Handouts

Language Specification &

Compiler Construction

Prof. Dr. Hanspeter Mssenbck
University of Linz

[email protected]












2008






http://ssw.jku.at/Misc/CC/




Course Contents

1. Overview
1.1 Motivation
1.2 Structure of a compiler
1.3 Grammars
1.4 Chomsky's classification of grammars
1.5 The MicroJ ava language

2. Scanning
2.1 Tasks of a scanner
2.2 Regular grammars and finite automata
2.3 Scanner implementation

3. Parsing
3.1 Context-free grammars and push-down automata
3.2 Recursive descent parsing
3.3 LL(1) property
3.4 Error handling

4. Semantic processing and attribute grammars

5. Symbol table
5.1 Overview
5.2 Objects
5.3 Scopes
5.4 Types
5.5 Universe

6. Code generation
6.1 Overview
6.2 The MicroJ ava VM
6.3 Code buffer
6.4 Items
6.5 Expressions
6.6 Assignments
6.7 J umps
6.8 Control structures
6.9 Methods

7. Building generators with Coco/R
7.1 Overview
7.2 Scanner specification
7.3 Parser specification
7.4 Error handling
7.5 LL(1) conflicts
7.6 Example


Main Literature

N. Wirth: Compiler Construction. Addison-Wesley 1996
A master of compiler constructions teaches how to write simple and efficient com-
pilers.
Also available under http://www.oberon.ethz.ch/WirthPubl/CBEAll.pdf
P.Terry: Compiling with C# and Java. Pearson Addison-Wesley 2005
A very good book that covers most of the topics of this course. It also describes
automatic compiler generation using the compiler generator Coco/R.
A.W.Appel: Modern Compiler Implementation in Java. Cambridge University Press
1998
Good and up-to-date book that treats the whole area of compiler construction in
depth.
H. Mssenbck: The Compiler Generator Coco/R. User Manual.
http://ssw.jku.at/Coco/Doc/UserManual.pdf


Further Reading

S.Muchnick: Advanced Compiler Design and Implementation. Morgan Kaufmann,
1997.
A very good and complete book which goes far beyond the scope of this introduc-
tory course. Not quite cheap but rewarding if you really want to become a compiler
expert.
H.Bal, D.Grune, C.Jacobs: Modern Compiler Design. John Wiley, 2000
Also a good books that describes the state of the art in compiler construction.
Aho, R. Sethi, J. Ullman: Compilers Principles, Techniques and Tools. Addison-
Wesley, 1986.
Old but still good to read. Does not cover recursive descent compilation in depth
but has chapters about optimisation and data flow analysis.
W.M.Waite, G.Goos: Compiler Construction. Springer-Verlag 1984
Theoretical book on compiler construction. Good chapter on attribute grammars.
Compiler Construction Lab


In this lab you will write a small compiler for a J ava-like language (MicroJava). You
will learn how to put the techniques from the compiler construction course into prac-
tice and study all the details involved in a real compiler implementation.
The project consists of four levels:
Level 1 requires you to implement a scanner and a parser for the language Micro-
J ava, specified in Appendix A of this document.
Level 2 deals with symbol table handling and type checking.
If you want to go to full length with your compiler you should also implement level
3, which deals with code generation for the MicroJ ava Virtual Machine specified in
Appendix B of this document. This level is (more or less) optional so that you can
get a good mark even if you do not implement it.
Level 4 finally requires you to use the compiler generator Coco/R to produce a
compiler-like program automatically.

The marking scheme will be as follows:
class test up to 45 points
project level 1 +20 points
project level 2 +20 points
project level 3 + 5 points
project level 4 +10 points
100 points

