NAME
CPAN - query, download and build perl modules from CPAN sites
SYNOPSIS
Interactive mode:
perl -MCPAN -e shell;
Batch mode:
use CPAN;
autobundle, clean, install, make, recompile, test
DESCRIPTION
The CPAN module is designed to automate the make and install of
perl modules and extensions. It includes some searching
capabilities and knows how to use Net::FTP or LWP (or lynx or an
external ftp client) to fetch the raw data from the net.
Modules are fetched from one or more of the mirrored CPAN
(Comprehensive Perl Archive Network) sites and unpacked in a
dedicated directory.
The CPAN module also supports the concept of named and versioned
'bundles' of modules. Bundles simplify the handling of sets of
related modules. See Bundles below.
The package contains a session manager and a cache manager.
There is no status retained between sessions. The session
manager keeps track of what has been fetched, built and
installed in the current session. The cache manager keeps track
of the disk space occupied by the make processes and deletes
excess space according to a simple FIFO mechanism.
For extended searching capabilities there's a plugin for CPAN
available, the CPAN::WAIT manpage. `CPAN::WAIT' is a full-text
search engine that indexes all documents available in CPAN
authors directories. If `CPAN::WAIT' is installed on your
system, the interactive shell of <CPAN.pm> will enable the `wq',
`wr', `wd', `wl', and `wh' commands which send queries to the
WAIT server that has been configured for your installation.
All other methods provided are accessible in a programmer style
and in an interactive shell style.
Interactive Mode
The interactive mode is entered by running
perl -MCPAN -e shell
which puts you into a readline interface. You will have the most
fun if you install Term::ReadKey and Term::ReadLine to enjoy
both history and command completion.
Once you are on the command line, type 'h' and the rest should
be self-explanatory.
The most common uses of the interactive modes are
Searching for authors, bundles, distribution files and modules
There are corresponding one-letter commands `a', `b', `d', and
`m' for each of the four categories and another, `i' for any
of the mentioned four. Each of the four entities is
implemented as a class with slightly differing methods for
displaying an object.
Arguments you pass to these commands are either strings
exactly matching the identification string of an object or
regular expressions that are then matched case-insensitively
against various attributes of the objects. The parser
recognizes a regular expression only if you enclose it between
two slashes.
The principle is that the number of found objects influences
how an item is displayed. If the search finds one item, the
result is displayed with the rather verbose method
`as_string', but if we find more than one, we display each
object with the terse method <as_glimpse>.
make, test, install, clean modules or distributions
These commands take any number of arguments and investigates
what is necessary to perform the action. If the argument is a
distribution file name (recognized by embedded slashes), it is
processed. If it is a module, CPAN determines the distribution
file in which this module is included and processes that,
following any dependencies named in the module's Makefile.PL
(this behavior is controlled by *prerequisites_policy*.)
Any `make' or `test' are run unconditionally. An
install <distribution_file>
also is run unconditionally. But for
install <module>
CPAN checks if an install is actually needed for it and prints
*module up to date* in the case that the distribution file
containing the module doesn't need to be updated.
CPAN also keeps track of what it has done within the current
session and doesn't try to build a package a second time
regardless if it succeeded or not. The `force' command takes
as a first argument the method to invoke (currently: `make',
`test', or `install') and executes the command from scratch.
Example:
cpan> install OpenGL
OpenGL is up to date.
cpan> force install OpenGL
Running make
OpenGL-0.4/
OpenGL-0.4/COPYRIGHT
[...]
A `clean' command results in a
make clean
being executed within the distribution file's working
directory.
readme, look module or distribution
These two commands take only one argument, be it a module or a
distribution file. `readme' unconditionally runs, displaying
the README of the associated distribution file. `Look' gets
and untars (if not yet done) the distribution file, changes to
the appropriate directory and opens a subshell process in that
directory.
Signals
CPAN.pm installs signal handlers for SIGINT and SIGTERM. While
you are in the cpan-shell it is intended that you can press
`^C' anytime and return to the cpan-shell prompt. A SIGTERM
will cause the cpan-shell to clean up and leave the shell
loop. You can emulate the effect of a SIGTERM by sending two
consecutive SIGINTs, which usually means by pressing `^C'
twice.
