UNIVERSITY OF CALIFORNIA

Department of Electrical Engineering

and Computer Sciences
Computer Science Division

CS61B P. N. Hilfinger
Fall 2008

Basic Compilation: javac and gmake

[The discussion in this section applies to Java 1.5 tools from Sun Microsystems. Tools from other
manufacturers and earlier tools from Sun differ in various details.]
Programming languages do not exist in a vacuum; any actual programming done in any language
one does within a programming environment that comprises the various programs, libraries, editors,
debuggers, and other tools needed to actually convert program text into action. This document
discusses the tools that translate programs into executable form and then execute them.

1 Compilation and Interpretation

The Scheme environment that you used in CS61A was particularly simple. It provided a component
called the reader, which read in Scheme-program text from files or command lines and converted it
into internal Scheme data structures. Then a component called the interpreter operated on these
translated programs or statements, performing the actions they denoted. You probably weren’t much
aware of the reader; it doesn’t amount to much because of Scheme’s very simple syntax.
Java’s more complex syntax and its static type structure (as discussed in lecture) require that
you be a bit more aware of the reader—or compiler, as it is called in the context of Java and most
other “production” programming languages. The Java compiler supplied by Sun Microsystems is a
program called javac on our systems. You first prepare programs in files (called source files) using
any appropriate text editor (Emacs, for example), giving them names that end in ‘.java’. Next you
compile them with the java compiler to create new, translated files, called class files, one for each
class, with names ending in ‘.class’. Once programs are translated into class files, there is a variety
of tools for actually executing them, including Sun’s java interpreter (called ‘java’ on our systems),
and interpreters built into products such as Netscape or Internet Explorer. The same class file format
works (or is supposed to) on all of these.
In the simplest case, if the class containing your main program or applet is called C, then you
should store it in a file called C.java, and you can compile it with the command

javac C.java

This will produce .class files for C and for any other classes that had to be compiled because they
were mentioned (directly or indirectly) in class C. For homework problems, this is often all you need
to know, and you can stop reading. However, things rapidly get complicated when a program consists
of multiple classes, especially when they occur in multiple packages. In this document, we’ll try to
deal with the more straightforward of these complications.

1
2 P. N. Hilfinger

2 Where ‘java’ and ‘javac’ find classes

Every Java class resides in a package (a collection of classes and subpackages). For example, the
standard class String is actually java.lang.String: the class named String that resides in the
subpackage named lang that resides in the outer-level package named java. You use a package
declaration at the beginning of a .java source file to indicate what package it is supposed to be in.
In the absence of such a declaration, the classes produced from the source file go into the anonymous
package, which you can think of as holding all the outer-level packages (such as java).

2.1 The interpreter’s classes

When the java program (the interpreter) runs the main procedure in a class, and that main procedure
uses some other classes, let’s say A and p.B, the interpreter looks for files A.class and B.class in
places that are dictated by things called class paths. Essentially, a class path is a list of directories
and archives (see §5 below for information on archives). If the interpreter’s class path contains,
let’s say, the directories D1 and D2 , then upon encountering a mention of class A, java will look
for a file named D1 /A.class or D2 /A.class. Upon encountering a mention of p.B, it will look for
D1 /p/B.class or D2 /p/B.class.
The class path is cobbled together from several sources. All Sun’s java tools automatically supply
a bootstrap class path, containing the standard libraries and such stuff. If you take no other steps,
the only other item on the class path will be the directory ‘.’ (the current directory). Otherwise, if
the environment variable CLASSPATH is set, it gets added to the bootstrap class path. In past years,
our standard class setup had ‘.’ and the directory containing the ucb package (with our own special
classes, lovingly concocted just for you), which we set up with the command

setenv CLASSPATH .:/home/ff/cs61b/lib/java/classes

(the colon is used in place of comma (for some reason) to separate directory names). The interpreter
and compiler would then find the definition of a class such as ucb.io.StdIO in

/home/ff/cs61b/lib/java/classes/ucb/io/StdIO.class

These days, we use archive files instead, as described below in §5.

2.2 The compiler’s classes

The compiler looks in the same places for .class files, but its life is more complicated, because it
also has to find source files. By default, when it needs to find the definition of a class A, it looks for
file A.java in the same directories it looks for A.class. This is the easiest case to deal with. If it
does not find A.class, it will automatically compile A.java. To use this default behavior, simply
make sure that the current directory (‘.’) is in your class path (as it is in our default setup) and put
the source for a class A (in the anonymous package) in A.java in the current directory, or for a class
p.B in p/B.java, etc., using the commands

javac A.java
javac p/A.java

respectively, to compile them.

