Thursday, August 06, 2009

Makefile how to

Here is a brief instruction of Makefile. Thanks the author.
http://www.wlug.org.nz/MakefileHowto


Makefiles are easy.
In fact, to build a simple program that doesn't depend on any libraries, you don't even need a makefile. make(1) is smart enough to figure it all out itself. For instance, if you have a file "foo.c" in the current directory:

$ ls foo.c  $ make foo cc foo.c -o foo

make(1) will detect the type of file and compile it for you, automatically naming the executable the same as the input file (gcc(1) foo.c will give you a file called a.out unless you manually specify a name for it). If you need libraries, you can specify them by setting the LDFLAGS variable on the command line.

Of course, most useful projects contain more than one file. A makefile describes the dependencies between files. It is called Makefile (with a capital M). Each line will typically consist of a filename, a colon and a list of dependencies. for instance, a simple make file to link together two object files foo.o and bar.o might look like:

program: foo.o bar.o

Each filename (before the colon) is called a target. You can make a specific target by executing

$ make target

make is smart enough to use the first rule in the Makefile as the default action, so:

$ ls Makefile foo.c bar.c  $ make cc bar.c -c -o bar.o cc foo.c -c -o foo.o cc foo.o bar.o -o program

See the CompilingHowto for more info on the steps required to turn source code into an executable.

You can get your Makefiles made automatigically for you using AutoTools.


Dynamic updating

Occasionally you might want to specify something special to happen, for a specific file. This can be done by providing some rules to build that target. This is done indented, on the next line after the dependencies are listed. Our sample make file again:

program: foo.o bar.o  bar.c:   echo 'char *builddate="' `date` '";' >bar.c

Note that the line that begins "echo" must be indented by one tab. If this isn't done make(1) will abort with a weird error message like "Missing delimiter". The echo line makes a one line C file with a variable called "builddate", set to the current date and time. This is a useful thing to do for your program if you wanted to know when this particular version was compiled. (Not that this is the only way, or in fact the best way to get this information, but it's a good example.)

Running this would produce:

$ make echo 'char *builddate="' `date` '"' >bar.c cc    -c -o bar.o bar.c cc    -c -o foo.o foo.c cc foo.o bar.o -o program

Phony targets

You can have "phony" targets -- targets which don't actually create a file, but do something. These are created like normal targets: for instance, to add a "all" target to our makefile we'd add (probably at the top, so it becomes the default target):

all: foo

This rule won't run if there exists a file called "all" in the directory (if someone was stupid enough to create one somehow). So we can tell make(1) that this is a phony target and should be rebuilt always this is by using the target .PHONY. so, we can add to our Makefile:

.PHONY: all

To add a clean target is fairly simple too, add:

clean:   rm -f bar.o bar.c foo.o foo.c

and add clean to the list of phony targets:

.PHONY: all clean

Selective building

Why use a makefile, instead of a script to rebuild everything from scratch?

If you have a rule that reads

objectfile.o: foo.c foo.h bar.c bar.h Makefile

then make(1) will check the last modification date of objectfile.o against the last modification date of all the files that follow it (foo.c, foo.h, bar.c, bar.h and the Makefile itself). If none of these things have changed, then it won't recompile objectfile.o.

Build lines like this with careful reference to #includes in your source - if your foo.h #includes bar.h, it has to be on the Makefile line - otherwise, changes to bar.h won't cause a recompile of objectfile.o and you might get confused as to why your constants aren't what you thought they should be.

Or, you could have make determine all your header file dependencies for you! If foo.h #includes bar.h, and bar.h #includes another.h, which #includes etc.h, it could very quickly become difficult to keep track of it all. Not to mention it may result in huge dependency lines! Instead, you can have a header file as a target and list its #included files as its dependencies. Then use the 'touch' command to update the timestamp. For example, if foo.c #includes foo.h, and both foo.h and bar.c #include bar.h, we could use this Makefile:

executable: foo.o bar.o         $(CC) foo.o bar.o -o executable foo.o: foo.c foo.h Makefile bar.o: bar.c bar.h Makefile foo.h: bar.h         touch foo.h bar.h:

So if you edit bar.h to change some constants or function definitions, Make will see that foo.h needs to be updated and 'touch' it. Then it will know it must also update foo.o in (in addition to bar.o) since foo.h appears new. This way each target only lists files that it is directly dependent on. Let make figure out the rest -- that's what it's supposed to do!