The project should be implemented in J ava using Sun Microsystem's J ava Develop-
ment Kit (J DK, http://java.sun.com/javase/downloads/index.jsp) or some other devel-
opment environment.
Level 1: Scanning and Parsing
In this part of the project you will implement a scanner and a recursive descent parser
for MicroJ ava. Start with the implementation of the scanner and do the following
steps:
1. Study the specification of MicroJ ava carefully (Appendix A). What are the tokens
of the MicroJ ava grammar? What is the format of names, numbers, character con-
stants and comments? What keywords and predeclared names do you need?
2. Create a package MJ and download the files Scanner.java and Token.java from
http://ssw.jku.at/Misc/CC/ into this package. Look at those files and try to under-
stand what they do.
3. Complete the skeleton file Scanner.java according to the slides of the course and
compile Token.java and Scanner.java.
4. Download the file TestScanner.java into the package MJ and compile it.
5. Download the MicroJ ava program sample.mj and run TestScanner on it.
Next you should write a recursive descent parser that uses your scanner to read the
input tokens. Do the following steps:
1. Download the file Parser.java into the package MJ and see what it does.
2. Complete the skeleton file Parser.java according to the slides of the course. Write
a recursive descent parsing method for every production of the MicroJ ava grammar
(see Appendix A). Compile Parser.java.
3. Download the file TestParser.java, compile it and run it on sample.mj. If your
parser is correct no errors should be reported.
4. Extend Parser.java with an error recovery according to the slides of the course.
Add synchronisation points at the beginning of statements and declarations.
5. Put some syntax errors into sample.mj and see if your parser detects them and if it
recovers correctly.
Level 2: Symbol Table Handling and Type Checking
Extend your parser with semantic processing. At every declaration the declared name
must be entered into the symbol table. When a name occurs in a statement it must be
looked up in the symbol table and the necessary context conditions must be checked
(see Appendix A.4). Do the following steps:
1. Create a new package MJ.SymTab (i.e. a package SymTab within the package MJ).
2. Download the files Obj.java, Struct.java, Scope.java and Tab.java into this pack-
age and see what they do.
3. Complete the skeleton file Tab.java (i.e. the symbol table) according to the slides
of the course.
4. Add semantic actions to Parser.java. These actions should enter every MicroJ ava
name into the symbol table when it is declared and retrieve it from the symbol ta-
ble when it is used. Try to check as many context conditions as you can from Ap-
pendix A.4 (some context conditions can only be checked during code generation).
5. Compile everything and run TestParser.java on sample.mj again to see if it works.
Insert some semantic errors into sample.mj. For example, use a few names without
declaring them or declare some names twice and see if your compiler detects those
errors.
6. In order to check whether you built the symbol table correctly you can call
Tab.dumpScope whenever you have processed all declarations of a scope in your
MicroJ ava program.
Level 3: Code Generation
The next task is to generate code for the MicroJ ava Virtual Machine. Before you start,
carefully study the specification of the VM (Appendix B) in order to become familiar
with the run-time data structures, the addressing modes, and the instructions. Then do
the following steps:
1. Create a new package MJ.CodeGen.
2. Download the files Code.java, Item.java and Decoder.java into this package.
3. Complete the skeleton file Code.java according to the slides of the course.
4. Add semantic actions to Parser.java. These actions should call the methods of
Code.java and Item.java as shown on the slides. Start with the actions for selectors
(e.g. obj.f and arr[i]), and continue with the semantic actions for expressions, as-
signments, if statements, while statements and method calls. Note that some con-
text conditions from Appendix A.4 have to be checked here as well.
5. Download the file Compiler.java into the package MJ. This is the main program of
your compiler that replaces TestParser.java. Compile it and run it on sample.mj.
This should produce a file sample.obj with the compiled program.
In order to run your compiled MicroJ ava programs download the file Run.java (i.e.
the MicroJ ava Virtual Machine) into the package MJ and compile it. You can invoke
it with
java MJ.Run sample.obj [-debug]
You can also decode a compiled MicroJ ava program by downloading the file De-
code.java to the package MJ and compiling it. You can invoke it with
java MJ.Decode sample.obj
Level 4: The Compiler Generator Coco/R
This task requires you to use the compiler generator Coco/R for building programs
that process structured input. It consists of two subtasks of which you have to imple-
ment at least one.
Task 1: Reading and Building a Binary Tree
Binary trees are dynamic data structures consisting of nodes, where every node has at
most 2 sons, which are again binary trees. Assume that we want to build the following
binary tree:
London
Paris
Madrid Rome
Vienna
Brussels
Data type for nodes
class Node {
String name;
Node left;
Node right;
}

We want to read the tree from an input file, which represents the tree structure with
brackets, i.e.:

(London
(Brussels)
(Paris
(Madrid)
(Rome
()
(Vienna)
)
)
)

Describe the input of such trees by a recursive EBNF grammar. Write a Coco/R com-
piler description using this grammar. Terminal symbols are identifiers as well as '('
and ')'. Add attributes and semantic actions to your compiler description in order to
build the corresponding binary tree. Write also a dump method that prints the tree
after it was built.
In order to use Coco/R go to http://ssw.jku.at/Coco/#J ava and download the files
Coco.jar, Scanner.frame and Parser.frame into a new directory Tree. If your compiler
description is in a file Tree.atg in the directory Tree go to this directory and type
java -jar Coco.jar Tree.atg
This will generate the files Scanner.java and Parser.java in the directory Tree. Write
a main program TreeBuilder.java that creates a scanner and a parser and calls the
parser (look at the slides in the course).
Task 2: Building a Phone Book
Assume that we have a text file with phone book entries. Every entry consists of a
person's name and one or more phone numbers for that person. A sample phone book
might look like this:
Boulder, J ohn M.
home 020 7815 1234
office 020 3465 234
Brown, Cynthia 1234567
Douglas, Ann Louise
office +44 (0)20 234 567
mobile +43 (0)664 7865 234
...
Names consist of letters and may be abbreviated with a dot.
Phone numbers consist of an optional country code (e.g. +44), an optional city code
(e.g. 020) and a phone number consisting of one or several digit sequences. Country
codes start with a '+' and must be followed by a city code (with a '0' in brackets).
City codes without country codes start with a '0'. If there is no country code the de-
fault is +44. If there is no city code the default is 020.
Phone numbers may be preceded by the words "home", "office" or "mobile". If
such a word is missing the default is "home".
Describe the syntax of such a phone book file by a grammar. Write a Coco/R com-
piler description that processes such input files by reading them and building a phone
book data structure in memory, where every entry of this data structure holds the fam-
ily name, the first names, the country code, the city code and the kind of phone num-
ber as separate fields. Write also a dump method that prints the whole phone book.
Appendix A. The MicroJava Language
This section describes the MicroJ ava language that is used in the practical part of the
compiler construction module. MicroJ ava is similar to J ava but much simpler.
A.1 General Characteristics
A MicroJ ava program consists of a single program file with static fields and static
methods. There are no external classes but only inner classes that can be used as da-
ta types.
The main method of a MicroJ ava program is always called main(). When a Micro-
J ava program is called this method is executed.
There are
- Constants of type int (e.g. 3) and char (e.g. 'x') but no string constants.
- Variables: all variables of the program are static.
- Primitive types: int, char (Ascii)
- Reference types: one-dimensional arrays like in J ava as well as classes with fields
but without methods.
- Static methods in the main class.
There is no garbage collector (allocated objects are only deallocated when the pro-
gram ends).
Predeclared procedures are ord, chr, len.
Sample program
program P
final int size = 10;

class Table {
int[] pos;
int[] neg;
}

Table val;

