Upgrade ExtUtils::MakeMaker to 7.33_03

[perl5.git] / cpan / ExtUtils-MakeMaker / lib / ExtUtils / MakeMaker / FAQ.pod

package ExtUtils::MakeMaker::FAQ;

our $VERSION = '7.33_03';
$VERSION = eval $VERSION;

1;
__END__

=head1 NAME

ExtUtils::MakeMaker::FAQ - Frequently Asked Questions About MakeMaker

=head1 DESCRIPTION

FAQs, tricks and tips for C<ExtUtils::MakeMaker>.


=head2 Module Installation

=over 4

=item How do I install a module into my home directory?

If you're not the Perl administrator you probably don't have
permission to install a module to its default location. Ways of handling
this with a B<lot> less manual effort on your part are L<perlbrew>
and L<local::lib>.

Otherwise, you can install it for your own use into your home directory
like so:

    # Non-unix folks, replace ~ with /path/to/your/home/dir
    perl Makefile.PL INSTALL_BASE=~

This will put modules into F<~/lib/perl5>, man pages into F<~/man> and
programs into F<~/bin>.

To ensure your Perl programs can see these newly installed modules,
set your C<PERL5LIB> environment variable to F<~/lib/perl5> or tell
each of your programs to look in that directory with the following:

    use lib "$ENV{HOME}/lib/perl5";

or if $ENV{HOME} isn't set and you don't want to set it for some
reason, do it the long way.

    use lib "/path/to/your/home/dir/lib/perl5";

=item How do I get MakeMaker and Module::Build to install to the same place?

Module::Build, as of 0.28, supports two ways to install to the same
location as MakeMaker.

We highly recommend the install_base method, its the simplest and most
closely approximates the expected behavior of an installation prefix.

1) Use INSTALL_BASE / C<--install_base>

MakeMaker (as of 6.31) and Module::Build (as of 0.28) both can install
to the same locations using the "install_base" concept.  See
L<ExtUtils::MakeMaker/INSTALL_BASE> for details.  To get MM and MB to
install to the same location simply set INSTALL_BASE in MM and
C<--install_base> in MB to the same location.

    perl Makefile.PL INSTALL_BASE=/whatever
    perl Build.PL    --install_base /whatever

This works most like other language's behavior when you specify a
prefix.  We recommend this method.

2) Use PREFIX / C<--prefix>

Module::Build 0.28 added support for C<--prefix> which works like
MakeMaker's PREFIX.

    perl Makefile.PL PREFIX=/whatever
    perl Build.PL    --prefix /whatever

We highly discourage this method.  It should only be used if you know
what you're doing and specifically need the PREFIX behavior.  The
PREFIX algorithm is complicated and focused on matching the system
installation.

=item How do I keep from installing man pages?

Recent versions of MakeMaker will only install man pages on Unix-like
operating systems.

For an individual module:

        perl Makefile.PL INSTALLMAN1DIR=none INSTALLMAN3DIR=none

If you want to suppress man page installation for all modules you have
to reconfigure Perl and tell it 'none' when it asks where to install
man pages.


=item How do I use a module without installing it?

Two ways.  One is to build the module normally...

        perl Makefile.PL
        make
        make test

...and then use L<blib> to point Perl at the built but uninstalled module:

        perl -Mblib script.pl
        perl -Mblib -e '...'

The other is to install the module in a temporary location.

        perl Makefile.PL INSTALL_BASE=~/tmp
        make
        make test
        make install

And then set PERL5LIB to F<~/tmp/lib/perl5>.  This works well when you
have multiple modules to work with.  It also ensures that the module
goes through its full installation process which may modify it.
Again, L<local::lib> may assist you here.

=item How can I organize tests into subdirectories and have them run?

Let's take the following test directory structure:

    t/foo/sometest.t
    t/bar/othertest.t
    t/bar/baz/anothertest.t

Now, inside of the C<WriteMakeFile()> function in your F<Makefile.PL>, specify
where your tests are located with the C<test> directive:

    test => {TESTS => 't/*.t t/*/*.t t/*/*/*.t'}

The first entry in the string will run all tests in the top-level F<t/> 
directory. The second will run all test files located in any subdirectory under
F<t/>. The third, runs all test files within any subdirectory within any other
subdirectory located under F<t/>.

