History of CPAN

CPAN History

Version 0.3

This document is a history of the Comprehensive Perl Archive Network, or CPAN. CPAN is a collection of open source Perl libraries, contributed by 6000+ developers over a period of 19 years. The archive is mirrored on 250+ servers around the world.

But this is also a history of the other services that make up the CPAN ecosystem, and the people who made it all happen.

This is the first version of this history, and it's been pulled together in something of a rush. Apologies to all those who should be here, but aren't yet. The master copy is on github, please send pull requests with additions and corrections. You can just add topics to the todo list at the end of this document.

The perl 4 days

With perl 4, we didn't have modules, we had .pl library files. For example, before the CGI module we had cgi-lib.pl. These libraries were collections of functions, which you included in your code with:

require 'cgi-lib.pl';

Perl 4.036 was the final release of Perl 4, on 5th Feb 1993. It came with 31 libraries, for example getopts.pl for processing command-line arguments.

There were also many third-party libraries, written by Perl users around the world. These were collected together in various ftp sites. There wasn't any single master site. The big ones were funet.fi, metronet, coombs, and ??

The libwww-perl project started at the first WWW conference in 1994 (Geneva) where Martijn Koster (MAK) discussed MOMspider with Roy Fielding (later of REST fame). Martijn suggested that it would be a good thing if the reusable components of this program were broken out into a library. The result was the libwww-perl library for perl4 that Roy maintained

Perl version 1 included tests that were based on what became the test anything protocol (TAP): printing "ok #" or "not ok #" to STDOUT. Up to Perl 5, tests just effectively did print "ok 1\n".

The USENET group comp.lang.perl is where most public perl discussion happened at this time, and where people announced new libraries, and where to find them.


The perl-packrats mailing list was created by Bill Middleton (Bill) on 2nd December 1993, for discussion between people who were maintaining ftp sites for perl. Founding members: Bill, Tom Christiansen (TOMC), Henk Penning (Henk), Jarkko Hietaniemi (JHI), Stephen Potter (SPP), Lee McLoughlin (LMJM), and Mark ?? (coombs).

On December 6th, Bill emailed a list of topics for discussion, which included "master archive". On the same day, an email from Jared Rhine (JARED) was forwarded to the list, in which he wrote "I propose that we cooperate to create a unified structure, much like the CTAN project". CTAN is the Comprehensive TeX Archive Network, a collection of packages for the TeX typesetting system.

On December 7th, Bill emailed the CTAN admins asking for any advice and any tools for running a CTAN-like thing. Sebastian Rahtz replied with some tips ("an agreed structure is crucial"), also saying that CTAN mostly ran on "luck and spit" :-)

On 10th December, JARED mailed a more concrete and detailed proposal, and used the name "CPAN" for the first time.

From there they got into more pragmatic concerns, like mirroring, and how to handle the transition to a standard structure without breaking all existing references to the different structures (symlinks!). Remember this was 1993, so there was discussion about what interfaces, protocols, and tools to use: archie, gopher, WAIS, http, ftp, ...

Then Christmas came, and things cooled for a while. Partly because everyone was playing with alpha and beta (and gamma, on Oct 14th 1994!) releases of Perl 5...

As people were kicking the tyres of Perl 5, they started writing modules. In August 1994, Tim Bunce (TIMB) created the Module List, which was a manually edited text file containing an index of all known Perl 5 modules. I don't have a copy of the first-ever list, but revision 2.2 was posted 7th March 1995. For a long time this was the place to look, to find out what modules were available for Perl 5.

In the Perl 4 days, using a database typically required you to compile a special version of Perl for each kind of database, each with an incompatible API. Buzz Moschetti, the author of the Interbase perl, emailed the other authors on 29th September 1992 to say "it might be a swell idea to standardize our perl-DB-API interfaces". They agreed and the "DBperl" project was born. TIMB joined the effort a few days later. After a year or so, TIMB became the lead for implementing the DBperl project for Perl 5, later renamed "DBI", and DBD::Oracle, the first driver. The first public release of DBI was on 12th October 1994, 5 days before the first public release of Perl 5.

Perl 5

Perl 5 was first released 17th October 1994, introducing modules, which could be object-oriented (OO) as well as collections of functions. That first release included ExtUtils::MakeMaker, which effectively defined what distros would look like.