Or you could decide that mindless drone work is a waste of time and just use makedepend to spare yourself the hassle. --AristotlePagaltzis

You should consider that using this touch could affect the configuration management system you are using (e.g. RCS or CVS), if it goes by the timestamp to determine the need to commit/checkin: you might suddenly have lots of files to commit, or lots of files locked! At the very least the new timestamp will confuse your friends and confound your enemies. However, makedepends can generate an unreadable and therefore unmaintainable monstrosity, partly because it cites every system dependency (e.g. stdio), and also as it recurses through the subdep files it cites each reference to stdio by the subdeps as if it were a separate dependency. So, depending on the size of your project, and how often you have to make major adjustments by hand to the makefiles, and how many headers each file uses, you may want to decide whether or not to use this touch method (which would indeed keep the dependencies nicely hierarchical), or use makedepends. To have it both ways, I believe you could precede the touch with "cp -p foo.h foo.h_preserveDate; touch foo.h" and then under the foo.o dependency you could after the compile then do "cc foo.c; mv foo.h_preserveDate foo.h" which would preserve the original date on the foo.h checked-out file. This would still keep the hierarchical nature, which is quite valuable because it eliminates redundancy in separate places (two distant places to maintain one fact is very bad). -- LindaBrock


Makefiles in subdirectories

With larger projects you often have subdirectories with their own Makefile. To allow make to run these Makefiles with the options passed to make use the $(MAKE) variable. This variable actually callse a second make process to make the Makefile in the subdirectory. To specify the Makefile's subdirectory use the -C option of make.

Example Makefile:

    all: Documentation/latex/refman.pdf      install: Documentation/latex/refman.pdf         cp Documentation/latex/refman.pdf Documentation/!KeithleyMeter.pdf      Documentation: Doxyfile Makefile src/keithleyMeter.cc hdr/keithleyMeter.h         # Dosen't use all the options you passed to make         make clean         # make the Documentation folder         /Applications/Doxygen.app/Contents/Resources/doxygen      Documentation/latex/refman.pdf: Documentation         # Uses the options you passed to make         $(MAKE) -C Documentation/latex      clean:         rm -rf Documentation

For a counter-argument against having separate make processes for sub-directories (and instead using makefile fragments but only one make process), see Recursive Makefile considered harmful (PDF)


Rules

The real power from makefiles comes when you want to add your own "rules" for files. If we have a program called "snozzle" that takes a ".snoz" file and produces a ".c" file we can add:

%.c: %.snoz    snozzle $< -o [email protected]

$< expands to the first dependency, and [email protected] the target. So, if foo.c is built from foo.snoz we can now:

$ ls Makefile foo.snoz $ make snozzle foo.snoz -o foo.c cc -c -o foo.o foo.c echo 'char *builddate="' `date` '"' >bar.c cc -c -o bar.o bar.c cc foo.o bar.o -o foo rm foo.c

Note that foo.c is removed by make at the end -- make(1) removes intermediate files itself when it's done. Smart, eh?


EnvironmentVariables

The only other major thing left to mention about Make is environmental variables. It uses $(variable) as an expando. thus the rule:

%c: %.snoz   snozzle $(SNOZFLAGS) $<

would let you specify the arguments to snozzle. This is useful if you call snozzle in multiple places, but want to be able to make one change to update the flags.

make(1) uses these variables for its compilers. The compiler it uses for compiling C is "CC", You can set the environment variable "CC" to your own favourite C compiler if you so wish. CFLAGS is used for the flags to the C compiler. Thus setting CFLAGS to "-g -Wall" will compile all programs with debugging (-g) and with all warnings enabled (-Wall). Environment variables can be defined in make by using "VARIABLE=value" for example:

CFLAGS=-g -Wall

So, our full make file would become:

CFLAGS=-g -Wall SNOZFLAGS=--with-extra-xyzzy  all: program  clean:    rm -f foo.c foo.o bar.c bar.o  .PHONY: clean all  program: foo.o bar.o  bar.c:   echo 'char *builddate="' `date` '";' >bar.c  %.c: %.snoz    snozzle $(SNOZFLAGS) $< -o [email protected]
  • CPPFLAGS command line flags to cpp
  • CFLAGS command line flags to cc
  • CXXFLAGS command line flags to c++
  • LDFLAGS command line flags to ld
  • ASFLAGS command line flags to as

If you specify your own command line you will have to explicitly include these variables in it.

You can also check if an environment variable has been set and initialise it to something if it has not. ie.

DESTDIR ?= /usr/local

will set DESTDIR to /usr/local if it is not already defined

To append to the environment variables use the += operator:

CFLAGS += -g -Wall

This allows the user to specify system specific optimizations in their shell environment.

Note : As you may have noticed, make uses $ to identify variables - both environment and defined in the file. To put a literal $ in a makefile, use $$. However, bash also uses $ to identify variables, and will consume the $ when it is passed to whatever program you're running. To therefore pass a literal $ to a program you must use \$$ - note the single \, not double. - OrionEdwards


An example makefile

 1: CXXFLAGS=-g  2:  3: sim: car.o road.o sim.o event.o  4:        g++ $(LDFLAGS)  sim.o car.o road.o event.o -lm   -o sim  5:  6: car.o: car.cc car.h sim.h event.h road.h Makefile  7: sim.o: sim.cc sim.h car.h road.h event.h Makefile  8: road.o: road.cc road.h sim.h event.h car.h Makefile  9: event.o: event.cc event.h sim.h Makefile

This makefile is for a car simulator written in C++. (It was written by Dr. Tony !McGregor? from TheUniversityOfWaikato).