Note that you do not have to use wildcards. You can specify explicitly which
subdirectories to run tests in:

    test => {TESTS => 't/*.t t/foo/*.t t/bar/baz/*.t'}

=item PREFIX vs INSTALL_BASE from Module::Build::Cookbook

The behavior of PREFIX is complicated and depends closely on how your
Perl is configured. The resulting installation locations will vary
from machine to machine and even different installations of Perl on the
same machine.  Because of this, its difficult to document where prefix
will place your modules.

In contrast, INSTALL_BASE has predictable, easy to explain installation
locations.  Now that Module::Build and MakeMaker both have INSTALL_BASE
there is little reason to use PREFIX other than to preserve your existing
installation locations. If you are starting a fresh Perl installation we
encourage you to use INSTALL_BASE. If you have an existing installation
installed via PREFIX, consider moving it to an installation structure
matching INSTALL_BASE and using that instead.

=item Generating *.pm files with substitutions eg of $VERSION

If you want to configure your module files for local conditions, or to
automatically insert a version number, you can use EUMM's C<PL_FILES>
capability, where it will automatically run each F<*.PL> it finds to
generate its basename. For instance:

    # Makefile.PL:
    require 'common.pl';
    my $version = get_version();
    my @pms = qw(Foo.pm);
    WriteMakefile(
      NAME => 'Foo',
      VERSION => $version,
      PM => { map { ($_ => "\$(INST_LIB)/$_") } @pms },
      clean => { FILES => join ' ', @pms },
    );

    # common.pl:
    sub get_version { '0.04' }
    sub process { my $v = get_version(); s/__VERSION__/$v/g; }
    1;

    # Foo.pm.PL:
    require 'common.pl';
    $_ = join '', <DATA>;
    process();
    my $file = shift;
    open my $fh, '>', $file or die "$file: $!";
    print $fh $_;
    __DATA__
    package Foo;
    our $VERSION = '__VERSION__';
    1;

You may notice that C<PL_FILES> is not specified above, since the default
of mapping each .PL file to its basename works well.

If the generated module were architecture-specific, you could replace
C<$(INST_LIB)> above with C<$(INST_ARCHLIB)>, although if you locate
modules under F<lib>, that would involve ensuring any C<lib/> in front
of the module location were removed.

=back

=head2 Common errors and problems

=over 4

=item "No rule to make target `/usr/lib/perl5/CORE/config.h', needed by `Makefile'"

Just what it says, you're missing that file.  MakeMaker uses it to
determine if perl has been rebuilt since the Makefile was made.  It's
a bit of a bug that it halts installation.

Some operating systems don't ship the CORE directory with their base
perl install.  To solve the problem, you likely need to install a perl
development package such as perl-devel (CentOS, Fedora and other
Redhat systems) or perl (Ubuntu and other Debian systems).

=back

=head2 Philosophy and History

=over 4

=item Why not just use <insert other build config tool here>?

Why did MakeMaker reinvent the build configuration wheel?  Why not
just use autoconf or automake or ppm or Ant or ...

There are many reasons, but the major one is cross-platform
compatibility.

Perl is one of the most ported pieces of software ever.  It works on
operating systems I've never even heard of (see perlport for details).
It needs a build tool that can work on all those platforms and with
any wacky C compilers and linkers they might have.

No such build tool exists.  Even make itself has wildly different
dialects.  So we have to build our own.


=item What is Module::Build and how does it relate to MakeMaker?

Module::Build is a project by Ken Williams to supplant MakeMaker.
Its primary advantages are:

=over 8

=item * pure perl.  no make, no shell commands

=item * easier to customize

=item * cleaner internals

=item * less cruft

=back

Module::Build was long the official heir apparent to MakeMaker.  The
rate of both its development and adoption has slowed in recent years,
though, and it is unclear what the future holds for it.  That said,
Module::Build set the stage for I<something> to become the heir to
MakeMaker.  MakeMaker's maintainers have long said that it is a dead
end and should be kept functioning, while being cautious about extending
with new features.

=back

=head2 Module Writing

=over 4

=item How do I keep my $VERSION up to date without resetting it manually?

Often you want to manually set the $VERSION in the main module
distribution because this is the version that everybody sees on CPAN
and maybe you want to customize it a bit.  But for all the other
modules in your dist, $VERSION is really just bookkeeping and all that's
important is it goes up every time the module is changed.  Doing this
by hand is a pain and you often forget.

Probably the easiest way to do this is using F<perl-reversion> in
L<Perl::Version>:

  perl-reversion -bump

If your version control system supports revision numbers (git doesn't
easily), the simplest way to do it automatically is to use its revision
number (you are using version control, right?).

In CVS, RCS and SVN you use $Revision$ (see the documentation of your
version control system for details).  Every time the file is checked
in the $Revision$ will be updated, updating your $VERSION.

SVN uses a simple integer for $Revision$ so you can adapt it for your
$VERSION like so:

    ($VERSION) = q$Revision$ =~ /(\d+)/;

In CVS and RCS version 1.9 is followed by 1.10.  Since CPAN compares
version numbers numerically we use a sprintf() to convert 1.9 to 1.009
and 1.10 to 1.010 which compare properly.

    $VERSION = sprintf "%d.%03d", q$Revision$ =~ /(\d+)\.(\d+)/g;

If branches are involved (ie. $Revision: 1.5.3.4$) it's a little more
complicated.

    # must be all on one line or MakeMaker will get confused.
    $VERSION = do { my @r = (q$Revision$ =~ /\d+/g); sprintf "%d."."%03d" x $#r, @r };

In SVN, $Revision$ should be the same for every file in the project so
they would all have the same $VERSION.  CVS and RCS have a different
$Revision$ per file so each file will have a different $VERSION.
Distributed version control systems, such as SVK, may have a different
$Revision$ based on who checks out the file, leading to a different $VERSION
on each machine!  Finally, some distributed version control systems, such
as darcs, have no concept of revision number at all.


=item What's this F<META.yml> thing and how did it get in my F<MANIFEST>?!

F<META.yml> is a module meta-data file pioneered by Module::Build and
automatically generated as part of the 'distdir' target (and thus
'dist').  See L<ExtUtils::MakeMaker/"Module Meta-Data">.

To shut off its generation, pass the C<NO_META> flag to C<WriteMakefile()>.


=item How do I delete everything not in my F<MANIFEST>?

Some folks are surprised that C<make distclean> does not delete
everything not listed in their MANIFEST (thus making a clean
distribution) but only tells them what they need to delete.  This is
done because it is considered too dangerous.  While developing your
module you might write a new file, not add it to the MANIFEST, then
run a C<distclean> and be sad because your new work was deleted.

If you really want to do this, you can use
C<ExtUtils::Manifest::manifind()> to read the MANIFEST and File::Find
to delete the files.  But you have to be careful.  Here's a script to
do that.  Use at your own risk.  Have fun blowing holes in your foot.

    #!/usr/bin/perl -w

    use strict;

    use File::Spec;
    use File::Find;
    use ExtUtils::Manifest qw(maniread);

    my %manifest = map  {( $_ => 1 )}
                   grep { File::Spec->canonpath($_) }
                        keys %{ maniread() };

    if( !keys %manifest ) {
        print "No files found in MANIFEST.  Stopping.\n";
        exit;
    }

    find({
          wanted   => sub {
              my $path = File::Spec->canonpath($_);

              return unless -f $path;
              return if exists $manifest{ $path };

              print "unlink $path\n";
              unlink $path;
          },
          no_chdir => 1
         },
         "."
    );


=item Which tar should I use on Windows?

We recommend ptar from Archive::Tar not older than 1.66 with '-C' option.

=item Which zip should I use on Windows for '[ndg]make zipdist'?

We recommend InfoZIP: L<http://www.info-zip.org/Zip.html>


=back

=head2 XS

=over 4

=item How do I prevent "object version X.XX does not match bootstrap parameter Y.YY" errors?

XS code is very sensitive to the module version number and will
complain if the version number in your Perl module doesn't match.  If
you change your module's version # without rerunning Makefile.PL the old
version number will remain in the Makefile, causing the XS code to be built
with the wrong number.

To avoid this, you can force the Makefile to be rebuilt whenever you
change the module containing the version number by adding this to your
WriteMakefile() arguments.

    depend => { '$(FIRST_MAKEFILE)' => '$(VERSION_FROM)' }


=item How do I make two or more XS files coexist in the same directory?

Sometimes you need to have two and more XS files in the same package.
There are three ways: C<XSMULTI>, separate directories, and bootstrapping
one XS from another.

=over 8

=item XSMULTI

Structure your modules so they are all located under F<lib>, such that
C<Foo::Bar> is in F<lib/Foo/Bar.pm> and F<lib/Foo/Bar.xs>, etc. Have your
top-level C<WriteMakefile> set the variable C<XSMULTI> to a true value.

Er, that's it.

=item Separate directories

Put each XS files into separate directories, each with their own
F<Makefile.PL>. Make sure each of those F<Makefile.PL>s has the correct
C<CFLAGS>, C<INC>, C<LIBS> etc. You will need to make sure the top-level
F<Makefile.PL> refers to each of these using C<DIR>.

=item Bootstrapping

Let's assume that we have a package C<Cool::Foo>, which includes
C<Cool::Foo> and C<Cool::Bar> modules each having a separate XS
file. First we use the following I<Makefile.PL>:

  use ExtUtils::MakeMaker;

  WriteMakefile(
      NAME              => 'Cool::Foo',
      VERSION_FROM      => 'Foo.pm',
      OBJECT              => q/$(O_FILES)/,
      # ... other attrs ...
  );

Notice the C<OBJECT> attribute. MakeMaker generates the following
variables in I<Makefile>:

  # Handy lists of source code files:
  XS_FILES= Bar.xs \
        Foo.xs
  C_FILES = Bar.c \
        Foo.c
  O_FILES = Bar.o \
        Foo.o

Therefore we can use the C<O_FILES> variable to tell MakeMaker to use
these objects into the shared library.

That's pretty much it. Now write I<Foo.pm> and I<Foo.xs>, I<Bar.pm>
and I<Bar.xs>, where I<Foo.pm> bootstraps the shared library and
I<Bar.pm> simply loading I<Foo.pm>.

The only issue left is to how to bootstrap I<Bar.xs>. This is done
from I<Foo.xs>:

  MODULE = Cool::Foo PACKAGE = Cool::Foo

  BOOT:
  # boot the second XS file
  boot_Cool__Bar(aTHX_ cv);

If you have more than two files, this is the place where you should
boot extra XS files from.

The following four files sum up all the details discussed so far.

  Foo.pm:
  -------
  package Cool::Foo;

  require DynaLoader;

  our @ISA = qw(DynaLoader);
  our $VERSION = '0.01';
  bootstrap Cool::Foo $VERSION;

  1;

  Bar.pm:
  -------
  package Cool::Bar;

  use Cool::Foo; # bootstraps Bar.xs

  1;

  Foo.xs:
  -------
  #include "EXTERN.h"
  #include "perl.h"
  #include "XSUB.h"

  MODULE = Cool::Foo  PACKAGE = Cool::Foo

  BOOT:
  # boot the second XS file
  boot_Cool__Bar(aTHX_ cv);

  MODULE = Cool::Foo  PACKAGE = Cool::Foo  PREFIX = cool_foo_

  void
  cool_foo_perl_rules()

      CODE:
      fprintf(stderr, "Cool::Foo says: Perl Rules\n");

  Bar.xs:
  -------
  #include "EXTERN.h"
  #include "perl.h"
  #include "XSUB.h"

  MODULE = Cool::Bar  PACKAGE = Cool::Bar PREFIX = cool_bar_

  void
  cool_bar_perl_rules()

      CODE:
      fprintf(stderr, "Cool::Bar says: Perl Rules\n");

And of course a very basic test:

  t/cool.t:
  --------
  use Test;
  BEGIN { plan tests => 1 };
  use Cool::Foo;
  use Cool::Bar;
  Cool::Foo::perl_rules();
  Cool::Bar::perl_rules();
  ok 1;

This tip has been brought to you by Nick Ing-Simmons and Stas Bekman.

An alternative way to achieve this can be seen in L<Gtk2::CodeGen>
and L<Glib::CodeGen>.

=back

=back

=head1 DESIGN

=head2 MakeMaker object hierarchy (simplified)

What most people need to know (superclasses on top.)

        ExtUtils::MM_Any
                |
        ExtUtils::MM_Unix
                |
        ExtUtils::MM_{Current OS}
                |
        ExtUtils::MakeMaker
                |
               MY

The object actually used is of the class MY which allows you to
override bits of MakeMaker inside your Makefile.PL by declaring
MY::foo() methods.

=head2 MakeMaker object hierarchy (real)

Here's how it really works:

                                    ExtUtils::MM_Any
                                            |
                                    ExtUtils::MM_Unix
                                            |
    ExtUtils::Liblist::Kid          ExtUtils::MM_{Current OS} (if necessary)
          |                                          |
    ExtUtils::Liblist     ExtUtils::MakeMaker        |
                    |     |                          |   
                    |     |   |-----------------------
                   ExtUtils::MM
                   |          |
        ExtUtils::MY         MM (created by ExtUtils::MM)
        |                                   |
        MY (created by ExtUtils::MY)        |
                    .                       |
                 (mixin)                    |
                    .                       |
               PACK### (created each call to ExtUtils::MakeMaker->new)

NOTE: Yes, this is a mess.  See
L<http://archive.develooper.com/makemaker@perl.org/msg00134.html>
for some history.

NOTE: When ExtUtils::MM is loaded it chooses a superclass for MM from
amongst the ExtUtils::MM_* modules based on the current operating
system.

NOTE: ExtUtils::MM_{Current OS} represents one of the ExtUtils::MM_*
modules except ExtUtils::MM_Any chosen based on your operating system.

NOTE: The main object used by MakeMaker is a PACK### object, *not*
ExtUtils::MakeMaker.  It is, effectively, a subclass of MY,
ExtUtils::Makemaker, ExtUtils::Liblist and ExtUtils::MM_{Current OS}

NOTE: The methods in MY are simply copied into PACK### rather than
MY being a superclass of PACK###.  I don't remember the rationale.

NOTE: ExtUtils::Liblist should be removed from the inheritance hiearchy
and simply be called as functions.

NOTE: Modules like File::Spec and Exporter have been omitted for clarity.


=head2 The MM_* hierarchy

                                MM_Win95   MM_NW5
                                     \      /
 MM_BeOS  MM_Cygwin  MM_OS2  MM_VMS  MM_Win32  MM_DOS  MM_UWIN
       \        |      |         |        /      /      /
        ------------------------------------------------
                           |       |
                        MM_Unix    |
                              |    |
                              MM_Any

NOTE: Each direct MM_Unix subclass is also an MM_Any subclass.  This
is a temporary hack because MM_Unix overrides some MM_Any methods with
Unix specific code.  It allows the non-Unix modules to see the
original MM_Any implementations.

NOTE: Modules like File::Spec and Exporter have been omitted for clarity.

=head1 PATCHING

If you have a question you'd like to see added to the FAQ (whether or
not you have the answer) please either:

=over 2

=item * make a pull request on the MakeMaker github repository

=item * raise a issue on the MakeMaker github repository

=item * file an RT ticket

=item * email makemaker@perl.org

=back

=head1 AUTHOR

The denizens of makemaker@perl.org.

=head1 SEE ALSO

L<ExtUtils::MakeMaker>

=cut