Releasing a module generally involved putting it on an ftp site somewhere, then posting about it to USENET group comp.lang.perl.

Test::Harness was written by ANDK and TIMB, and came with the first release of Perl 5. This allowed modules to write tests based around Perl's core testing model.

12th Feb 1995, JHI emailed packrats saying "let's do this CPAN thing", proposing a directory structure, and that he was going to get on with it.

For me this is a key moment. It needed someone to say "right, I'm gonna do it"

"Nothing gets done without someone doing it" — Larry Wall 1994-11-08 (ie not in response to this).

JARED followed up on the 20th with some proposed principles (easy to maintain, easy to submit, things should have an abstract, tagging ahem).

March 17th Ian Burrell (IBURRELL) posted some questions and thoughts about how scripts and modules would be indexed.

15th April, TOMC posted a list of extensions ("module" and "extension" were both widely used), which he'd pulled together, saying they were hard to find. Kenneth Albanowski (KJALB) replied it would be easier if authors of modules had "some place standard to deposit them", and this place could be used to populate the repositories.

17th April, Andreas König (ANDK) suggested a "MASTER site for modules", which would be mirrored by the official repositories. He said "I have written 33% of the programs needed". Up to this point different modules and packages had different master sites.

This is the second key "I'm gonna do it" moment.

29th April: ANDK emailed with details of his "first cut". This wasn't for uploads at this point: it was a central database of URLs pointing to the master copy of perl modules (I think!). The goal at this point was the automated maintenance of TIMB's module list, or at least a central database that held the information needed to build the list.

May 4th: TIMB announced an incoming/ directory on ftp.demon.co.uk where authors could upload modules (and other perl things).

May 12th: TIMB expressed some frustration with the lack of progress, and desire to get on with things. Bill noted that there was no-one "in charge", and that maybe part of the problem was trying to cope with everything, so why not focus on modules and perl source. JHI apologised for the delay in his (CPAN) efforts, but said his Master's thesis was taking priority, for just a little while longer...

In Perl 4, cgi-lib.pl was the de facto standard used when creating CGI scripts. In June 1995 Lincoln Stein (LDS) created the CGI module, a 'port' of cgi-lib.pl. For many years it was the way people created dynamic web interfaces in Perl. So much so that it was "moved into the core", and bundled with Perl from 5.004 onwards (it was dropped from the core in Perl 5.20, in 2014).

CPAN and PAUSE take shape

5th June, JHI announced "CPAN v0.000" to perl-packrats, saying it was built from "about a dozen ftp sites". There was still discussion about what the "proper name" should be (REEPH, ARK, EYEKAP).

1st August, JHI announced the "CPAN private showing".

15th August, ANDK emailed saying that he'd rethought what was needed for upload, asking for input, but that he was coding NOW. He started working on a database and upload server. During his development the first true upload was logged in the database: Symdump 1.20, on 16th August 1995. Initially people were uploading files into a directory structure that mirrored the module list.

17th August, KJALB said "Username directory seem to make sense in general", to which ANDK replied "!!!! light bulb on !!!!", to which KJALB in turn replied, "I think an author tree is a good idea, especially as automatically generating a symlinked module tree from the author tree and the module list should be quite simple".

If you look at the early PAUSE uploads, most days there are just one or two uploads. But on 20th August it looks as if 14 different people uploaded 28 releases. But in fact this was ANDK testing: he grabbed releases from the early pioneers, and was doing 'uploads by proxy'. Those early pioneers: AMOSS, ANDK, FMC, GBARR, GBOSS, ILYAZ, JACKS, JKAST, KJALB, MICB, NWINT, SPIDB, TYEMQ, WPS.

Those early uploads were proper modules, with a Makefile.PL based on ExtUtils::MakeMaker.

Aug 21st, ANDK emailed an update on the "module list server", which was starting to sound a bit like modern PAUSE by this point. It used Msql for the database and LWP for network stuff. Author directories were CPAN/mod/pml/byauthor/ (what did PML stand for? Perl module list?)

Sep 20th: ANDK emailed an update, where I think he established himself and TIMB as the first "PAUSE admins" (though they weren't referred to as that). At this point there was a close linkage between the module list and the nascent PAUSE.

Sep 25th: TIMB asked JHI and ANDK if they were ready to announce CPAN and the module upload server?

Sep 27th: JARED proposed a change in directory structure for the module part of CPAN, to have three key trees: by-author, by-category, and by-module. Category was thematic (eg www would contain LWP), where by-module was by the namespace of modules (so LWP would be under LWP/). In response ANDK suggested pml/byauthor/.

29th Sep: JHI did a bunch of reorg on CPAN, and replied that he wasn't quite ready to announce it: "RSN" :-)

Oct 4th: ANDK created by-module/ and by-category/. Lots of discussion followed...

Oct 8th: ANDK created authors/id/ as the root of the author directories, so his directory was authors/id/ANDK/. This structure remained until May 2000.

Oct 12th: ANDK noted that the Module List Server had become a Perl Authors Upload Server, or CPAN Lounge, or Open CPAN. He didn't use the PAUSE acronym just yet ...

Oct 15th: ANDK -> JHI "isn't it time?". JHI: "I guess so". Such unbridled enthusiasm!

Oct 16th: ANDK "I'm tempted to call franz the PAUSE, the Perl Authors Upload SErver, indicating, that each author has deserved some break after having coded bravely for that many hours". The first use of the name PAUSE.

When was PAUSE announced to the public? It wasn't mentioned in the CPAN announcement.

The CPAN years

Oct 26th: JHI announced CPAN to c.l.p.a

May 1996: Gisle Aas (GAAS) released the first version of libwww-perl-5 (aka LWP). This was a Perl 5 rewrite of the libwww-perl library for Perl 4. LWP was written by MAK and GAAS, but subsequently maintained by Gisle for many years. For a long time this was the library for writing HTTP clients in Perl.

August 19-21 1997: O'Reilly ran the first Perl Conference. Of note here is that this was the first time that JHI and ANDK met face to face. It was the first time many key players in Perl met face to face.

Does anyone have any photos of ANDK and JHI together at the conference?

The Test module was released by Joshua Pritikin (JPRIT) in January 1998. This introduced the notion of structuring tests using a module. It wasn't very widely used, but was the start of what became a large movement in Perl / CPAN history.

In May 1998, Graham Barr (GBARR) and Chris Nandor (CNANDOR) had the idea for CPAN Testers: multi-platform testing of distributions uploaded to CPAN. The service launched in 1999, and the volunteer testers had to upload test reports by hand, sending them to a mailing list. Initially the "database" was just the archive of that list, but that didn't scale once it got into hundreds of thousands of messages.

19th May 1999: GBARR emailed P5P a link to a web site which had docs for all modules, crosslinked. This early version of search.cpan.org was running on a machine at his home. For many years, search.cpan.org was the only search engine for CPAN.

In July 1999, ANDK and Loic Dachary (LDACHARY) did some major work on PAUSE, including a rework of the database schema. From this point forwards, we know the date and time for when PAUSE accounts were created. All of the existing accounts had their creation date set to Thu Jun 10 16:48:40 1999 (epoch time: 929033320), so we know there were 694 PAUSE accounts by this point.

In August 1999, the UI for PAUSE was overhauled, and it switched to using MySQL (from Msql).

In August 1999, Andy Wardley (ABW) released the Template Toolkit. This was his second templating system, and was driven by the lessons he'd learned writing Text::MetaText.

For a good while, funet.fi was the master site for CPAN. Jarkko had tools to mirror from various Perl ftp sites around the world, putting their content into the appropriate place. PAUSE was just one of those sources, but eventually became the only real source.

BackPAN was started by ANDK on the 4th February 2000. This holds a copy of everything ever released to CPAN, including things that are no longer on CPAN. CPAN Testers has a longer description of BackPAN.

In May 2000, the author directories were changed from authors/id/ANDK/ to the current format format authors/id/A/AN/ANDK/.

Michael Schwern (MSCHWERN) posted the initial idea for CPANTS in August 2000. It was conceived as a 'CPAN Testing Service' that would measure the 'kwalitee' of a CPAN distribution, a measure of how well-behaved a dist is, in terms of working with the rest of the CPAN ecosystem. The original vision was broader than the current system. It would be a few years before anything got implemented...

In October 2000, the PAUSE database was converted to UTF-8.

December 2000: Mark Fowler started the first Perl Module Advent Calendar.

In 2001, MSCHWERN released Test::Simple, Test::Tutorial and Test::More. These were designed to provide a simple clear interface, to encourage more people to write tests.

Between 2000 and 2002, the testing culture in Perl, and CPAN, really started to take off. Many more core modules had tests, and CPAN modules started to switch from a test.pl testsuite (whose output wasn't parsed), to the t/*.t system.

Paul Johnson (PJCJ) had written code coverage tools for Hardware Description Languages and wanted something similar for Perl. However, it wasn't until Perl 5.6.1 that perl provided the infrastructure to make that feasible. On 9th April 2001, the day after 5.6.1 was released, Devel::Cover 0.01 was announced on the perl-qa mailing list.

20th May 2001: PAUSE's official hostname became pause.perl.org.

Following discussions at YAPC::EU in 2001, Jos Boumans (KANE) wrote CPANPLUS, an alternative installer to CPAN.pm. One feature of CPANPLUS was that it automated the sending of test reports to cpan-testers.

In Sept 2001, chromatic (chromatic) and SCHWERN released Test::Builder. This moved the guts of the testing model / framework into a separate library, opening the door for more specialised testing libraries that could be used together in the same testsuite.

While working on Devel::Cover, PJCJ started testing it with more and more CPAN distributions. Eventually this grew into cpancover, which aims to run coverage tests for as much of CPAN as possible. It started running in October 2002, and was announced on the perl-qa mailing list in December 2002.

Inspired by MSCHWERN's vision, Léon Brocard (LBROCARD) implemented the first version of part of CPANTS, which was released 24th March 2003. Module::CPANTS held the result data, and was created by Module::CPANTS::Generator.

In 2003, the Phalanx 100 project was started by Andy Lester (PETDANCE) and others (who?). The goal was to identify the top 100 CPAN dists, in terms of most widely used by other dists. Then improve test coverage and fix bugs, to improve the overall quality / reliability of CPAN. Results?

In August 2003 Ask Bjørn Hansen (ABH) announced a beta of cpanratings.perl.org, a site where perl users could review CPAN modules, giving comments and a 1-5 star rating. When you see stars ratings on MetaCPAN, they're coming from CPAN ratings.

In September 2003, Thomas Klausner (DOMM) took over CPANTS and rewrote everything, so the data was stored in a database. This resulted in Module::CPANTS::Analyse and the other modules, which are still part of CPANTS today (2014).

In February 2004, Simon Cozens (SIMON) released Maypole. Was this the first popular web application framework for Perl 5?

Sebastian Riedel (SRI) had been working on Maypole with others, but had different ideas from SIMON on how it should evolve, and left to start his own project. This resulted in Catalyst, which was first released in February 2005.

As noted earlier, originally funet.fi was the master site, but at some point after 2005, perl.org started mirroring modules directly from PAUSE, and the rest from funet. Sometime after that PAUSE itself moved to perl.org servers as well. Since 2001 the perl.org server has been maintaining all of CPAN.

In February 2005, Audrey Tang (AUDREYT) started the Pugs project; Pugs was an implementation of Perl6 in Haskell.

8th August 2005: first release of DBIx::Class

Stevan Little (STEVAN) spent some months working with AUDREYT on the Perl6 project. One of the things he worked on was the object system, described in Synopsis 12. Working in Perl 5, he wanted similar capabilities, so started working on them. Several failed prototypes later, he released Class::MOP in January 2006. Two months later he released Moose 0.01, which revolutionised the Perl OO world.

In August 2006, David Golden (DAGOLDEN) released CPAN::Reporter, which enabled CPAN.pm to submit test reports to CPAN Testers when people install things from CPAN.

In April 2008, the first QA Hackathon was held in Oslo, Norway, organised by Salve Nilsen (SJN) and Oslo.pm. The purpose of the meeting was given as:

Quality Assurance-related problems that are easier to solve when everyone is gathered in the same physical location. This can include issues with packaging, testing modules, community support or with tools.

This went on to become a yearly event.

After time away from doing Ruby, SRI returned to Perl in 2008, and started the Mojo project, as a new foundation for Catalyst. That didn't work out, and Mojolicious was created as a web toolkit, to demonstrate the capabilities of Mojo. Mojolicious took on a life of its own.

In 2008, people realised that Perl's testing model might be of interest outside the Perl community. It was at this point that it got the name TAP, and people started talking about it more widely.

Having got to the point where he was maintaining a lot of CPAN dists, Rik Signes (RJBS) became frustrated at the amount of time spent typing the same boilerplate over and over, and how many rote tasks were involved in releasing code. Existing release automation tools couldn't be customized to work the way he wanted, so over the course of a few hours on the bus, Dist::Zilla was born ("Dist::Zilla is a tool for writing less crap and more code when building CPAN distributions") and the first version released to CPAN in June 2008.

In the middle of 2009, Tatsuhiko Miyagawa (MIYAGAWA) was frustrated trying to build something on HTTP::Engine, and finding it too much work. In September, he brainstormed with Tokuhiro Matsuno (TOKUHIROM) and Kazuhiro Osawa (YAPPO), and came up with an idea based on WSGI and Rack. This resulted in PSGI (a gateway protocol between web servers and perl web applications or frameworks. Plack is a toolkit which implements the PSGI stack. All of Perl's modern web frameworks are now built on Plack.

In February 2010, MIYAGAWA was frustrated (again with the frustration!) trying to install Perl on memory-limited box - CPAN.pm kept running out of memory. He knocked up a script which would find a CPAN dist, download it, then run perl Makefile.PL && make install. After putting it on github, it rapidly evolved, and version 1 of cpanm was released on April 24th 2010.

In late 2010, Olaf Alders (OALDERS) wanted an API to CPAN, for an iOS app he was working on. Following a chat with Toronto Perl Mongers in the pub, he ended up creating the CPAN API, and Mark Jubenville (IONCACHE) then knocked up a web interface on top of that. A little while later, metacpan.org was born. This is the default CPAN search engine for many of us, these days. Even so, many search engines like Google and Bing still continued to direct people to search.cpan.org.

In 2008, Kenichi Ishigaki (ISHIGAKI) wrote a blog post titled "CPAN is not a sandbox for kindergarten children" (CPANは幼稚園児の砂場じゃないよね), which prompted a lot of discussion amongst Japanese developers. Inspired by this, Kentaro Kuribayashi (KENTARO) thought it would be helpful to have a place where you could discuss ideas for CPAN modules before you release them. He created PrePAN, which was announced in Japan on 4th October 2011, and then announced more widely on 24th October 2011. Yuki Shibazaki (SHIBAZAKI) subsequently joined the team, and helps KENTARO maintain PrePAN.

On 13th June 2012, brian d foy (BDFOY) gave a tutorial at YAPC::NA, part of which involved the attendees uploading their first ever CPAN distribution. There were 150 releases uploaded to PAUSE that day, which was the record number of uploads, until CPAN Day 2014.

In late 2012, ISHIGAKI resurrected CPANTS, while preparing a talk for YAPC Asia 2012. He ported it to Mojolicious::Lite, leading to the site you see today: cpants.cpanauthors.org.

On 4th June 2014, Philippe Bruhat (BOOK) had the idea for CPAN Day, and proposed (to NEILB) that it be based on the date of the first recorded upload to CPAN.

In 2017, the QA Hackathon was renamed to be the Perl Toolchain Summit, to more accurately reflect the scope and purpose of the annual gathering. It's now established as a four-day summit where we bring together the people who are working on the Perl and CPAN toolchain.

On 25th June 2018, all traffic to search.cpan.org was redirected to MetaCPAN, making MetaCPAN the only CPAN search engine.


This history was edited by Neil Bowers (NEILB) from a multitude of sources, and email discussions with many of the people listed above. There are likely errors in the above, since it was dredged from various memories, and then filtered through NEILB. If you spot an error, please raise a github issue, or submit a pull request.

Key sources:

Todo list

These are things I think should be added:

Document created on 2014-08-16 09:37:09 +0100 by Neil Bowers. Document last modified on 2018-06-26 16:43:38 +0100 by Neil Bowers.