

Create a Map (initially empty)

Insert a new key-value pair in the Map; if a key-value

pair with the same key is already there, update the
value part of the pair
Find the key-value pair in the Map corresponding to a
given key; return the value, or null if none
Delete the key-value pair corresponding to a given
key

Traverse all key-value pairs in the Map

33

Implementing the Map ADT

A Map can be implemented in various ways

In each case:

using a list, binary search tree, hashtable (well talk

about these later), etc., etc.
the implementing data structure has to be able to hold
key-value pairs
the implementing data structure has to be able to do
insert, find, and delete operations depending on the
key, but storing both the key and its data value

the find operation should return the value associated

with a given key

Using Java generics, you can specify the type of the keys, and
the type of the values associated with keys
34

The Map ADT and symbol tables

The Map ADT is useful in any situation where you want to

store, retrieve, and manipulate data based on keys associated
with pieces of data
One important application is a symbol table in a programming
language compiler or interpreter

A symbol table associates identifiers in a program with related

data such as whether the identifier is the name of a variable,
function, constant, or class; and if it is a variable, what its value
is, etc.

35

Symbol table operations

Typical symbol table operations include:

When a variable is declared, an "insert" operation is

done in the table to add an entry saying what type the
variable is, and what its initial value is, etc.
When a variable is referenced, a "find" operation is
done in the symbol table, to see what the variables
value is at the present time
When a variable is assigned to, an "insert" or "update"
operation is done to change the value associated with
the variable
When a variable passes out of scope, a "delete"
operation is done to remove it and its information from
the table
36

Symbol tables and side effects

Some expression languages may permit only "pure evaluation":

evaluating a node in the parse tree involves no side effects
However, if the language contains variables and permits
assignment of values to variables, the interpreter must correctly
handle these side effects, and a symbol table can do that
The symbol table maintains associations between variables (as
keys) and their values
In a simple expression language, variables are the only
symbols, and once introduced they do not go out of scope. So
the only symbol table operations needed are:

When a variable is evaluated, the symbol table is accessed to find

the value associated with the variable
When an assignment expression is evaluated, the symbol table is
updated to associate the new assigned value with the variable
37

Next time

The Map ADT

Implementations of the Map ADT
Hashing and Hash Tables
Collisions and Collision Resolution Strategies
Hash Functions
Hash Table time costs

Reading: Gray, Ch 12

38