It is also possible to put source files, input class files, and output class files (i.e., those created by
the compiler) in three different directories, if you really want to (I don’t think we’ll need this). See
the -sourcepath and -d options in the on-line documentation for javac, if you are curious.
Basic Compilation: javac and gmake 3

3 Multiple classes in one source file

In general, you should try to put a class named A in a file named A.java (in the appropriate
directory). For one thing, this makes it possible for the compiler to find the class’s definition. On the
other hand, although public classes must go into files named in this way, other classes don’t really
need to. If you have a non-public class that really is used only by class A, then you can put it, too,
into A.java. The compiler will still generate a separate .class file for it.

4 Compiling multiple files

Java source files depend on each other; that is, the text of one file will refer to definitions in other
files. As I said earlier, if you put these source files in the right places, the compiler often will
automatically compile all that are needed even if it is only actually asked to compile one “root” class
(the one containing the main program or main applet). However, it is possible for the compiler to
get confused when (a) some .java files have already been compiled into .class files, and then (b)
subsequently changed. Sometimes the compiler will recompile all the necessary files (that is, the ones
whose source files have changed or that use classes whose source files have changed), but it is a bit
dangerous to rely on this for the Sun compiler. The compiler also can’t find class definitions if you
“hide” them by putting, say, several classes into one file. The compiler guesses that class A.B is in
file A/B.java. If it isn’t, then it gives up. You can avoid both of these problems by asking listing all
the necessary files for javac explicitly:

javac A.java p/B.java root.java

Since this is tedious to write, it is best to rely on a makefile to do it for you, as described below in
§6.

5 Archive files
For the purposes of this course, it will be sufficient to have separate .class files in appropriate
directories, as I have been describing. However in real life, when one’s application consists of large
numbers of .class files scattered throughout a bunch of directories, it becomes awkward to ship
it elsewhere (say to someone attempting to run your Web applet remotely). Therefore, it is also
possible to bundle together a bunch of .class files into a single file called a Java archive (or jar file).
You can put the name of a jar file as one member of a class path (instead of a directory), and all its
member classes will be available just as if they were unpacked into the directory structure described
in previous sections.
The utility program ‘jar’, provided by Sun, can create or examine jar files. Typical usage: to
form a jar file stuff.jar out of all the classes in package myPackage, plus the files A.class and
B.class, use the command

jar cvf stuff.jar A.class B.class myPackage

This assumes that myPackage is a subdirectory containing just .class files in package myPackage.
To use this bundle of classes, you might set your class path like this:

setenv CLASSPATH .:stuff.jar:other directories and archives

4 P. N. Hilfinger

6 The make utility

Even relatively small software systems can require rather involved, or at least tedious, sequences
of instructions to translate them from source to executable forms. Furthermore, since translation
takes time (more than it should) and systems generally come in separately translatable parts, it
is desirable to save time by updating only those portions whose source has changed since the last
compilation. However, keeping track of and using such information is itself a tedious and error-prone
task, if done by hand. Therefore, most programming environments provide some kind of project or
compilation-control facility. The UNIX make utility is a conceptually simple and general example.
It accepts as input a description of the interdependencies of a set of source files and the commands
necessary to compile them, known as a makefile; it examines the ages of the appropriate files; and it
executes whatever commands are necessary, according to the description. For further convenience,
it will supply certain standard actions and dependencies by default, making it unnecessary to state
them explicitly.
There are numerous dialects of make, both among UNIX installations and (under other names)
in programming environments for personal computers. In this course, I will use a version known as
gmake1 . Though conceptually simple, the gmake utility has accreted features with age and use, and
is rather imposing in the glory of its full definition. This document describes only the simple use of
gmake.
In addition to compilation (or re-compilation) control, there are other uses for gmake. It is
useful in cases where one needs some kind of preprocessing: where the Java source files themselves
result from applying some program to other inputs. The makefiles themselves also serve as a useful
repository for scripts that perform numerous tasks incidental to compilation. I use them to build
course material and copy it to where others can get to it.

6.1 Basic Operation and Syntax

Figure 1 is a sample makefile for compiling a simple editor program, edit, from eight .java files.
This file consists five rules. A rule consists of a line containing two lists of names separated by a
colon, followed by one or more lines beginning with tab characters. Any line may be continued, as
illustrated, by putting a backslash at the very end, which essentially acts like a space, combining the
line with its successor. The ‘#’ character indicates the start of a comment that goes to the end of
that line.
The names preceding the colons are known as targets; they are most often the names of files that
are to be produced. The names following the colons are known as dependencies of the targets. They
usually denote other files (possibly, other targets) that must be present and up-to-date before the
target can be processed. The lines starting with tabs2 that follow the first line of a rule are called
actions. They are shell commands (that is, commands that you could type in response to the Unix
prompt) that get executed in order to create or bring up to date the target of the rule (we’ll use the
generic term update for the process of determining whether action is necessary on a particular target
and then (if needed) building or rebuilding it).
Each rule says, in effect, that to update the targets, each of the dependencies must first be updated
(recursively). Next, if a target does not exist (that is, if no file by that name exists) or if it does
exist but is older than one of its dependencies (so that one of its dependencies was changed after the
target was last updated), the actions of the rule are executed to create or update that target. The
1
For “GNU make,” GNU being an acronym for “GNU’s Not Unix.” gmake is “copylefted” (it has a license that
requires free use of any product containing it). It is also more powerful than the standard make utility.
2
Tabs, not blanks. Yes, I know: this is a really irritating design, because if you ever make the mistake of substituting
blanks for the tab, you get errors (with very unhelpful messages). The make utility is of rather ancient lineage, and the
file format has been this way since before most of you were born (literally).
Basic Compilation: javac and gmake 5

# Makefile for a simple editor

# The jar file contains the entire collection of classes

# constituting the editor.
edit.jar: edit.class commands.class display.class files.class
jar cf edit.jar edit.class commands.class display.class \
files.class

edit.class: edit.java commands.java display.java files.java

javac -g edit.java commands.java display.java \
files.java

commands.class: edit.java commands.java display.java files.java

javac -g edit.java commands.java display.java \
files.java

display.class: edit.java commands.java display.java files.java

javac -g edit.java commands.java display.java \
files.java

files.class: edit.java commands.java display.java files.java

javac -g edit.java commands.java display.java \
files.java

Figure 1: Sample makefile for an editor program. Adapted from “GNU Make: A Program for Directing Recompi-
lation” by Richard Stallman and Roland McGrath, 1988.

program will complain if any dependency does not exist and there is no rule for creating it. To start
the process off, the user who executes the gmake utility specifies one or more targets to be updated.
The first target of the first rule in the file is the default.
In the example above, edit.jar is the default target. The first step in updating it is to update
all the files listed as dependencies (a bunch of .class files). The remaining rules tell how to update
each of these .class files. As you can see, they all look pretty much the same, and say that to
update X.class:

First update all the source files edit.java, command.java, etc. Next, if X.class is
missing or is older than any of these source files, then execute javac on all the source
files.

We chose to compile all the source files together like this because otherwise it is possible for the
compiler to get confused by old .class files that are still lying around.
Updating the source (.java) files is easy. There are no rules for any of them, so gmake simply
insists that they all exist in order to be considered up to date.
Now edit.class, for example, is up to date if it is younger (was created more recently) than all
the files edit.java, command.java, and so forth. If instead it is older, gmake assumes that that one
of those source files has been changed since the last compilation that produced edit.class and must
be “rebuilt.” Of course, if edit.class does not exist, then gmake also knows it has to be rebuilt.
If rebuilding is necessary, gmake executes the action “javac -g edit.java commands.java ...”,
producing new .class files. In our particular case, if any one of the .class files needs to be rebuilt,
6 P. N. Hilfinger

they are all rebuilt. If two of them need to be rebuilt (let’s say edit.class and files.class), then
gmake will execute the action for one of them and then, when it checks the other .class file, will
discover that it has already been updated. Thus, you needn’t worry that the javac command will
be executed more than once.
Once all the .class files are up-to-date, gmake will check to see if any of them are younger than
edit.jar (or if edit.jar does not exist). If any of the class files had to be rebuilt, then of course it
will be younger, and gmake will execute the indicated action: “jar cf edit.jar....”
To invoke gmake for this example, one issues the command

gmake -f makefile-name target-names

where the target-names are the targets that you wish to update and the makefile-name given in the
-f switch is the name of the makefile. By default, the target is that of the first rule in the file.
Furthermore, you may (and usually do) leave off -f makefile-name, in which case it defaults to either
Makefile, makefile, or (in the case of gmake only) GNUmakefile, whichever exists. It is typical to
arrange that each directory contains the source code for a single principal program. By adopting
the convention that the rule with that program as its target goes first, and that the makefile for the
directory is named Makefile, you can arrange that, by convention, issuing the command gmake with
no arguments in any directory will update the principal program of that directory.
It is possible to have more than one rule with the same target, as long as no more than one rule
for each target has an action. Thus, I can also write the latter part of the example above as follows:

edit.class:
javac -g edit.java commands.java display.java files.java

commands.class:
javac -g edit.java commands.java display.java files.java

display.class:
javac -g edit.java commands.java display.java files.java

files.class:
javac -g

edit.class: edit.java commands.java display.java files.java

commands.class: edit.java commands.java display.java files.java
display.class: edit.java commands.java display.java files.java
files.class: edit.java commands.java display.java files.java

The order in which these rules are written is irrelevant. Which order or grouping you choose is largely
a matter of taste, aside from which is the first (default) target.
Next, you can combine rules with the same dependencies and action. For example:

edit.class commands.class display.class files.class

javac -g edit.java commands.java display.java files.java

edit.class commands.class display.class files.class: edit.java \

commands.java display.java files.java

or just
Basic Compilation: javac and gmake 7

edit.class commands.class display.class files.class: edit.java \

commands.java display.java files.java
javac -g edit.java commands.java display.java files.java
The example of this section illustrates the concepts underlying gmake. The rest of gmake’s features
exist mostly to enhance the convenience of using it.

6.2 Variables
You can clarify the example from §6.1 considerably and eliminate redundancy by defining variables
to contain the names of the files.
# Makefile for simple editor

JFLAGS = -g

JAVA_SRCS = edit.java \
commands.java \
display.java \
files.java

CLASSES = edit.class commands.class display.class files.class

edit.jar : $(CLASSES)
jar cf edit.jar $(CLASSES)

$(CLASSES): $(JAVA_SRCS)
javac $(JFLAGS) $(JAVA_SRCS)
The (continued) line beginning “JAVA SRCS =” defines the variable JAVA SRCS, which can later be
referenced as “$(JAVA SRCS)”. These later references cause the definition of JAVA SRCS to be sub-
stituted verbatim before the rule is processed. It is somewhat unfortunate that both gmake and the
shell use ‘$’ to prefix variable references; gmake defines ‘$$’ to be simply ‘$’, thus allowing you to
send ‘$’s to the shell in actions, where needed.
You will sometimes find that you need a value that is just like that of some variable, with a certain
systematic substitution. For example, given a variable listing the names of all source files, you might
want to get the names of all resulting .class files. You can rewrite the definition of CLASSES above
to get this.
CLASSES = $(JAVA_SRCS:.java=.class)
The substitution suffix ‘:.java=.class’ specifies the desired substitution. I now have variables for
both the names of all sources and the names of all class files without having to repeat a lot of file
names (and possibly make a mistake). (I have assumed here that each source file contains a single
class whose name is derived from the source file. You can’t use this trick if that isn’t so.)
Variables may also be set in the command line that invokes gmake. For example, the makefile
above contains what might look like an unnecessary definition of JFLAGS. However, defining it like
that allows one to write:
gmake JFLAGS="-g -deprecation" ...
which passes an extra flag to javac (this one happens to give a fuller explanation of certain warning
messages). Variable definitions in the command lines override those in the makefile, which allows the
makefile to supply defaults.
8 P. N. Hilfinger

6.3 Phony targets

It is often useful to have targets for which there are never any corresponding files. If the actions for
a target do not create a file by that name, it follows from the definition of how gmake works that
the actions for that target will be executed each time gmake is applied to that target (because it will
think the target is missing). A common use is to put a standard “clean-up” operation into each of
your makefiles, specifying how to get rid of files that can be reconstructed, if necessary. For example,
you will often see a rule like this in a makefile.
.PHONY: clean

clean:
rm -f *.class *~
Every time you issue the shell command “gmake clean,” this action will execute, removing all .class
files and Emacs old-version files.
The special .PHONY target tells gmake that clean is not a file, and is instead just the name of a
target that is always out of date. Therefore, when you make the “clean” target, gmake will always
execute the rm command, regardless of what files happen to be lying around. In effect, .PHONY tells
gmake to treat clean as a command.
Another possible use is to provide a standard way to run a set of tests on your program—what
are typically known as regression tests—to see that it is working and has not “regressed” as a result
of some change you’ve made. For example, to cause the command
make check
to feed a test file through our editor program and check that it produces the right result, use:
.PHONY: check

check: edit
rm -f test-file1
java edit < test-commands1
diff test-file1 expected-test-file1
where the test input file test-commands1 presumably contains editor commands that are supposed
to produce a file test-file1, and the file expected-test-file1 contains what is supposed to be
in test-file1 after executing those commands. The first action line of the rule clears away any old
copy of test-file1; the second runs the editor and feeds in test-commands1 through the standard
input, and the third compares the resulting file with its expected contents. If either the second or
third action fails, gmake will report that it encountered an error.

6.4 Details of actions

By default, each action line specified in a rule is executed by the Bourne shell (as opposed to the
C shell, which, most unfortunately, is more commonly used here). For the simple makefiles we are
likely to use, this will make little difference, but be prepared for surprises if you get ambitious.
The gmake program usually prints each action as it is executed, but there are times when this is
not desirable. Therefore, a ‘@’ character at the beginning of an action suppresses the default printing.
Here is an example of a common use.
edit.jar : $(CLASSES)
@echo Creating edit.jar ...
@jar cf edit.jar $(CLASSES)
@echo Done
Basic Compilation: javac and gmake 9

The result of these actions is that when gmake executes this final step for the edit program, the only
thing you’ll see printed is a line reading “Creating edit.jar ...” and, at the end of the step, a
line reading “Done”.
When gmake encounters an action that returns a non-zero exit code, the UNIX convention for
indicating an error, its standard response is to end processing and exit. The error codes of action
lines that begin with a ‘-’ sign (possibly preceded by a ‘@’) are ignored. Also, the -k switch to gmake
will cause it to abandon processing only of the current rule (and any that depend on its target) upon
encountering an error, allowing processing of “sibling” rules to proceed.

6.5 Including makefiles

A good way to create makefiles is to have a template that you include in your particular makefile—
something like the example in Figure 2. We’ve prepared one like this already, so that in the very
simplest case, your makefile can contain just:
JAVA_SRCS = edit.java commands.java display.java files.java

include $(MASTERDIR)/lib/java.Makefile.std
As you can probably guess, the include line is a special command that essentially gets replaced by
the contents of the named file.
Figure 2 illustrates what such a template file might look like. It uses one obscure new feature
that makes it possible to partially define an action, and allow others to add to it. The definition
of the phony target clean uses two colons rather than one. This is a signal that there may be
other “double-colon” rules for clean, complete with actions. They will all get used (in the order
encountered). For example, if you include this particular template in a place where you want to
define additional clean-up actions besides the ones defined in the template, you can write:

JAVA_SRCS = edit.java commands.java display.java files.java

include $(MASTERDIR)/lib/java.Makefile.std

clean::
rm -rf test-output

which will cause gmake clean to remove the directory test-output as well as the class files and
Emacs-generated files removed in the template.
10 P. N. Hilfinger

# Standard definitions for make utility: Java version.

# Assumes that this file is included from a Makefile that defines

# JAVA_SRCS to be a list of Java source files to be compiled.
# It may optionally define OTHER_CLASSES to contain names of classes
# that aren’t derivable from the names of the JAVA_SRCS files.
# The including Makefile may subsequently override JFLAGS (flags to
# the Java compiler), and JAVAC (the Java compiler’s name), by putting
# these definitions after the "include".

# Targets defined:
# default:Default entry. Compiles classes from all source files.
# clean:: Remove back-up files and files that make can reconstruct.
# You can add additional clean-up actions by adding more
# ’clean::’ targets (note the double colon) to your makefile.
# check: Look in the subdirectory tests for all files whose name ends
# in ’.sh’. Each of these should be an executable shell script
# (a file of commands such as you could enter at the command
# prompt) that performs some test of the program. Run each
# and report all that fail (return a non-zero exit code).
#

JAVAC = javac

JFLAGS = -g

CLASSES = $(JAVA_SRCS:.java=.class) $(OTHER_CLASSES)

.PHONY: clean check default

# Default entry
default: $(CLASSES)

$(CLASSES): $(JAVA_SRCS)
$(JAVAC) $(JFLAGS) $(JAVA_SRCS)

clean::
/bin/rm -f $(CLASSES) *~

check: $(CLASSES)
cd tests; for test in *.sh; do \
if ./$${test}; then \
echo "$${tests}: OK."; \
else \
echo "$${tests}: FAILED."; \
fi; \
done

Figure 2: An example of a file of standard makefile definitions that can be included from a specific makefile to
compile many simple collections of Java programs.