{
void main()
int x, i;
{ //---------- Initialize val ------------
val = new Table;
val.pos = new int[size];
val.neg = new int[size];
i = 0;
while (i < size) {
val.pos[i] = 0; val.neg[i] = 0;
i = i + 1;
}
//------------ Read values -------------
read(x);
while (x != 0) {
if (x >= 0) {
val.pos[x] = val.pos[x] + 1;
} else if (x < 0) {
val.neg[-x] = val.neg[-x] + 1;
}
read(x);
}
}
}
A.2 Syntax
Program = "program" ident {ConstDecl | VarDecl | ClassDecl}
"{" {MethodDecl} "}".
ConstDecl = "final" Type ident "=" (number | charConst) ";".
VarDecl = Type ident {"," ident } ";".
ClassDecl = "class" ident "{" {VarDecl} "}".
MethodDecl = (Type | "void") ident "(" [FormPars] ")" {VarDecl} Block.
FormPars = Type ident {"," Type ident}.
Type = ident ["[" "]"].
Block = "{" {Statement} "}".
Statement = Designator ("=" Expr | ActPars) ";"
| "if" "(" Condition ")" Statement ["else" Statement]
| "while" "(" Condition ")" Statement
| "return" [Expr] ";"
| "read" "(" Designator ")" ";"
| "print" "(" Expr ["," number] ")" ";"
| Block
| ";".
ActPars = "(" [ Expr {"," Expr} ] ")".
Condition = Expr Relop Expr.
Relop = "==" | "!=" | ">" | ">=" | "<" | "<=".
Expr = ["-"] Term {Addop Term}.
Term = Factor {Mulop Factor}.
Factor = Designator [ActPars]
| number
| charConst
| "new" ident ["[" Expr "]"]
| "(" Expr ")".
Designator = ident {"." ident | "[" Expr "]"}.
Addop = "+" | "-".
Mulop = "*" | "/" | "%".
Lexical structure
Character classes: letter = 'a'..'z' | 'A'..'Z'.
digit = '0'..'9'.
whiteSpace = ' ' | '\t' | '\r' | '\n'.
Terminal classes: ident = letter {letter | digit}.
number = digit {digit}.
charConst = "'" char "'". // including '\r', '\t', '\n'
Keywords: program class
if else while read print return
void final new
Operators: + - * / %
== != > >= < <=
( ) [ ] { }
= ; , .
Comments: // to the end of line
A.3 Semantics
All terms in this document that have a definition are underlined to emphasize their
special meaning. The definitions of these terms are given here.
Reference type
Arrays and classes are called reference types.
Type of a constant
The type of an integer constant (e.g. 17) is int.
The type of a character constant (e.g. 'x') is char.
Same type
Two types are the same
if they are denoted by the same type name, or
if both types are arrays and their element types are the same.
Type compatibility
Two types are compatible
if they are the same, or
if one of them is a reference type and the other is the type of null.
Assignment compatibility
A type src is assignment compatible with a type dst
if src and dst are the same, or
if dst is a reference type and src is the type of null.
Predeclared names
int the type of all integer values
char the type of all character values
null the null value of a class or array variable, meaning "pointing to no value"
chr standard method; chr(i) converts the int expression i into a char value
ord standard method; ord(ch) converts the char value ch into an int value
len standard method; len(a) returns the number of elements of the array a
Scope
A scope is the textual range of a method or a class. It extends from the point after the
declaring method or class name to the closing curly bracket of the method or class
declaration. A scope excludes other scopes that are nested within it. We assume that
there is an (artificial) outermost scope (called the universe), to which the main class is
local and which contains all predeclared names. The declaration of a name in an inner
scope hides the declarations of the same name in outer scopes.
Note
Indirectly recursive methods are not allowed, since every name must be declared
before it is used. This would not be possible if indirect recursion were allowed.
A predeclared name (e.g. int or char) can be redeclared in an inner scope (but this is
not recommended).
A.4 Context Conditions
General context conditions
Every name must be declared before it is used.
A name must not be declared twice in the same scope.
A program must contain a method named main. It must be declared as a void
method and must not have parameters.
Context conditions for standard methods
chr(e) e must be an expression of type int.
ord(c) c must be of type char.
len(a) a must be an array.
Context conditions for the MicroJ ava productions
Program = "program" ident {ConstDecl | VarDecl | ClassDecl} "{" {MethodDecl} "}".

ConstDecl = "final" Type ident "=" (number | charConst) ";".
The type of number or charConst must be the same as the type of Type.

VarDecl = Type ident {"," ident } ";".

ClassDecl = "class" ident "{" {VarDecl} "}".

MethodDecl = (Type | "void") ident "(" [FormPars] ")" {VarDecl} "{" {Statement} "}".
If a method is a function it must be left via a return statement (this is checked at run
time).

FormPars = Type ident {"," Type ident}.

Type = ident ["[" "]"].
ident must denote a type.

Statement = Designator "=" Expr ";".
Designator must denote a variable, an array element or an object field.
The type of Expr must be assignment compatible with the type of Designator.

Statement = Designator ActPars ";".
Designator must denote a method.

Statement = "read" "(" Designator ")" ";".
Designator must denote a variable, an array element or an object field.
Designator must be of type int or char.

Statement = "print" "(" Expr ["," number] ")" ";".
Expr must be of type int or char.

Statement = "return" [Expr] .
The type of Expr must be assignment compatible with the function type of the cur-
rent method.
If Expr is missing the current method must be declared as void.

Statement = "if" "(" Condition ")" Statement ["else" Statement]
| "while" "(" Condition ")" Statement
| "{" {Statement} "}"
| ";".

ActPars = "(" [ Expr {"," Expr} ] ")".
The numbers of actual and formal parameters must match.
The type of every actual parameter must be assignment compatible with the type of
every formal parameter at corresponding positions.

Condition = Expr Relop Expr.
The types of both expressions must be compatible.
Classes and arrays can only be checked for equality or inequality.

Expr = Term.

Expr = "-"Term.
Term must be of type int.

Expr = Expr Addop Term.
Expr and Term must be of type int.

Term = Factor.

Term = Term Mulop Factor.
Term and Factor must be of type int.

Factor = Designator | number | charConst| "(" Expr ")".

Factor = Designator ActPars.
Designator must denote a method.

Factor = "new" Type .
Type must denote a class.

Factor = "new" Type "[" Expr "]".
The type of Expr must be int.

Designator = Designator "." ident .
The type of Designator must be a class.
ident must be a field of Designator.

Designator = Designator "[" Expr "]".
The type of Designator must be an array.
The type of Expr must be int.

Relop = "==" | "!=" | ">" | ">=" | "<" | "<=".

Addop = "+" | "-".

Mulop = "*" | "/" | "%".
A.5 Implementation Restrictions
There must not be more than 127 local variables.
There must not be more than 32767 global variables.
A class must not have more than 32767 fields.
Appendix B. The MicroJava VM
This section describes the architecture of the MicroJ ava Virtual Machine that is used
in compiler lab. The MicroJ ava VM is similar to the J ava VM but has less and simpler
instructions. Whereas the J ava VM uses operand names from the constant pool that
are resolved by the loader, the MicroJ ava VM uses fixed operand addresses. J ava in-
structions encode the types of their operands so that a verifyer can check the consis-
tency of an object file. MicroJ ava instructions do not encode operand types.
B.1 Memory Layout
The memory areas of the MicroJ ava VM are as follows.

code
(byte array)
code
pc
data heap
free
ra
dl
pstack
fp
sp
estack
esp
data
(word array)
heap
(word array)
pstack
(word array)
estack
(word array)

code This area contains the code of the methods. The register pc contains the in-
dex of the currently executed instruction. mainpc contains the start address of
the method main().
data This area holds the (static or global) data of the main program. It is an array
of variables. Every variable holds a single word (32 bits). The addresses of
the variables are indexes into the array.
heap This area holds the dynamically allocated objects and arrays. The blocks are
allocated consecutively. free points to the beginning of the still unused area
of the heap. Dynamically allocated memory is only returned at the end of the
program. There is no garbage collector. All object fields hold a single word
(32 bits). Arrays of char elements are byte arrays. Their length is a multiple
of 4. Pointers are word offsets into the heap. Array objects start with an in-
visible word, containing the array length.
pstack This area (the procedure stack) maintains the activation frames of the in-
voked methods. Every frame consists of an array of local variables, each
holding a single word (32 bits). Their addresses are indexes into the array. ra
is the return address of the method, dl is the dynamic link (a pointer to the
frame of the caller). A newly allocated frame is initialized with all zeroes.
estack This area (the expression stack) is used to store the operands of the instruc-
tions. After every MicroJ ava statement estack is empty. Method parameters
are passed on the expression stack and are removed by the Enter instruction
of the invoked method. The expression stack is also used to pass the return
value of the method back to the caller.
All data (global variables, local variables, heap variables) are initialized with a null
value (0 for int, chr(0) for char, null for references).
B.2 Instruction Set
The following tables show the instructions of the MicroJ ava VM together with their
encoding and their behaviour. The third column of the tables show the contents of
estack before and after every instruction, for example
..., val, val
..., val
means that this instruction removes two words from estack and pushes a new word
onto it. The operands of the instructions have the following meaning:
b a byte
s a short int (16 bits)
w a word (32 bits)
Variables of type char are stored in the lowest byte of a word and are manipulated
with word instructions (e.g. load, store). Array elements of type char are stored in a
byte array and are loaded and stored with special instructions.
Loading and storing of local variables
1 load b ... Load
..., val push(local[b]);
2..5 load_n ... Load (n =0..3)
..., val push(local[n]);
6 store b ..., val Store
... local[b] =pop();
7..10 store_n ..., val Store (n =0..3)
... local[n] =pop();
Loading and storing of global variables
11 getstatic s ... Load static variable
..., val push(data[s]);
12 putstatic s ..., val Store static variable
... data[s] =pop();
Loading and storing of object fields
13 getfield s ..., adr Load object field
..., val adr =pop()/4; push(heap[adr+s]);
14 putfield s ..., adr, val Store object field
... val =pop(); adr =pop()/4;
heap[adr+s] =val;


Loading of constants
15..20 const_n ... Load constant (n =0..5)
..., val push(n);
21 const_m1 ... Load minus one
..., -1 push(-1);
22 const w ... Load constant
..., val push(w);
Arithmetic
23 add ..., val1, val2 Add
..., val1+val2 push(pop() +pop());
24 sub ..., val1, val2 Subtract
..., val1-val2 push(-pop() +pop());
25 mul ..., val1, val2 Multiply
..., val1*val2 push(pop() * pop());
26 div ..., val1, val2 Divide
..., val1/val2 x =pop(); push(pop() / x);
27 rem ..., val1, val2 Remainder
..., val1%val2 x =pop(); push(pop() % x);
28 neg ..., val Negate
..., - val push(-pop());
29 shl ..., val, x Shift left
..., val1 x =pop(); push(pop() <<x);
30 shr ..., val, x Shift right (arithmetically)
..., val1 x =pop(); push(pop() >>x);
Object creation
31 new s ... New object
..., adr allocate area of s words;
initialize area to all 0;
push(adr(area));
32 newarray b ..., n New array
..., adr n =pop();
if (b==0)
alloc. array with n elems of byte size;
else if (b==1)
alloc. array with n elems of word size;
initialize array to all 0;
push(adr(array))


Array access
33 aload ..., adr, i Load array element
..., val i =pop(); adr =pop();
push(heap[adr+1+i]);
34 astore ..., adr, i, val Store array element
... val =pop(); i =pop(); adr =pop();
heap[adr+1+i] =val;
35 baload ..., adr, i Load byte array element
..., val i =pop(); adr =pop();
x =heap[adr+1+i/4];
push(byte i%4 of x);
36 bastore ..., adr, i, val Store byte array element
... val =pop(); i =pop(); adr =pop();
x =heap[adr+1+i/4];
set byte i%4 in x;
heap[adr+1+i/4] =x;
37 arraylength ..., adr Get array length
..., len adr =pop();
push(heap[adr]);
Stack manipulation
38 pop ..., val Remove topmost stack element
... dummy =pop();
Jumps
39 jmp s J ump unconditionally
pc =s;
40..45 j<cond> s ..., x, y J ump conditionally (eq, ne, lt, le, gt, ge)
... y =pop(); x =pop();
if (x cond y) pc =s;


Method call (PUSH and POP work on pstack)
46 call s Call method
PUSH(pc+3); pc =s;
47 return Return
pc =POP();
48 enter b1, b2 Enter method
psize =b1; lsize =b2; // in words
PUSH(fp); fp =sp; sp =sp +lsize;
initialize frame to 0;
for (i=psize-1;i>=0;i--) local[i] =pop();
49 exit Exit method
sp =fp; fp =POP();
Input/Output
50 read ... Read
..., val readInt(x); push(x);
51 print ..., val, width Print
... width =pop(); writeInt(pop(), width);
52 bread ... Read byte
..., val readChar(ch); push(ch);
53 bprint ..., val, width Print byte
... width =pop(); writeChar(pop(), width);
Miscellaneous
54 trap b Generate run time error
print error message depending on b;
stop execution;
B.3 Object File Format
2 bytes: "MJ "
4 bytes: code size in bytes
4 bytes: number of words for the global data
4 bytes: mainPC: the address of main() relative to the beginning of the code area
n bytes: the code area (n =code size specified in the header)
B.4 Run Time Errors
1 Missing return statement in a function.