Basic compilation control with gmake

"Footnotes"

The name gmake stands 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.

The example used in this document is adapted from "GNU Make: A Program for Directing Recompilation" by Richard Stallman and Roland McGrath, 1988.

Most of this document was taken nearly verbatim from Paul Hilfinger. Any errors introduced, however, are either David Wolfe's or Karl Knight's.

Introduction

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.

The UNIX make utility is a conceptually-simple and general solution to these problems. 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, we will use a version known as gmake. Though conceptually simple, the make 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.

Basic Operation and Syntax

The following is a sample makefile for compiling a simple editor program, edit, from eight .cc files and three header (.h) files. 
    # Makefile for simple editor

    edit : edit.o kbd.o commands.o display.o \
           insert.o search.o files.o utils.o
            g++ -g -o edit edit.o kbd.o commands.o display.o \
                       insert.o search.o files.o utils.o

    edit.o : edit.cc defs.h
            g++ -g -c -Wall edit.cc
    kbd.o : kbd.cc defs.h command.h
            g++ -g -c -Wall kbd.cc
    commands.o : command.cc defs.h command.h
            g++ -g -c -Wall commands.cc
    display.o : display.cc defs.h buffer.h
            g++ -g -c -Wall display.cc
    insert.o : insert.cc defs.h buffer.h
            g++ -g -c -Wall insert.cc
    search.o : search.cc defs.h buffer.h
            g++ -g -c -Wall search.cc
    files.o : files.cc defs.h buffer.h command.h
            g++ -g -c -Wall files.cc
    utils.o : utils.cc defs.h
            g++ -g -c -Wall utils.cc
This file consists of a sequence of nine rules. Each 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 ending it with a backslash-newline combination, 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 the 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 (generally, other targets) that must be present and up-to-date before the target can be processed. The lines starting with tabs that follow the first line of a rule we will call 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 update the target of the rule (we'll use the generic term update for both).

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 it was last updated), the actions of the rule are executed to create or update that target. The 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 is the default target. The first step in updating it is to update all the object (.o) files listed as dependencies. To update edit.o, in turn, requires first that edit.cc and defs.h be updated. Presumably, edit.cc is the source file that produces edit.o and defs.h is a header file that edit.cc includes. There are no rules targeting these files; therefore, they merely need to exist to be up-to-date. Now edit.o is up-to-date if it is younger than either edit.cc or defs.h (if it were older, it would mean that one of those files had been changed since the last compilation that produced edit.o). If edit.o is older than its dependencies, gmake executes the action "g++ -g -c -Wall edit.cc", producing a new edit.o. Once edit.o and all the other .o files are updated, they are combined by the action "g++ -g -o edit ..." to produce the program edit, if either edit does not already exist or if any of the .o files are younger than the existing edit file.

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. You may (and usually do) leave off -f makefile-name, in which case it defaults to either makefile or Makefile, 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, we can also write the latter part of the example above as follows. 

    edit.o : edit.cc
            g++ -g -c -Wall edit.cc
    kbd.o : kbd.cc
            g++ -g -c -Wall kbd.cc
    commands.o : command.cc
            g++ -g -c -Wall commands.cc
    display.o : display.cc
            g++ -g -c -Wall display.cc
    insert.o : insert.cc
            g++ -g -c -Wall insert.cc
    search.o : search.cc
            g++ -g -c -Wall search.cc
    files.o : files.cc
            g++ -g -c -Wall files.cc
    utils.o : utils.cc
            g++ -g -c -Wall utils.cc

    edit.o kbd.o commands.o display.o \
        insert.o search.o files.o utils.o: defs.h
    kbd.o commands.o files.o : command.h
    display.o insert.o search.o files.o : buffer.h
The order in which these rules are written is irrelevant. Which order or grouping you choose is largely a matter of taste.

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.

Variables

The dependencies of the target edit in our example are also the arguments to the command that links them. One can avoid this redundancy by defining a variable that contains the names of all object files. 
    # Makefile for simple editor

    OBJS = edit.o kbd.o commands.o display.o \
           insert.o search.o files.o utils.o

    edit : $(OBJS)
            g++ -g -o edit $(OBJS)
The (continued) line beginning ``OBJS ='' defines the variable OBJS, which can later be referenced as $(OBJS) or ${OBJS}. These later references cause the definition of OBJ to be substituted verbatim before the rule is processed. It is somewhat unfortunate that both gmake and the shell use `$' to prefix variable references; gmake defines `$$' to be simply `$', thus allowing you to send `$'s to the shell, 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 .o files. We can rewrite the definition of OBJS above to get this. 

SRCS = edit.cc kbd.cc commands.cc display.cc \
       insert.cc search.cc files.cc utils.cc
OBJS = $(SRCS:.cc=.o)
The substitution suffix `:.cc=.o' specifies the desired substitution. We now have variables for both the names of all sources and the names of all object files without having to repeat a lot of file names (and possibly make a mistake).

Variables may also be set in the command line that invokes gmake. For example, if the makefile contains 

    edit.o: edit.cc
            g++ $(DEBUG) -c -Wall edit.cc
Then a command such as 
    gmake DEBUG=-g ...
will cause the compilations to use the -g (add symbolic debugging information) switch, while leaving off the DEBUG=-g will not use the -g switch. Variable definitions in the command lines override those in the makefile, which allows the makefile to supply defaults.

Variables not set by either of these methods may be set as UNIX environment variables. Thus, the sequence of commands 

    setenv DEBUG -g
    gmake ...
for this last example will also use the -g switch during compilations.

Implicit rules

In our example, all of the compilations that produced .o files have the same form. It is tedious to have to duplicate them; it merely gives you the opportunity to type something wrong. Therefore, gmake can be told about---and for some standard cases, already knows about---the default files and actions needed to produce files having various extensions. For our purposes, the most important is that it knows how to produce a file file.o given a file of the form file.cc, and knows that the file.o file depends on the file file.cc. Specifically, gmake automatically introduces (in effect) the rule 
file.o : file.cc
	$(CXX) -c -Wall $(CXXFLAGS) file.cc
when called upon to produce file.o when there is a C++ file file.cc present, but no explicitly specified actions exist for producing file.o. The use of the prefix ``CXX'' is a naming convention for variables that have to do with C++. It also creates the command 
file : file.o
	$(CC) $(LDFLAGS) file.o $(LOADLIBES) -o file
to tell how to create an executable file named file from file.o.

As a result, we may abbreviate the example as follows. 

    # Makefile for simple editor

    SRCS = edit.cc kbd.cc commands.cc display.cc \
	   insert.cc search.cc files.cc utils.cc

    OBJS = $(SRCS:.cc=.o)

    CC = gcc

    CXX = g++

    CXXFLAGS = -g

    LOADLIBES = 

    edit : $(OBJS)
    edit.o : defs.h
    kbd.o : defs.h command.h
    commands.o : defs.h command.h
    display.o : defs.h buffer.h
    insert.o : defs.h buffer.h
    search.o : defs.h buffer.h
    files.o : defs.h buffer.h command.h
    utils.o : defs.h
There are quite a few other such implicit rules built into gmake. The -p switch will cause gmake to list them somewhat cryptically, if you are at all curious. We are most likely to be using the rules for creating .o files from .cc (C++) files. It is also possible to supply your own default rules and to suppress the standard rules; for details, see the full documentation.

Special actions

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. 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     # This first line isn't really necessary.
    clean:
            rm -f *.o
Every time you issue the shell command gmake clean, this action will execute, removing all .o files.

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 gmake 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 : $(OBJS)
            @echo Linking edit ...
            @g++ -g -o edit $(OBJS)
            @echo Done
The result of these actions is that when gmake executes this final editing step for the edit program, the only thing you'll see printed is a line reading ``Linking edit...'' 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.

Creating makefiles

A good way to create makefiles is to have a template that you tailor to your particular program, such as the following template. By simply copying this template and replacing the angle-bracketed portions with the particulars of your program (as in the following filled in template) you get a working makefile.

As a final convenience, you may use the -MM option to g++ to create the dependencies automatically, and then add them into your modified makefile. This is the purpose of the depend special target. 

PROGRAM = <REPLACE WITH PROGRAM NAME>

LOADLIBES = <LOAD LIBRARIES>

CXX_SRCS = <C++ SOURCE FILE NAMES>

CC = gcc

LDFLAGS = -g

CXX = g++

CXXFLAGS = -g -Wall -fno-builtins 

OBJS = $(CXX_SRCS:.cc=.o)

$(PROGRAM) : $(OBJS)
	$(CC) $(LDFLAGS) $(OBJS) $(LOADLIBES) -o $(PROGRAM)

clean:
	/bin/rm -f *.o $(PROGRAM) *~

depend:
	$(CXX) -MM $(CXX_SRCS)

###
<DEPENDENCIES ON .h FILES GO HERE>
An example of a Makefile template that can be tailored to many simple C++ programs. 
PROGRAM = edit

LOADLIBES = 

CXX_SRCS = edit.cc kbd.cc commands.cc display.cc \
	   insert.cc search.cc files.cc utils.cc

CC = gcc

LDFLAGS = -g

CXX = g++

CXXFLAGS = -g -Wall -fno-builtins 

OBJS = $(CXX_SRCS:.cc=.o)

$(PROGRAM) : $(OBJS)
	$(CC) $(LDFLAGS) $(OBJS) $(LOADLIBES) -o $(PROGRAM)

clean:
	/bin/rm -f *.o $(PROGRAM) *~

depend:
	$(CXX) -MM $(CXX_SRCS)

###
edit.o : defs.h
kbd.o : defs.h command.h
commands.o : defs.h command.h
display.o : defs.h buffer.h
insert.o : defs.h buffer.h
search.o : defs.h buffer.h
files.o : defs.h buffer.h command.h
utils.o : defs.h
A Makefile created by modifying the previous template.