CPAN.pm ignores a SIGPIPE. If the user sets
inactivity_timeout, a SIGALRM is used during the run of the
`perl Makefile.PL' subprocess.
CPAN::Shell
The commands that are available in the shell interface are
methods in the package CPAN::Shell. If you enter the shell
command, all your input is split by the
Text::ParseWords::shellwords() routine which acts like most
shells do. The first word is being interpreted as the method to
be called and the rest of the words are treated as arguments to
this method. Continuation lines are supported if a line ends
with a literal backslash.
autobundle
`autobundle' writes a bundle file into the `$CPAN::Config-
>{cpan_home}/Bundle' directory. The file contains a list of all
modules that are both available from CPAN and currently
installed within @INC. The name of the bundle file is based on
the current date and a counter.
recompile
recompile() is a very special command in that it takes no
argument and runs the make/test/install cycle with brute force
over all installed dynamically loadable extensions (aka XS
modules) with 'force' in effect. The primary purpose of this
command is to finish a network installation. Imagine, you have a
common source tree for two different architectures. You decide
to do a completely independent fresh installation. You start on
one architecture with the help of a Bundle file produced
earlier. CPAN installs the whole Bundle for you, but when you
try to repeat the job on the second architecture, CPAN responds
with a `"Foo up to date"' message for all modules. So you invoke
CPAN's recompile on the second architecture and you're done.
Another popular use for `recompile' is to act as a rescue in
case your perl breaks binary compatibility. If one of the
modules that CPAN uses is in turn depending on binary
compatibility (so you cannot run CPAN commands), then you should
try the CPAN::Nox module for recovery.
The four `CPAN::*' Classes: Author, Bundle, Module, Distribution
Although it may be considered internal, the class hierarchy does
matter for both users and programmer. CPAN.pm deals with above
mentioned four classes, and all those classes share a set of
methods. A classical single polymorphism is in effect. A
metaclass object registers all objects of all kinds and indexes
them with a string. The strings referencing objects have a
separated namespace (well, not completely separated):
Namespace Class
words containing a "/" (slash) Distribution
words starting with Bundle:: Bundle
everything else Module or Author
Modules know their associated Distribution objects. They always
refer to the most recent official release. Developers may mark
their releases as unstable development versions (by inserting an
underbar into the visible version number), so the really hottest
and newest distribution file is not always the default. If a
module Foo circulates on CPAN in both version 1.23 and 1.23_90,
CPAN.pm offers a convenient way to install version 1.23 by
saying
install Foo
This would install the complete distribution file (say BAR/Foo-
1.23.tar.gz) with all accompanying material. But if you would
like to install version 1.23_90, you need to know where the
distribution file resides on CPAN relative to the authors/id/
directory. If the author is BAR, this might be BAR/Foo-
1.23_90.tar.gz; so you would have to say
install BAR/Foo-1.23_90.tar.gz
The first example will be driven by an object of the class
CPAN::Module, the second by an object of class
CPAN::Distribution.
Programmer's interface
If you do not enter the shell, the available shell commands are
both available as methods (`CPAN::Shell->install(...)') and as
functions in the calling package (`install(...)').
There's currently only one class that has a stable interface -
CPAN::Shell. All commands that are available in the CPAN shell
are methods of the class CPAN::Shell. Each of the commands that
produce listings of modules (`r', `autobundle', `u') also return
a list of the IDs of all modules within the list.
expand($type,@things)
The IDs of all objects available within a program are strings
that can be expanded to the corresponding real objects with
the `CPAN::Shell->expand("Module",@things)' method. Expand
returns a list of CPAN::Module objects according to the
`@things' arguments given. In scalar context it only returns
the first element of the list.
Programming Examples
This enables the programmer to do operations that combine
functionalities that are available in the shell.