  • Line 1 sets up the environment variables to the C++ compiler, ensuring everything is compiled with debugging info on.
  • Line 3 is the first target in the file, so when you run 'make' it will 'make sim'. sim depends on car.o, road.o etc (targets that are defined on lines 6-9).
  • Line 4 is indented; because we want to add extra smarts to the compiling of sim (we want to link to the math library libm.a); so when 'make sim' is executed and the .o's are up to date, that line will be executed.
  • Lines 6-9 are targets for the various object files that will be generated. They say that car.o is built from car.cc, car.h etc. This probably means that car.h somewhere #include's event.h, road.h... Every time you run 'make car.o', it will compare the last modification date on all the files listed against the modification date of car.o. If car.o is newer, it is up to date and no compiling is necessary. Otherwise, make will recompile everything it needs to.

Functions

It is possible to call some predefined functions in makefiles. A full list of them can be found in the manual, of course (http://www.gnu.org/software/make/manual/html_chapter/make_8.html#SEC83).

Perhaps you want to find all the .c files in directory for later use:

SOURCES := $(wildcard *.c)

Given these, maybe you want to know the names of their corresponding .o files:

OBJS := $(patsubst %.c, %.o, $(SOURCES))

You can do things like adding prefixes and suffixes, which comes in handy quite often. For example, you could have at the top of the makefile a variable where you set the libraries to be included:

LIBS := GL SDL stlport

And then use

$(addprefix -l,$(LIBS))

in a later rule to add a -l prefix for every library mentioned in LIBS above.

Finding files in multiple directories is a good example of the usage of foreach

DIRS := src obj headers FILES := $(foreach dir, $(DIRS), $(wildcard $(dir)/*))

Automatic dependency calculation

Based on manuals and research papers, if you are creating a Makefile for C/C++ gcc can calculate dependency information for you. The quickest way to get this going is to add the -MD flag to your CFLAGS first. You will then need to know the names of the .d files in your makefile. I do something like this:

DEPS := $(patsubst %.o,%.d,$(OBJS))

Then near the end of the makefile, add an

-include $(DEPS)

It might also help to make a 'deps' target:

deps: $(SOURCES)      $(CC) -MD -E $(SOURCES) > /dev/null

'-E' tells gcc to stop after preprocessing. When using -E, the processed C file is sent to STDOUT. Therefore to avoid the mess on the screen, send it to /dev/null instead. Using this command all of the *.d files will be made.


Comments

As your Makefile gets longer, you may want to insert comments to explain what the file is supposed to do. Comment lines, which are ignored by make, begin with a '#':

deps: $(SOURCES)      # the following line makes all of the .d files.      $(CC) -MD -E $(SOURCES) > /dev/null

You can also put a comment on the same line as another statement:

-include $(DEPS)  # this includes everything in DEPS

Gotchas

Some OperatingSystems use filesystems (such as MicrosoftWindows FAT, FAT32 and NTFS(?) and Apple's HFS) that are case insensitive, and can cause problems. (This implies you are using Cygwin on windows or Darwin on MacOSX). Particularly as unix packages often have a file named INSTALL which has installation instructions, the command "make install" says

make: Nothing to be done for install

You can fix this in your make(1) files by adding:

.PHONY: install

This will tell make that the "install" target is a "phony" target, and doesn't actually refer to a file and should always be rebuilt.

If you are on an OS such as FreeBSD you might need to invoke 'gmake' for a GNU compatible make.

Shell variables in Makefiles

There may come a time you need to use shell scripting complicated enough to require shell vars in a Makefile but make has issues since $ is the prefix for Make vars too, to escape the $, just use $$, so this:

for e in * ; do echo $e ; done

becomes:

for e in * ; do echo $$e ; done

It's a simple change but I didn't see it written anywhere obvious :)


See Also:


Requests

I'd like to know something about makedepend and such things. Maybe some links to other or "official" make HOWTOs would be useful as well. Thanks. -- Someone

Dear Someone, Take a look at the make manual, especially section 4.14. Basically 'make depend' is not really needed anymore.


I cannot find info about the meaning of '@[email protected]' variables in a Makefile. At the moment i get the error
make: [email protected]: Kommando nicht gefunden make: *** [arandom.lo? Fehler 127

thx, !FlorianKonnertz?

This isn't really anything to do with make. The autoconf/configure methods that many projects use take a template file (such as Makefile.in) and use that to create a makefile. autoconf uses things like @[email protected] for its variables, and should replace @[email protected] vars with something that makes sense to make. If you have a makefile that still has @[email protected] variables in it, then it's a bug and there is a bug in the package.


I have a question. I have a directory called src. Within this directory, a group publishes designs inside directories:

Release_010405 Release_010505 Release_010605 Release_032304 Release_082903
If there is a file called baja.c inside one of these directories that is newer than baja.o, I want to compile. I was able to make a list of all the baja.c files within the Release directories using wildcard
NEWSOURCES = $(wildcard Release_*/baja.c)

However, I don't know how to tell Make which is the latest file. The following grabs the first of the list.

baja.local: $(NEWSOURCES)

cp $< .

You could try using $? which gives you the names of the prerequisites which are newer than the target. If there can be several of those and you only need the latest, though, you have to do it in the recipe, using shell tools. --AristotlePagaltzis

No comments: