Perl Package Manager (PPM)


Preamble

Perl has a strong historical connection with Unix, and the distribution
methods that currently exist serve this community well: 'do-it-yourself'
types who are accustomed to obtaining their software over the Internet and
building it from source.  CPAN has greatly simplified the process of locating
and building Perl extensions from source, and having this well-designed
source repository has undoubtedly contributed to Perl's past and present
popularity.

But platforms such as Macintosh and Windows make up the vast majority of
desktop workstations, as well as an ever-growing percentage of the server
market, and the users of these platforms typically have a different set
of expectations regarding their software than does an experienced Perl
programmer.  These users typically install binary versions of software
onto their systems, usually with the aid of an installation program.

We now have a number of binary distributions for the Perl core for these
platforms.  But one problem remains: how to give users on these platforms
access to the vast library of Perl modules and extensions that have been
written and made available on CPAN?  This document describes how we intend
to solve this problem for Windows users.  We offer it as a proposal for a
method of providing pre-built versions of Perl modules and extensions in
general, and welcome comments from the Perl community on how to make it as
useful as possible.


Introduction

PPM is a program and API intended to simplify the process of locating,
installing, removing, upgrading and otherwise maintaining pre-built Perl
modules and extensions (packages).

To obtain information about available software packages, PPM refers to files
that conform to the Open Software Description (OSD) specification. These
description files are referred to as Perl Package Description (PPD) files,
and are written in Extensible Markup Language (XML).  The glossary at the
end of this document contains further information about these terms.

Each package has an individual PPD file describing attributes such as package
location, installation script, abstract, licensing information, etc.  The
person providing the pre-built version of the package for a particular
platform generates the PPD file during the build process ("make ppd").  It is
constructed using information available in the package's Makefile.pl file or
its POD.

The PPD is then uploaded to a PPD repository, using the Perl Authors Upload
Server (PAUSE) as an authentication scheme, where is it merged with other
PPDs describing available pre-built packages for various platforms.  Since
the PPD files are written in XML, the merging can be automatically performed
by scripts, reducing the burden on repository maintainers.  The merged PPD
is then exposed on the repository to users of PPM.


PPM Usage

In a typical scenario, PPM will look at a central PPD repository to obtain a
list of available packages.  The PPD repository may be a remote location
accessed using Internet protocols such as HTTP or FTP, or a locally mounted
(or shared) directory.  Since the PPD for a package contains the information
needed to retrieve the installation archive, the software packages themselves
need not be at the same location as the PPDs.

To perform a package installation, PPM consults the package's PPD to determine
the location of the pre-built package for the platform on which it is running.
The package will usually be in
    'package-version-bin-binversion-archname-compiler.tar.gz'
format (the binary upload naming convention suggested by Andreas Koenig in
'CPAN/modules/05bindist.convention.html'), but may be in any format the
install script is written to handle.  The file is retrieved and placed in a
local 'build' directory, where it is uncompressed and unarchived if necessary.
PPM will then use installation instructions provided in the PPD in the form
of a Perl script.  If there is no installation script, the required files are
copied to the appropriate locations on the local system.

PPM performs basic verifying and upgrading by comparing information from the
PPD available at the repository to information saved during a previous
installation of the same package.  PPM provides the ability to verify that
the most recent version of a package is installed, and to upgrade a specific
package (or all currently installed packages) if it is outdated.


Features

PPM provides the ability to search for specific package capabilities using
keywords in the abstract, title, or other attributes of the PPD file. This
makes, for example, locating a package that provides a broad range of
functionality (e.g., 'LWP') much easier than by using existing tools. PPM
can perform simple installations of pre-built distributions for specific
platforms.  This greatly simplifies the task of installing and upgrading
packages on platforms for which there is no native 'make' command.

PPM has the concept of software dependencies, and thus the ability to
automatically install any additional packages on which a package depends.
This concept of a dependency extends to commands such as 'make', so that
a package that is in source form will not be installed on a system without
a 'make' command.

Finally, PPM has been designed to be used with any of a number of user
interfaces: command-line, interactive text-mode, GUI, Web browser, etc. This
allows for the development of tools to appeal to a broad range of users with
various levels of sophistication.


Samples

The following is a sample PPD file.

<SOFTPKG NAME="Math-MatrixBool" VERSION="4,2,0,0">
    <TITLE>Math-MatrixBool</TITLE>
    <ABSTRACT>Easy manipulation of matrices of booleans (Boolean Algebra)</ABSTRACT>
    <AUTHOR>Steffen Beyer (sb@sdm.de)</AUTHOR>
    <LICENSE HREF="http://www.ActiveState.com/packages/Math-MatrixBool/license.html" />
    <IMPLEMENTATION>
        <OS VALUE="WinNT" />
        <OS VALUE="Win95" />
        <PROCESSOR VALUE="x86" />
        <CODEBASE HREF="http://www.ActiveState.com/packages/Math-MatrixBool/Math-MatrixBool-4.2-bin-1-Win32.tar.gz" />
        <DEPENDENCY NAME="Bit-Vector" />
        <INSTALL>
        </INSTALL>
        <UNINSTALL>
        </UNINSTALL>
    </IMPLEMENTATION>
    
    <IMPLEMENTATION>
        <DEPENDENCY NAME="Bit-Vector" />
        <CODEBASE HREF="&CPAN;/CPAN/modules/by-module/Math/Math-MatrixBool-4.2.tar.gz" />
        <INSTALL>
            system("make"); ;;
            system("make test"); ;;
            system("make install"); ;;
        </INSTALL>
    </IMPLEMENTATION>
<</SOFTPKG>


In this example, the package 'Math-MatrixBool' is installed as a pre-built
extension on Windows NT or 95 on an Intel x86 chip, since no install script
has been provided.  On other platforms, the source package is retrieved from
a CPAN mirror site and built according to the installation script. The
dependency on 'Bit-Vector' means that 'Math-MatrixBool' will only be
installed if 'Bit-Vector' has been, or can be, installed.


Issues

1. How to reliably provide uninstall capability?  In practice, what is the
   likelihood of filename conflicts with existing modules?
   
   Possible solution: The repository could maintain a list of all installed
   files for all packages, and as part of the process of adding a new
   IMPLEMENTATION to a PPD, we could check that it causes no conflicts
   with existing files.  This would still be subject to potential problems 
   if the user installs packages outside of PPM.
   
   Alternatively, we could use the .packlist scheme, but do the removing 
   ourselves rather than using uninstall, only removing files that have 
   been installed under the package's own directory, ignoring others.  
   This would leave the possibility of orphaned files remaining.
   
2. How to retain configuration information (e.g. lib/site/Net/Config.pm)
   for installed packages during an upgrade?

   Possible solution: The existing configuration could be saved by commands
   in the /INSTALL script in the PPD.


Glossary of terms

dependency
    - a software package necessary for installation or operation of another
      package

OSD 
    - Open Software Description (see http://www.w3.org/TR/NOTE-OSD.html)

PAUSE
    - Perl Authors Upload Server (see CPAN/modules/04pause.html)
    
PPD
    - Perl Package Description file

PPD repository
    - an Internet location or local directory, to which PPM looks for PPD files

PPM
    - Perl Package Manager

pre-built package
    - a software package consisting of ready-to-install files (e.g. compiled
      C objects) for use on a specific platform

XML
    - Extensible Markup Language (see http://www.sil.org/sgml/xml.html)