# install everything that is outdated on my disk:
perl -MCPAN -e 'CPAN::Shell->install(CPAN::Shell->r)'
# install my favorite programs if necessary:
for $mod (qw(Net::FTP MD5 Data::Dumper)){
my $obj = CPAN::Shell->expand('Module',$mod);
$obj->install;
}
# list all modules on my disk that have no VERSION number
for $mod (CPAN::Shell->expand("Module","/./")){
next unless $mod->inst_file;
# MakeMaker convention for undefined $VERSION:
next unless $mod->inst_version eq "undef";
print "No VERSION in ", $mod->id, "\n";
}
Or if you want to write a cronjob to watch The CPAN, you could
list all modules that need updating:
perl -e 'use CPAN; CPAN::Shell->r;'
If you don't want to get any output if all modules are up to
date, you can parse the output of above command for the
regular expression //modules are up to date// and decide to
mail the output only if it doesn't match. Ick?
If you prefer to do it more in a programmer style in one
single process, maybe something like this suites you better:
# list all modules on my disk that have newer versions on CPAN
for $mod (CPAN::Shell->expand("Module","/./")){
next unless $mod->inst_file;
next if $mod->uptodate;
printf "Module %s is installed as %s, could be updated to %s from CPAN\n",
$mod->id, $mod->inst_version, $mod->cpan_version;
}
If that gives you too much output every day, you maybe only
want to watch for three modules. You can write
for $mod (CPAN::Shell->expand("Module","/Apache|LWP|CGI/")){
as the first line instead. Or you can combine some of the
above tricks:
# watch only for a new mod_perl module
$mod = CPAN::Shell->expand("Module","mod_perl");
exit if $mod->uptodate;
# new mod_perl arrived, let me know all update recommendations
CPAN::Shell->r;
Methods in the four Classes
Cache Manager
Currently the cache manager only keeps track of the build
directory ($CPAN::Config->{build_dir}). It is a simple FIFO
mechanism that deletes complete directories below `build_dir' as
soon as the size of all directories there gets bigger than
$CPAN::Config->{build_cache} (in MB). The contents of this cache
may be used for later re-installations that you intend to do
manually, but will never be trusted by CPAN itself. This is due
to the fact that the user might use these directories for
building modules on different architectures.
There is another directory ($CPAN::Config->{keep_source_where})
where the original distribution files are kept. This directory
is not covered by the cache manager and must be controlled by
the user. If you choose to have the same directory as build_dir
and as keep_source_where directory, then your sources will be
deleted with the same fifo mechanism.
Bundles
A bundle is just a perl module in the namespace Bundle:: that
does not define any functions or methods. It usually only
contains documentation.
It starts like a perl module with a package declaration and a
$VERSION variable. After that the pod section looks like any
other pod with the only difference being that *one special pod
section* exists starting with (verbatim):
=head1 CONTENTS
In this pod section each line obeys the format
Module_Name [Version_String] [- optional text]
The only required part is the first field, the name of a module
(e.g. Foo::Bar, ie. *not* the name of the distribution file).
The rest of the line is optional. The comment part is delimited
by a dash just as in the man page header.
The distribution of a bundle should follow the same convention
as other distributions.
Bundles are treated specially in the CPAN package. If you say
'install Bundle::Tkkit' (assuming such a bundle exists), CPAN
will install all the modules in the CONTENTS section of the pod.
You can install your own Bundles locally by placing a conformant
Bundle file somewhere into your @INC path. The autobundle()
command which is available in the shell interface does that for
you by including all currently installed modules in a snapshot
bundle file.
Prerequisites
If you have a local mirror of CPAN and can access all files with
"file:" URLs, then you only need a perl better than perl5.003 to
run this module. Otherwise Net::FTP is strongly recommended. LWP
may be required for non-UNIX systems or if your nearest CPAN
site is associated with an URL that is not `ftp:'.
If you have neither Net::FTP nor LWP, there is a fallback
mechanism implemented for an external ftp command or for an
external lynx command.
Finding packages and VERSION
This module presumes that all packages on CPAN
* declare their $VERSION variable in an easy to parse manner. This
prerequisite can hardly be relaxed because it consumes far too
much memory to load all packages into the running program just
to determine the $VERSION variable. Currently all programs
that are dealing with version use something like this
perl -MExtUtils::MakeMaker -le \
'print MM->parse_version(shift)' filename
If you are author of a package and wonder if your $VERSION can
be parsed, please try the above method.
* come as compressed or gzipped tarfiles or as zip files and contain
a Makefile.PL (well, we try to handle a bit more, but without
much enthusiasm).
Debugging
The debugging of this module is pretty difficult, because we
have interferences of the software producing the indices on
CPAN, of the mirroring process on CPAN, of packaging, of
configuration, of synchronicity, and of bugs within CPAN.pm.
In interactive mode you can try "o debug" which will list
options for debugging the various parts of the package. The
output may not be very useful for you as it's just a by-product
of my own testing, but if you have an idea which part of the
package may have a bug, it's sometimes worth to give it a try
and send me more specific output. You should know that "o debug"
has built-in completion support.
Floppy, Zip, Offline Mode
CPAN.pm works nicely without network too. If you maintain
machines that are not networked at all, you should consider
working with file: URLs. Of course, you have to collect your
modules somewhere first. So you might use CPAN.pm to put
together all you need on a networked machine. Then copy the
$CPAN::Config->{keep_source_where} (but not $CPAN::Config-
>{build_dir}) directory on a floppy. This floppy is kind of a
personal CPAN. CPAN.pm on the non-networked machines works
nicely with this floppy. See also below the paragraph about CD-
ROM support.
CONFIGURATION
When the CPAN module is installed, a site wide configuration
file is created as CPAN/Config.pm. The default values defined
there can be overridden in another configuration file:
CPAN/MyConfig.pm. You can store this file in
$HOME/.cpan/CPAN/MyConfig.pm if you want, because $HOME/.cpan is
added to the search path of the CPAN module before the use() or
require() statements.
Currently the following keys in the hash reference $CPAN::Config
are defined:
build_cache size of cache for directories to build modules
build_dir locally accessible directory to build modules
index_expire after this many days refetch index files
cpan_home local directory reserved for this package
gzip location of external program gzip
inactivity_timeout breaks interactive Makefile.PLs after this
many seconds inactivity. Set to 0 to never break.
inhibit_startup_message
if true, does not print the startup message
keep_source_where directory in which to keep the source (if we do)
make location of external make program
make_arg arguments that should always be passed to 'make'
make_install_arg same as make_arg for 'make install'
makepl_arg arguments passed to 'perl Makefile.PL'
pager location of external program more (or any pager)
prerequisites_policy
what to do if you are missing module prerequisites
('follow' automatically, 'ask' me, or 'ignore')
scan_cache controls scanning of cache ('atstart' or 'never')
tar location of external program tar
unzip location of external program unzip
urllist arrayref to nearby CPAN sites (or equivalent locations)
wait_list arrayref to a wait server to try (See CPAN::WAIT)
ftp_proxy, } the three usual variables for configuring
http_proxy, } proxy requests. Both as CPAN::Config variables
no_proxy } and as environment variables configurable.
You can set and query each of these options interactively in the
cpan shell with the command set defined within the `o conf'
command:
`o conf <scalar option>'
prints the current value of the *scalar option*
`o conf <scalar option> <value>'
Sets the value of the *scalar option* to *value*
`o conf <list option>'
prints the current value of the *list option* in MakeMaker's
neatvalue format.
`o conf <list option> [shift|pop]'
shifts or pops the array in the *list option* variable
`o conf <list option> [unshift|push|splice] <list>'
works like the corresponding perl commands.
Note on urllist parameter's format
urllist parameters are URLs according to RFC 1738. We do a
little guessing if your URL is not compliant, but if you have
problems with file URLs, please try the correct format. Either:
file://localhost/whatever/ftp/pub/CPAN/
or
file:///home/ftp/pub/CPAN/
urllist parameter has CD-ROM support
The `urllist' parameter of the configuration table contains a
list of URLs that are to be used for downloading. If the list
contains any `file' URLs, CPAN always tries to get files from
there first. This feature is disabled for index files. So the
recommendation for the owner of a CD-ROM with CPAN contents is:
include your local, possibly outdated CD-ROM as a `file' URL at
the end of urllist, e.g.
o conf urllist push file://localhost/CDROM/CPAN
CPAN.pm will then fetch the index files from one of the CPAN
sites that come at the beginning of urllist. It will later check
for each module if there is a local copy of the most recent
version.
Another peculiarity of urllist is that the site that we could
successfully fetch the last file from automatically gets a
preference token and is tried as the first site for the next
request. So if you add a new site at runtime it may happen that
the previously preferred site will be tried another time. This
means that if you want to disallow a site for the next transfer,
it must be explicitly removed from urllist.
SECURITY
There's no strong security layer in CPAN.pm. CPAN.pm helps you
to install foreign, unmasked, unsigned code on your machine. We
compare to a checksum that comes from the net just as the
distribution file itself. If somebody has managed to tamper with
the distribution file, they may have as well tampered with the
CHECKSUMS file. Future development will go towards strong
authentication.
EXPORT
Most functions in package CPAN are exported per default. The
reason for this is that the primary use is intended for the cpan
shell or for oneliners.
POPULATE AN INSTALLATION WITH LOTS OF MODULES
To populate a freshly installed perl with my favorite modules is
pretty easiest by maintaining a private bundle definition file.
To get a useful blueprint of a bundle definition file, the
command autobundle can be used on the CPAN shell command line.
This command writes a bundle definition file for all modules
that are installed for the currently running perl interpreter.
It's recommended to run this command only once and from then on
maintain the file manually under a private name, say
Bundle/my_bundle.pm. With a clever bundle file you can then
simply say
cpan> install Bundle::my_bundle
then answer a few questions and then go out for a coffee.
Maintaining a bundle definition file means to keep track of two
things: dependencies and interactivity. CPAN.pm sometimes fails
on calculating dependencies because not all modules define all
MakeMaker attributes correctly, so a bundle definition file
should specify prerequisites as early as possible. On the other
hand, it's a bit annoying that many distributions need some
interactive configuring. So what I try to accomplish in my
private bundle file is to have the packages that need to be
configured early in the file and the gentle ones later, so I can
go out after a few minutes and leave CPAN.pm unattained.
WORKING WITH CPAN.pm BEHIND FIREWALLS
Thanks to Graham Barr for contributing the following paragraphs
about the interaction between perl, and various firewall
configurations.
Firewalls can be categorized into three basic types.
http firewall
This is where the firewall machine runs a web server and to
access the outside world you must do it via the web server.
If you set environment variables like http_proxy or
ftp_proxy to a values beginning with http:// or in your web
browser you have to set proxy information then you know you
are running a http firewall.
To access servers outside these types of firewalls with perl
(even for ftp) you will need to use LWP.
ftp firewall
This where the firewall machine runs a ftp server. This kind
of firewall will only let you access ftp serves outside the
firewall. This is usually done by connecting to the firewall
with ftp, then entering a username like
"user@outside.host.com"
To access servers outside these type of firewalls with perl
you will need to use Net::FTP.
One way visibility
I say one way visibility as these firewalls try to make
themselve look invisible to the users inside the firewall.
An FTP data connection is normally created by sending the
remote server your IP address and then listening for the
connection. But the remote server will not be able to
connect to you because of the firewall. So for these types
of firewall FTP connections need to be done in a passive
mode.
There are two that I can think off.
SOCKS If you are using a SOCKS firewall you will need to compile
perl and link it with the SOCKS library, this is what is
normally called a ``socksified'' perl. With this
executable you will be able to connect to servers
outside the firewall as if it is not there.
IP Masquerade
This is the firewall implemented in the Linux kernel, it
allows you to hide a complete network behind one IP
address. With this firewall no special compiling is need
as you can access hosts directly.
BUGS
We should give coverage for all of the CPAN and not just the
PAUSE part, right? In this discussion CPAN and PAUSE have become
equal -- but they are not. PAUSE is authors/ and modules/. CPAN
is PAUSE plus the clpa/, doc/, misc/, ports/, src/, scripts/.
Future development should be directed towards a better
integration of the other parts.
If a Makefile.PL requires special customization of libraries,
prompts the user for special input, etc. then you may find CPAN
is not able to build the distribution. In that case, you should
attempt the traditional method of building a Perl module package
from a shell.
AUTHOR
Andreas Koenig <andreas.koenig@anima.de>
SEE ALSO
perl(1), CPAN::Nox(3)