++ed by:

22 PAUSE users
14 non-PAUSE users.

Ken Williams


Module::Build - Build and install Perl modules


 Standard process for building & installing modules:
   perl Build.PL
   ./Build test
   ./Build install


This is a beta version of a new module I've been working on, Module::Build. It is meant to be a replacement for ExtUtils::MakeMaker.

To install Module::Build, and any other module that uses Module::Build for its installation process, do the following:

  perl Build.PL       # 'Build.PL' script creates the 'Build' script
  ./Build             # Need ./ to ensure we're using this "Build" script
  ./Build test        # and not another one that happens to be in the PATH
  ./Build install

This illustrates initial configuration and the running of three 'actions'. In this case the actions run are 'build' (the default action), 'test', and 'install'. Actions defined so far include:

  build                          help        
  clean                          install     
  dist                           manifest    
  distcheck                      realclean   
  distclean                      skipcheck   
  distdir                        test        
  disttest                       testdb      

You can run the 'help' action for a complete list of actions.

It's like the MakeMaker metaphor, except that Build is a short Perl script, not a long Makefile. State is stored in a directory called _build/.

Any customization can be done simply by subclassing Module::Build and adding a method called (for example) ACTION_test, overriding the default 'test' action. You could also add a method called ACTION_whatever, and then you could perform the action Build whatever.

More actions will certainly be added to the core - it should be easy to do everything that the MakeMaker process can do. It's going to take some time, though. In the meantime, I may implement some pass-through functionality so that unknown actions are passed to MakeMaker.

For information on providing backward compatibility with ExtUtils::MakeMaker, see Module::Build::Compat.


I list here some of the most important methods in Module::Build. Normally you won't need to deal with these methods unless you want to subclass Module::Build. But since one of the reasons I created this module in the first place was so that subclassing is possible (and easy), I will certainly write more docs as the interface stabilizes.

$m = Module::Build->new(...)

Creates a new Module::Build object. Arguments to the new() method are listed below. Most arguments are optional, but you must provide either the module_name argument, or dist_name and one of dist_version or dist_version_from. In other words, you must provide enough information to determine both a distribution name and version.

  • module_name

    The module_name is a shortcut for setting default values of dist_name and dist_version_from, reflecting the fact that the majority of CPAN distributions are centered around one "main" module. For instance, if you set module_name to Foo::Bar, then dist_name will default to Foo-Bar and dist_version_from will default to lib/Foo/Bar.pm. dist_version_from will in turn be used to set dist_version.

    Setting module_name won't override a dist_* parameter you specify explicitly.

  • dist_name

    Specifies the name for this distribution. Most authors won't need to set this directly, they can use module_name to set dist_name to a reasonable default. However, some agglomerative distributions like libwww-perl or bioperl have names that don't correspond directly to a module name, so dist_name can be set independently.

  • dist_version

    Specifies a version number for the distribution. See module_name or dist_version_from for ways to have this set automatically from a $VERSION variable in a module. One way or another, a version number needs to be set.

  • dist_version_from

    Specifies a file to look for the distribution version in. Most authors won't need to set this directly, they can use module_name to set it to a reasonable default.

    The version is extracted from the specified file according to the same rules as ExtUtils::MakeMaker and CPAN.pm. It involves finding the first line that matches the regular expression


    , eval()-ing that line, then checking the value of the $VERSION variable. Quite ugly, really, but all the modules on CPAN depend on this process, so there's no real opportunity to change to something better.

  • license

    Specifies the licensing terms of your distribution. Valid options include:

    • perl

      The distribution may be copied and redistributed under the same terms as perl itself (this is by far the most common licensing option for modules on CPAN). This is a dual license, in which the user may choose between either the GPL or the Artistic license.

    • gpl

      The distribution is distributed under the terms of the Gnu Public License.

    • artistic

      The distribution is licensed under the Artistic License, as specified by the Artistic file in the standard perl distribution.

    • restrictive

      The distribution may not be redistributed without special arrangement with the author.

    Note that you must still include the terms of your license in your documentation - this field only lets automated tools figure out your licensing restrictions. Humans still need something to read.

    If you use a licensing option unknown to Module::Build, an unknown license type will be used. Please let me know if you need another license to be recognized - I just started out with a small set to keep things simple, figuring I'd let people with actual working knowledge in this area tell me what to do.

  • requires

    An optional requires argument specifies any module prerequisites that the current module depends on. The prerequisites are given in a hash reference, where the keys are the module names and the values are version specifiers:

     requires => {Foo::Module => '2.4',
                  Bar::Module => 0,
                  Ken::Module => '>= 1.2, != 1.5, < 2.0',
                  perl => '5.6.0'},

    These four version specifiers have different effects. The value '2.4' means that at least version 2.4 of Foo::Module must be installed. The value 0 means that any version of Bar::Module is acceptable, even if Bar::Module doesn't define a version. The more verbose value '>= 1.2, != 1.5, < 2.0' means that Ken::Module's version must be at least 1.2, less than 2.0, and not equal to 1.5. The list of criteria is separated by commas, and all criteria must be satisfied.

    A special perl entry lets you specify the versions of the Perl interpreter that are supported by your module. The same version dependency-checking semantics are available, except that we also understand perl's new double-dotted version numbers.

    One note: currently Module::Build doesn't actually require the user to have dependencies installed, it just strongly urges. In the future we may require it. There's now a recommends section for things that aren't absolutely required.

    Automated tools like CPAN.pm should refuse to install a module if one of its dependencies isn't satisfied, unless a "force" command is given by the user. If the tools are helpful, they should also offer to install the dependencies.

    A sysnonym for requires is prereq, to help succour people transitioning from ExtUtils::MakeMaker. The requires term is preferred, but the prereq term will remain valid in future distributions.

  • recommends

    This is just like the prereq argument, except that modules listed in this section aren't essential, just a good idea. We'll just print a friendly warning if one of these modules aren't found, but we'll continue running.

    If a module is recommended but not required, all tests should still pass if the module isn't installed. This may mean that some tests will be skipped if recommended dependencies aren't present.

    Automated tools like CPAN.pm should inform the user when recommended modules aren't installed, and it should offer to install them if it wants to be helpful.

  • build_requires

    Modules listed in this section are necessary to build and install the given module, but are not necessary for regular usage of it. This is actually an important distinction - it allows for tighter control over the body of installed modules, and facilitates correct dependency checking on binary/packaged distributions of the module.

  • conflicts

    Modules listed in this section conflict in some serious way with the given module. Module::Build will refuse to install the given module if

  • c_source

    An optional c_source argument specifies a directory which contains C source files that the rest of the build may depend on. Any .c files in the directory will be compiled to object files. The directory will be added to the search path during the compilation and linking phases of any C or XS files.

  • autosplit

    An optional autosplit argument specifies a file which should be run through the Autosplit::autosplit() function. In general I don't consider this a great idea, and I may even go so far as to remove this feature later. Let me know if I shouldn't.

  • dynamic_config

    A boolean flag indicating whether the Build.PL file must be executed, or whether this module can be built, tested and installed solely from consulting its metadata file. The default value is 0, reflecting the fact that "most" of the modules on CPAN just need to be copied from one place to another. The main reason to set this to a true value is that your module performs some dynamic configuration as part of its build/install process.

    Currently Module::Build doesn't actually do anything with this flag - it's probably going to be up to tools like CPAN.pm to do something useful with it. It can potentially bring lots of security, packaging, and convenience improvements.


Creates an executable script called Build in the current directory that will be used to execute further user actions. This script is roughly analogous (in function, not in form) to the Makefile created by ExtUtils::MakeMaker. This method also creates some temporary data in a directory called _build/. Both of these will be removed when the realclean action is performed.


A Module::Build method may call $self->add_to_cleanup(@files) to tell Module::Build that certain files should be removed when the user performs the Build clean action. I decided to make this a dynamic method, rather than a static list of files, because these static lists can get difficult to manage. I preferred to keep the responsibility for registering temporary files close to the code that creates them.


You'll probably never call this method directly, it's only called from the auto-generated Build script. The new() method is only called once, when the user runs perl Build.PL. Thereafter, when the user runs Build test or another action, the Module::Build object is created using the resume() method.


This method is also called from the auto-generated Build script. It parses the command-line arguments into an action and an argument list, then calls the appropriate routine to handle the action. Currently (though this may change), an action foo will invoke the ACTION_foo method. All arguments (including everything mentioned in ACTIONS below) are contained in the $self->{args} hash reference.


If you're subclassing Module::Build and some code needs to alter its behavior based on the current platform, you may only need to know whether you're running on Windows, Unix, MacOS, VMS, etc. and not the fine-grained value of Perl's $^O variable. The os_type() method will return a string like Windows, Unix, MacOS, VMS, or whatever is appropriate. If you're running on an unknown platform, it will return undef - there shouldn't be many unknown platforms though.


Returns a data structure containing information about any failed prerequisites (of any of the types described above), or undef if all prerequisites are met.

The data structure returned is a hash reference. The top level keys are the type of prerequisite failed, one of "requires", "build_requires", "conflicts", or "recommends". The associated values are hash references whose keys are the names of required (or conflicting) modules. The associated values of those are hash references indicating some information about the failure. For example:

  have => '0.42',
  need => '0.59',
  message => 'Version 0.42 is installed, but we need version 0.59',


  have => '<none>',
  need => '0.59',
  message => 'Prerequisite Foo isn't installed',

This hash has the same structure as the hash returned by the check_installed_status() method, except that in the case of "conflicts" dependencies we change the "need" key to "conflicts" and construct a proper message.


  # Check a required dependency on Foo::Bar
  if ( $m->prereq_failures->{requires}{Foo::Bar} ) { ...

  # Check whether there were any failures
  if ( $m->prereq_failures ) { ...
  # Show messages for all failures
  my $failures = $m->prereq_failures;
  while (my ($type, $list) = each %$failures) {
    while (my ($name, $hash) = each %$list) {
      print "Failure for $name: $hash->{message}\n";

$m->check_installed_status($module, $version)

This method returns a hash reference indicating whether a version dependency on a certain module is satisfied. The $module argument is given as a string like "Data::Dumper" or "perl", and the $version argument can take any of the forms described in prereq above. This allows very fine-grained version checking.

The returned hash reference has the following structure:

  ok => $whether_the_dependency_is_satisfied,
  have => $version_already_installed,
  need => $version_requested, # Same as incoming $version argument
  message => $informative_error_message,

If no version of $module is currently installed, the have value will be the string "<none>". Otherwise the have value will simply be the version of the installed module. Note that this means that if $module is installed but doesn't define a version number, the have value will be undef - this is why we don't use undef for the case when $module isn't installed at all.

$m->check_installed_version($module, $version)

Like check_installed_status(), but simply returns true or false depending on whether module $module statisfies the dependency $version.

If the check succeeds, the return value is the actual version of $module installed on the system. This allows you to do the following:

 my $installed = $m->check_installed_version('DBI', '1.15');
 if ($installed) {
   print "Congratulations, version $installed of DBI is installed.\n";
 } else {
   die "Sorry, you must install DBI.\n";

If the check fails, we return false and set $@ to an informative error message.

If $version is any nontrue value (notably zero) and any version of $module is installed, we return true. In this case, if $module doesn't define a version, or if its version is zero, we return the special value "0 but true", which is numerically zero, but logically true.

In general you might prefer to use check_installed_status if you need detailed information, or this method if you just need a yes/no answer.


There are some general principles at work here. First, each task when building a module is called an "action". These actions are listed above; they correspond to the building, testing, installing, packaging, etc. tasks.

Second, arguments are processed in a very systematic way. Arguments are always key=value pairs. They may be specified at perl Build.PL time (i.e. perl Build.PL sitelib=/my/secret/place), in which case their values last for the lifetime of the Build script. They may also be specified when executing a particular action (i.e. Build test verbose=1), in which case their values last only for the lifetime of that command. Per-action command-line parameters take precedence over parameters specified at perl Build.PL time.

The build process also relies heavily on the Config.pm module, and all the key=value pairs in Config.pm are available in

$self->{config}. If the user wishes to override any of the values in Config.pm, she may specify them like so:

  perl Build.PL config='siteperl=/foo perlpath=/wacky/stuff'

Not the greatest interface, I'm looking for alternatives. Speak now! Maybe:

  perl Build.PL config-siteperl=/foo config-perlpath=/wacky/stuff

or something.

The following build actions are provided by default.

  • help

    This action will simply print out a message that is meant to help you use the build process. It will show you a list of available build actions too.

  • build

    If you run the Build script without any arguments, it runs the build action.

    This is analogous to the MakeMaker 'make all' target. By default it just creates a blib/ directory and copies any .pm and .pod files from your lib/ directory into the blib/ directory. It also compiles any .xs files from lib/ and places them in blib/. Of course, you need a working C compiler (probably the same one that built perl itself) for this to work properly.

    The build action also runs any .PL files in your lib/ directory. Typically these create other files, named the same but without the .PL ending. For example, a file lib/Foo/Bar.pm.PL could create the file lib/Foo/Bar.pm. The .PL files are processed first, so any .pm files (or other kinds that we deal with) will get copied correctly.

    If your .PL scripts don't create any files, or if they create files with unexpected names, or even if they create multiple files, you should tell us that so that we can clean up properly after these created files. Use the PL_files parameter to new():

     PL_files => { 'lib/Foo/Bar_pm.PL' => 'lib/Foo/Bar.pm',
                   'lib/something.PL'  => ['/lib/something', '/lib/else'],
                   'lib/funny.PL'      => [] }

    Note that in contrast to MakeMaker, the build action only (currently) handles .pm, .pod, .PL, and .xs files. They must all be in the lib/ directory, in the directory structure that they should have when installed. We also handle .c files that can be in the place of your choosing - see the c_source argument to new().

    The .xs support is currently in alpha. Please let me know whether it works for you.

  • test

    This will use Test::Harness to run any regression tests and report their results. Tests can be defined in the standard places: a file called test.pl in the top-level directory, or several files ending with .t in a t/ directory.

    If you want tests to be 'verbose', i.e. show details of test execution rather than just summary information, pass the argument verbose=1.

    If you want to run tests under the perl debugger, pass the argument debugger=1.

    In addition, if a file called visual.pl exists in the top-level directory, this file will be executed as a Perl script and its output will be shown to the user. This is a good place to put speed tests or other tests that don't use the Test::Harness format for output.

  • testdb

    This is a synonym for the 'test' action with the debugger=1 argument.

  • clean

    This action will clean up any files that the build process may have created, including the blib/ directory (but not including the _build/ directory and the Build script itself).

  • realclean

    This action is just like the clean action, but also removes the _build directory and the Build script. If you run the realclean action, you are essentially starting over, so you will have to re-create the Build script again.

  • install

    This action will use ExtUtils::Install to install the files from blib/ into the correct system-wide module directory. The directory is determined from the sitelib entry in the Config.pm module. To install into a different directory, pass a different value for the sitelib parameter, like so:

     Build install sitelib=/my/secret/place/

    Alternatively, you could specify the sitelib parameter when you run the Build.PL script:

     perl Build.PL sitelib=/my/secret/place/

    Under normal circumstances, you'll need superuser privileges to install into the default sitelib directory.

  • fakeinstall

    This is just like the install action, but it won't actually do anything, it will just report what it would have done if you had actually run the install action.

  • manifest

    This is an action intended for use by module authors, not people installing modules. It will bring the MANIFEST up to date with the files currently present in the distribution. You may use a MANIFEST.SKIP file to exclude certain files or directories from inclusion in the MANIFEST. MANIFEST.SKIP should contain a bunch of regular expressions, one per line. If a file in the distribution directory matches any of the regular expressions, it won't be included in the MANIFEST.

    The following is a reasonable MANIFEST.SKIP starting point, you can add your own stuff to it:


    See the distcheck and skipcheck actions if you want to find out what the manifest action would do, without actually doing anything.

  • dist

    This action is helpful for module authors who want to package up their module for distribution through a medium like CPAN. It will create a tarball of the files listed in MANIFEST and compress the tarball using GZIP compression.

  • distcheck

    Reports which files are in the build directory but not in the MANIFEST file, and vice versa. (See manifest for details)

  • skipcheck

    Reports which files are skipped due to the entries in the MANIFEST.SKIP file (See manifest for details)

  • distclean

    Performs the 'realclean' action and then the 'distcheck' action.

  • distdir

    Creates a directory called $(DISTNAME)-$(VERSION) (if that directory already exists, it will be removed first). Then copies all the files listed in the MANIFEST file to that directory. This directory is what people will see when they download your distribution and unpack it.

    While performing the 'distdir' action, a file containing various bits of "metadata" will be created. The metadata includes the module's name, version, dependencies, license, and the dynamic_config flag. This file is created as META.yaml in YAML format, so you must have the YAML module installed in order to create it. You should also ensure that the META.yaml file is listed in your MANIFEST - if it's not, a warning will be issued.

  • disttest

    Performs the 'distdir' action, then switches into that directory and runs a perl Build.PL, followed by the 'build' and 'test' actions in that directory.


One advantage of Module::Build is that since it's implemented as Perl methods, you can invoke these methods directly if you want to install a module non-interactively. For instance, the following Perl script will invoke the entire build/install procedure:

 my $m = new Module::Build (module_name => 'MyModule');

If any of these steps encounters an error, it will throw a fatal exception.

You can also pass arguments as part of the build process:

 my $m = new Module::Build (module_name => 'MyModule');
 $m->dispatch('test', verbose => 1);
 $m->dispatch('install', sitelib => '/my/secret/place/');

Building and installing modules in this way skips creating the Build script.


Module::Build creates a class hierarchy conducive to customization. Here is the parent-child class hierarchy in classy ASCII art:

   |   Your::Parent     |  (If you subclass Module::Build)
   /--------------------\  (Doesn't define any functionality
   |   Module::Build    |   of its own - just figures out what
   \--------------------/   other modules to load.)
   /-----------------------------------\  (Some values of $^O may
   |   Module::Build::Platform::$^O    |   define specialized functionality.
   \-----------------------------------/   Otherwise it's ...::Default, a
            |                              pass-through class.)
   |   Module::Build::Base    |  (Most of the functionality of 
   \--------------------------/   Module::Build is defined here.)


Right now, there are two ways to subclass Module::Build. The first way is to create a regular module (in a .pm file) that inherits from Module::Build, and use that module's class instead of using Module::Build directly:

  ------ in Build.PL: ----------
  use lib qw(/nonstandard/library/path);
  use My::Builder;  # Or whatever you want to call it
  my $m = My::Builder->new(module_name => 'Next::Big::Thing');

This is relatively straightforward, and is the best way to do things if your My::Builder class contains lots of code. The create_build_script() method will ensure that the current value of @INC (including the /nonstandard/library/path) is propogated to the Build script, so that My::Builder can be found when running build actions.

For very small additions, Module::Build provides a subclass() method that lets you subclass Module::Build more conveniently, without creating a separate file for your module:

  ------ in Build.PL: ----------
  my $class = Module::Build->subclass
     class => 'My::Builder',
     code => q{
      sub ACTION_foo {
        print "I'm fooing to death!\n";
  my $m = $class->new(module_name => 'Module::Build');

Behind the scenes, this actually does create a .pm file, since the code you provide must persist after Build.PL is run if it is to be very useful.


There are several reasons I wanted to start over, and not just fix what I didn't like about MakeMaker:

  • I don't like the core idea of MakeMaker, namely that make should be involved in the build process. Here are my reasons:


    When a person is installing a Perl module, what can you assume about their environment? Can you assume they have make? No, but you can assume they have some version of Perl.


    When a person is writing a Perl module for intended distribution, can you assume that they know how to build a Makefile, so they can customize their build process? No, but you can assume they know Perl, and could customize that way.

    For years, these things have been a barrier to people getting the build/install process to do what they want.

  • There are several architectural decisions in MakeMaker that make it very difficult to customize its behavior. For instance, when using MakeMaker you do use MakeMaker, but the object created in WriteMakefile() is actually blessed into a package name that's created on the fly, so you can't simply subclass ExtUtils::MakeMaker. There is a workaround MY package that lets you override certain MakeMaker methods, but only certain explicitly predefined (by MakeMaker) methods can be overridden. Also, the method of customization is very crude: you have to modify a string containing the Makefile text for the particular target.

  • It is risky to make major changes to MakeMaker, since it does so many things, is so important, and generally works. Module::Build is an entirely seperate package so that I can work on it all I want, without worrying about backward compatibility.

  • Finally, Perl is said to be a language for system administration. Could it really be the case that Perl isn't up to the task of building and installing software? Even if that software is a bunch of stupid little .pm files that just need to be copied from one place to another? Are you getting riled up yet??

Please contact me if you have any questions or ideas.


The current method of relying on time stamps to determine whether a derived file is out of date isn't likely to scale well, since it requires tracing all dependencies backward, it runs into problems on NFS, and it's just generally flimsy. It would be better to use an MD5 signature or the like, if available. See cons for an example.

The current dependency-checking is prone to errors. You can make 'widowed' files by doing Build, perl Build.PL, and then Build realclean. Should be easy to fix, but it's got me wondering whether the dynamic declaration of dependencies is a good idea.

- make man pages and install them. - append to perllocal.pod - write .packlist in appropriate location (needed for un-install)


Ken Williams, ken@mathforum.org


perl(1), ExtUtils::MakeMaker(3), YAML(3)