Elliot Berk Department of Computer Science, Princeton University
Version 1.2, May 5, 1997
Manual revision October 29, 1997
A lexical analyzer breaks an input stream of characters into tokens. Writing lexical analyzers by hand can be a tedious process, so software tools have been developed to ease this task.
Perhaps the best known such utility is Lex. Lex is a lexical analyzer generator for the UNIX operating system, targeted to the C programming language. Lex takes a specially-formatted specification file containing the details of a lexical analyzer. This tool then creates a C source file for the associated table-driven lexer.
The JLex utility is based upon the Lex lexical analyzer generator model. JLex takes a specification file similar to that accepted by Lex, then creates a Java source file for the corresponding lexical analyzer.
A JLex input file is organized into three sections,
separated by double-percent directives (``%%'').
A proper JLex specification has the following format.
user code
%%
JLex directives
%%
regular expression rules
The ``%%'' directives distinguish sections of the input
file and must be placed at the beginning of their line.
The remainder of the line containing the ``%%'' directives
may be discarded and should not be used to house
additional declarations or code.
The user code section - the first section of the specification file - is copied directly into the resulting output file. This area of the specification provides space for the implementation of utility classes or return types.
The JLex directives section is the second part of the input file. Here, macros definitions are given and state names are declared.
The third section contains the rules of lexical analysis, each of which consists of three parts: an optional state list, a regular expression, and an action.
User code precedes the first double-percent directive (``%%'). This code is copied verbatim into the lexical analyzer source file that JLex outputs, at the top of the file. Therefore, if the lexer source file needs to begin with a package declaration or with the importation of an external class, the user code section should begin with the corresponding declaration. This declaration will then be copied onto the top of the generated source file.
The JLex directive section begins after the first ``%%'' and continues until the second ``%%'' delimiter. Each JLex directive should be contained on a single line and should begin that line.
The %{...%} directive allows the user to write
Java code to be copied into the lexical analyzer class.
This directive is used as follows.
%{
<code>
%}
To be properly recognized, the %{ and %}
should each be situated at the beginning of a line.
The specified Java code in <code> will be then copied into
the lexical analyzer class created by JLex.
class Yylex {
... <code> ...
}
This permits the declaration of variables and functions
internal to the generated lexical analyzer class.
Variable names beginning with yy should be
avoided, as these are reserved for use by the generated
lexical analyzer class.
The %init{ ... %init} directive allows the user to write
Java code to be copied into the constructor for the
lexical analyzer class.
%init{
<code>
%init}
The %init{ and %init} directives
should be situated at the beginning of a line.
The specified Java code in <code> will be then copied into
the lexical analyzer class constructor.
class Yylex {
Yylex () {
... <code> ...
}
}
This directive permits one-time initializations
of the lexical analyzer class from inside its constructor.
Variable names beginning with yy should be
avoided, as these are reserved for use by the generated
lexical analyzer class.
The code given in the %init{ ... %init} directive
may potentially throw an exception, or propagate it from
another function. To declare this exception, use
the %initthrow{ ... %initthrow} directive.
%initthrow{
<exception[1]>[, <exception[2]>, ...]
%initthrow}
The Java code specified here will be copied
into the declaration of the lexical analyzer
constructor.
Yylex ()
throws <exception[1]>[, <exception[2]>, ...]
{
... <code> ...
}
If the Java code given in the %init{ ... %init}
directive throws an exception that is not declared,
the resulting lexical analyzer source file may not compile
successfully.
The %eof{ ... %eof} directive
allows the user to write Java code to be
copied into the lexical analyzer class
for execution after the end-of-file is reached.
%eof{
<code>
%eof}
The %eof{ and %eof} directives
should be situated at the beginning of a line.
The specified Java code in <code>
will be executed at most once, and immediately
after the end-of-file is reached for the input file
the lexical analyzer class is processing.
The code given in the %eof{ ... %eof} directive
may potentially throw an exception, or propagate it from
another function. To declare this exception, use
the %eofthrow{ ... %eofthrow} directive.
%eofthrow{
<exception[1]>[, <exception[2]>, ...]
%eofthrow}
The Java code specified here will be copied
into the declaration of the lexical analyzer function
called to clean-up upon reaching end-of-file.
private void yy_do_eof ()
throws <exception[1]>[, <exception[2]>, ...]
{
... <code> ...
}
The Java code in <code> that makes up
the body of this function will, in part,
come from the code given in the
%eof{ ... %eof} directive.
If this code throws an exception that is not declared
using the %eofthrow{ ... %eofthrow} directive,
the resulting lexer may not compile successfully.
Macro definitions are given in the JLex directives section
of the specification.
Each macro definition is contained on a single line and
consists of a macro name followed by an equal sign (=),
then by its associated definition.
The format can therefore be summarized as follows.
<name> = <definition>
Non-newline white space, e.g. blanks and tabs,
is optional between the macro name and the equal sign
and between the equal sign and the macro definition.
Each macro definition should be contained on a
single line.
Macro names should be valid identifiers, e.g. sequences of letters, digits, and underscores beginning with a letter or underscore.
Macro definitions should be valid regular expressions, the details of which are described in another section below.
Macro definitions can contain other macro expansions,
in the standard
{<name>} format for macros
within regular expressions.
However, the user should note that these expressions
are macros - not functions or nonterminals - so
mutually recursive constructs using macros are illegal.
Therefore, cycles in macro definitions will have
unpredictable results.
Lexical states are used to control when certain
regular expressions are matched.
These are declared in the JLex directives
in the following way.
%state state[0][, state[1], state[2], ...]
Each declaration of a series of lexical states should
be contained on a single line.
Multiple declarations can be included in the same
JLex specification, so the declaration of many states
can be broken into many declarations over multiple lines.
State names should be valid identifiers, e.g. sequences of letters, digits, and underscores beginning with a letter or underscore.
A single lexical state is implicitly declared by JLex. This state is called YYINITIAL, and the generated lexer begins lexical analysis in this state.
Rules of lexical analysis begin with an optional state list. If a state list is given, the lexical rule is matched only when the lexical analyzer is in one of the specified states. If a state list is not given, the lexical rule is matched when the lexical analyzer is in any state.
If a JLex specification does not make use of states, by neither declaring states nor preceding lexical rules with state lists, the resulting lexer will remain in state YYINITIAL throughout execution. Since lexical rules are not prefaced by state lists, these rules are matched in all existing states, including the implicitly declared state YYINITIAL. Therefore, everything works as expected if states are not used at all.
States are declared as constant integers within the generated lexical analyzer class. The constant integer declared for a declared state has the same name as that state. The user should be careful to avoid name conflict between state names and variables declared in the action portion of rules or elsewhere within the lexical analyzer class. A convenient convention would be to declare state names in all capitals, as a reminder that these identifiers effectively become constants.
Character counting is turned off by default, but can be activated
with the %char directive.
%char
The zero-based character index of the first character in
the matched region of text is then placed in the
integer variable yychar.
Line counting is turned off by default, but can be activated
with the %line directive.
%line
The zero-based line index at the beginning of the
matched region of text is then placed in the
integer variable yyline.
Java CUP is a parser generator for Java written
by Scott Hudson of Georgia Tech University.
Details of this software tool are on the World Wide Web
at
http://www.cs.princeton.edu/~appel/modern/java/CUP/.
Java CUP compatibility is turned off by default, but can
be activated with the following JLex directive.
%cup
At this point, the desired effect of this directive is undecided,
so no changes are made when it is given.
Note that features titled as Lexical Analyzer Component
Titles and Specifying the Return Value on End-of-File
may be useful when
using JLex with Java CUP.
The following directives can be used to change the name of
the generated lexical analyzer class, the tokenizing function, and
the token return type. To change the name of the lexical
analyzer class from Yylex, use the
%class directive.
%class <name>
To change the name of the tokenizing function from yylex,
use the %function directive.
%function <name>
To change the name of the return type from the tokenizing
function from Yytoken, use the %type
directive.
%type <name>
If the default names are not altering using these directives,
the tokenizing function is envoked with a call to
Yylex.yylex(), which returns the Ytoken type.
To avoid scoping conflicts, names beginning with yy are normally reserved for lexical analyzer internal functions and variables.
To make the 32-bit primitive integer type int,
the return type for the tokenizing function
(and therefore the token type),
use the %integer directive.
%integer
Under default settings, Yytoken is the return
type of the tokenizing function
Yylex.yylex(),
as in the following code fragment.
class Yylex { ...
public Yytoken yylex () {
... }
The %integer directive replaces the previous code
with a revised declaration, in which the token type
has been changed to int.
class Yylex { ...
public int yylex () {
... }
This declaration allows lexical actions to return
integer codes, as in the following code fragment
from a hypothetical lexical action.
{ ...
return 7;
... }
The integer return type forces changes the behavior at end of file. Under default settings, objects - subclasses of the java.lang.Object class - are returned by Yylex.yylex(). During execution of the generated lexer Yylex, a special object value must be reserved for end-of-file. Therefore, when the end-of-file is reached for the processed input file (and from then onward), Yylex.yylex() returns null.
When int is the return type of Yylex.yylex(),
null can no longer be returned. Instead,
Yylex.yylex() returns the value -1, corresponding
to constant integer
Yylex.YYEOF.
When %integer is given, this member of
Yylex is declared as follows.
public final int YYEOF = -1;
To make java.lang.Integer the return type for the
tokenizing function (and therefore the token type),
use the %intwrap directive.
%intwrap
Under default settings, Yytoken is the return
type of the tokenizing function
Yylex.yylex(),
as in the following code fragment.
class Yylex { ...
public Yytoken yylex () {
... }
The %intwrap directive replaces the previous code
with a revised declaration, in which the token type
has been changed to java.lang.Integer.
class Yylex { ...
public java.lang.Integer yylex () {
... }
This declaration allows lexical actions to return
wrapped integer codes, as in the following code fragment
from a hypothetical lexical action.
{ ...
return new java.lang.Integer(0);
... }
Notice that the effect of %intwrap directive can be
equivalently accomplished using the %type
directive, as follows.
%type java.lang.Integer
This manually changes the name of the return type
from Yylex.yylex() to
java.lang.Integer.
The %yyeof directive causes the constant
integer Yylex.YYEOF to be declared and
returned upon end-of-file.
%yyeof
This constant integer is discussed in more detail
in a previous section. Note also that Yylex.YYEOF
is a int, so the user should make sure
that this return value is compatible
with the return type of Yylex.yylex().
In UNIX operating systems, the character code sequence
representing a newline is the single character ``\n''.
Conversely, in DOS-based operating systems, the newline is
the two-character sequence ``\r\n''
consisting of the carriage return followed by the newline.
The %notunix directive results in either the carriage
return or the newline being recognized as a newline.
%notunix
This issue of recognizing the proper sequence of characters
as a newline is important in ensuring Java platform independence.
The default settings support an alphabet of character codes between 0 and 127 inclusive. If the generated lexical analyzer receives an input character code that falls outside of these bounds, the lexer may fail.
The %full directive can be used
to extend this alphabet to include
all 8-bit values.
%full
If the %full directive is given,
JLex will generate a lexical analyzer
that supports an alphabet of character codes
between 0 and 255 inclusive.
The following %unicode feature is unimplemented.
The %unicode can be used
to extend the alphabet to include the
full 16-bit Unicode alphabet.
%unicode
If the %unicode directive is given,
JLex will generate a lexical analyzer
that supports an alphabet of character codes
between 0 and 2^16-1 inclusive.
The %ignorecase directive can be given to generate
case-insensitive lexers.
%ignorecase
If the %ignorecase directive is given, CUP will expand all
character classes in a unicode-friendly way to match both upper,
lower, and title-case letters.
Under the status quo, JLex and the lexical analyzer it generates read from and write to Ascii text files, with byte sized characters. However, to support further extensions on the JLex tool, all internal processing of characters is done using the 16-bit Java character type, although the full range of 16-bit values is not supported.
The code given in the action portion of
the regular expression rules,
in section three of the JLex specification,
may potentially throw an exception, or propagate
it from another function.
To declare these exceptions, use
the %yylexthrow{ ... %yylexthrow} directive.
%yylexthrow{
<exception[1]>[, <exception[2]>, ...]
%yylexthrow}
The Java code specified here will be copied
into the declaration of the lexical analyzer
tokenizing function Yylex.yylex(), as follows.
public Yytoken yylex ()
throws <exception[1]>[, <exception[2]>, ...]
{
...
}
If the code given in the action portion of
the regular expression rules
throws an exception that is not declared
using the %yylexthrow{ ... %yylexthrow} directive,
the resulting lexer may not compile successfully.
The %eofval{ ... %eofval} directive
specifies the return value on end-of-file.
This directive allows the user to write Java code to be
copied into the lexical analyzer tokenizing
function Yylex.yylex()
for execution when the end-of-file is reached.
This code must return a value compatible with
the type of the tokenizing function Yylex.yylex().
%eofval{
<code>
%eofval}
The specified Java code in <code>
determines the return value of Yylex.yylex()
when the end-of-file is reached for the input file
the lexical analyzer class is processing.
This will also be the value returned by Yylex.yylex()
each additional time this function is called
after end-of-file is initially reached,
so <code> may be executed more than once.
Finally, the %eofval{ and %eofval} directives
should be situated at the beginning of a line.
An example of usage is given below.
Suppose the return value desired on end-of-file is
(new token(sym.EOF)) rather than
the default value null.
The user adds the following declaration to the
specification file.
%eofval{
return (new token(sym.EOF));
%eofval}
The code is then copied into Yylex.yylex()
into the appropriate place.
public Yytoken yylex () { ...
return (new token(sym.EOF));
... }
The value returned by Yylex.yylex() upon
end-of-file and from that point onward is now
(new token(sym.EOF)).
JLex allows the user to specify an interface which the Yylex
class will implement. By adding the following declaration to the input
file:
%implements <classname>
the user specifies that Yylex will implement classname. The
generated parser class declaration will look like:
class Yylex implements classname { ...
The %public directive causes the lexical analyzer class
generated by JLex to be a public class.
%public
The default behavior adds no access specifier to the generated
class, resulting in the class being visible only from the current
package.
The third part of the JLex specification consists of a series of rules for breaking the input stream into tokens. These rules specify regular expressions, then associate these expressions with actions consisting of Java source code.
The rules have three distinct parts:
the optional state list, the regular expression,
and the associated action.
This format is represented as follows.
[<states>] <expression> { <action> }
Each part of the rule is discussed in a section below.
If more than one rule matches strings from its input, the generated lexer resolves conflicts between rules by greedily choosing the rule that matches the longest string. If more than one rule matches strings of the same length, the lexer will choose the rule that is given first in the JLex specification. Therefore, rules appearing earlier in the specification are given a higher priority by the generated lexer.
The rules given in a JLex specification should match all possible input. If the generated lexical analyzer receives input that does not match any of its rules, an error will be raised.
Therefore, all input should be matched by at least one rule.
This can be guaranteed by placing the following rule
at the bottom of a JLex specification:
. { java.lang.System.out.println("Unmatched input: " + yytext());
}
The dot (.), as described below, will match any input
except for the newline.
An optional lexical state list preceeds each rule.
This list should be in the following form:
<state[0][, state[1], state[2], ...]>
The outer set of brackets ([]) indicate that multiple states are optional.
The greater than (<) and less than (>) symbols
represent themselves and should surround the state
list, preceding the regular expression.
The state list specifies under which initial states
the rule can be matched.
For instance, if yylex() is called with the lexer at state A, the lexer will attempt to match the input only against those rules that have A in their state list.
If no state list is specified for a given rule, the rule is matched against in all lexical states.
Regular expressions should not contain any white space, as white space is interpreted as the end of the current regular expression. There is one exception; if (non-newline) white space characters appear from within double quotes, these characters are taken to represent themselves. For instance, `` '' is interpreted as a blank space.
The alphabet for JLex is the Ascii character set, meaning character codes between 0 and 127 inclusive.
The following characters are metacharacters, with
special meanings in JLex regular expressions.
? * + | ( ) ^ $ / ; . = < > [ ] { } " \
ef Consecutive regular expressions represents their concatenation.
e|f The vertical bar (|) represents an option between the regular expressions that surround it, so matches either expression e or f.
The following escape sequences are recognized and expanded:
| \b | Backspace |
| \n | newline |
| \t | Tab |
| \f | Formfeed |
| \r | Carriage return |
| \ddd | The character code corresponding to the number formed by three octal digits ddd |
| \xdd | The character code corresponding to the number formed by two hexadecimal digits dd |
| \udddd | The Unicode character code corresponding to the number formed by four hexidecimal digits dddd. The support of Unicode escape sequences of this type is unimplemented. |
| \^C | Control character |
| \c | A backslash followed by any other character c matches itself |
. The dot (.) matches any character except the newline, so this expression is equivalent to [^\n].
"..." Metacharacters lose their meaning within
double quotes and represent themselves.
The sequence \" (which represents the
single character ") is the only exception.
{name} Curly braces denote a macro expansion, with name the declared name of the associated macro.
* The star (*) represents Kleene closure and matches zero or more repetitions of the preceding regular expression.
+ The plus (+) matches one or more repetitions of the preceding regular expression, so e+ is equivalent to ee*.
? The question mark (?) matches zero or one repetitions of the preceding regular expression.
(...) Parentheses are used for grouping within regular expressions.
[...] Square backets denote a class of characters and match any one character enclosed in the backets. If the first character following the left bracket ([) is the up arrow (^), the set is negated and the expression matches any character except those enclosed in the backets. Different metacharacter rules hold inside the backets, with the following expressions having special meanings:
| {name} | Macro expansion |
| a - b | Range of character codes from a to b to be included in character set |
| "..." | All metacharacters within double quotes lose
their special meanings. The sequence \" (which represents the
single character ") is the only exception. |
| \ | Metacharacter following backslash(\) loses its special meaning |
For example, [a-z] matches any lower-case letter, [^0-9] matches anything except a digit, and [0-9a-fA-F] matches any hexadecimal digit. Inside character class brackets, a metacharacter following a backslash loses its special meaning. Therefore, [\-\\] matches a dash or a backslash. Likewise ["A-Z"] matches one of the three characters A, dash, or Z. Leading and trailing dashes in a character class also lose their special meanings, so [+-] and [-+] do what you would expect them to (ie, match only '+' and '-').
The action associated with a lexical rule consists
of Java code enclosed inside block-delimiting curly braces.
{ action }
The Java code action is copied, as given, into
the state-driven lexical analyzer produced by JLex.
All curly braces contained in action not part of strings or comments should be balanced.
If no return value is returned in an action, the lexical analyzer will loop, searching for the next match from the input stream and returning the value associated with that match.
The lexical analyzer can be made to recur explicitly
with a call to yylex(), as in the following
code fragment.
{ ...
return yylex();
... }
This code fragment causes the lexical analyzer to recur,
searching for the next match in the input
and returning the value associated with that match.
The same effect can be had, however, by simply
not returning from a given action.
This results in the lexer searching for the next match,
without the additional overhead of recursion.
The preceding code fragment is an example of tail recursion,
since the recursive call comes at the end of
the calling function's execution.
The following code fragment is an example of a recursive
call that is not tail recursive.
{ ...
next = yylex();
... }
Recursive actions that are not tail-recursive work
in the expected way,
except that variables such as
yyline and yychar
may be changed during recursion.
If lexical states are declared in the JLex
directives section, transitions on these states
can be declared within the regular expression actions.
State transitions are made by the following
function call.
yybegin(state);
The void function yybegin() is passed the state
name state and effects a transition to
this lexical state.
The state state must be declared within the JLex directives section, or this call will result in a compiler error in the generated source file. The one exception to this declaration requirement is state YYINITIAL, the lexical state implicitly declared by JLex. The generated lexer begins lexical analysis in state YYINITIAL and remains in this state until a transition is made.
The following values, internal to the Yylex class, are available within the action portion of the lexical rules.
| Variable or Method | ActivationDirective | Description |
|---|---|---|
| java.lang.String yytext(); | Always active. | Matched portion of the character input stream. |
| int yychar; | %char | Zero-based character index of the first character in the matched portion of the input stream |
| int yyline; | %line | Zero-based line number of the start of the matched portion of the input stream |
JLex will take a properly-formed specification and transform it into a Java source file for the corresponding lexical analyzer.
The generated lexical analayzer resides in the class Yylex. There are two constructors to this class, both requiring a single argument: the input stream to be tokenized. The input stream may either be of type java.io.InputStream or java.io.Reader (such as StringReader).
The access function to the lexer is Yylex.yylex(),
which returns the next token from the input stream.
The return type is Yytoken and the function is declared
as follows.
class Yylex { ...
public Yytoken yylex () {
... }
The user must declare the type of Yytoken and
can accomplish this conveniently in the
first section of the JLex specification, the user
code section. For instance, to make Yylex.yylex()
return the java.lang.Integer integer wrapper type,
the user would enter the following code somewhere
preceding the first ``%%''.
class Yytoken extends java.lang.Integer { }
Then, in the lexical actions, wrapped integers would
be returned, in something like this way.
{ ...
return new java.lang.Integer(0);
... }
Likewise, in the user code section, a class could be defined
declaring constants that correspond to each of the token types.
class TokenCodes { ...
public static final STRING = 0;
public static final INTEGER = 1;
... }
Then, in the lexical actions, these token codes could be
returned.
{ ...
return new java.lang.Integer(STRING);
... }
These are simplified examples; in actual use, one would probably
define a token class containing more information than an integer
code.
These examples begin to illustrate the object-oriented
techniques a user could employ to define an arbitrarily
complex token type to be returned by Yylex.yylex().
In particular, inheritance permits the user to return more
than one token type. If a distinct token type was needed
for strings and integers, the user could make the
following declarations.
class Yytoken { ... }
class IntegerToken extends Yytoken { ... }
class StringToken extends Yytoken { ... }
Then the user could return both IntegerToken and
StringToken types from the lexical actions.
The names of the lexical analyzer class, the tokening function, and its return type each may be altered using the JLex directives.
A benchmark experiment was conducted, comparing the performance of a lexical analyzer generated by JLex to that of a hand-written lexical analyzer. The comparison was made for lexical analyzers of a simple ``toy'' programming language. The hand-written lexical analyzer, like the lexical analyzer generated by JLex, was written in Java.
The experiment consists of running each lexical analyzer on two source files written in the toy language, then measuring the time required to process these files. Each lexical analyzer was invoked by a dummy driver also written in Java.
The generated lexical analyzer proved to be quite quick, as the following results show.
| Size of Source File | JLex-Generated Lexical Analyzer: Execution Time | Hand-Written Lexical Analyzer: Execution Times |
|---|---|---|
| 177 lines | 0.42 seconds | 0.53 seconds |
| 897 lines | 0.98 seconds | 1.28 seconds |
The JLex lexical analyzer soundly outperformed the hand-written lexer.
One of the biggest complaints about table-driven lexical analyzers generated by programs like JLex is that these lexical analyzers do not perform as well as hand-written ones. Therefore, this experiment is particularly important in demonstrating the relative speed of JLex lexical analyzers.
The following is a (possibly incomplete) list of unimplemented features of JLex.
In contrast to the 8-bit character type (char) mandated by Ansi C, Java supports a 16-bit char and the Unicode character set. Java provides a built-in String class to manipulate these Unicode characters.
Along with the 16-bit char type, Java has an 8-bit primitive type (byte), as well as a host of utilities for manipulating character information in this 8-bit type.
As nearly all operating systems support the 8-bit Ascii character set, rather than the 16-bit Unicode character set, JLex assumes that text files - in particular, the JLex specification file and the lexical analyzer source file produced by JLex - are saved in Ascii format.
Commas between state names in declaration lists and lexical rules are optional. These lists will be correctly parsed with white space between state names and without comma separators.
The following minor features would be nice to have as part of JLex, but have not been implemented due to their scope or their negative impact upon performance.
The treatment of lexical analyzer generators given in Alan Holub's Compiler Design in C (Prentice-Hall, 1990) provided a starting point for my implementation.
Discussions with Professor Andrew Appel of the Princeton University Computer Science Department provided guidance in the design of JLex.
Java is a trademark of Sun Microsystems Incorporated.
JLex COPYRIGHT NOTICE, LICENSE AND DISCLAIMER.
Copyright 1996 by Elliot Joel Berk.
Permission to use, copy, modify, and distribute this software and its documentation for any purpose and without fee is hereby granted, provided that the above copyright notice appear in all copies and that both the copyright notice and this permission notice and warranty disclaimer appear in supporting documentation, and that the name of Elliot Joel Berk not be used in advertising or publicity pertaining to distribution of the software without specific, written prior permission.
Elliot Joel Berk disclaims all warranties with regard to this software, including
all implied warranties of merchantability and fitness. In no event
shall Elliot Joel Berk be liable for any special, indirect or consequential
damages or any damages whatsoever resulting from loss of use, data or
profits, whether in an action of contract, negligence or other
tortious action, arising out of or in connection with the use or
performance of this software.