GNU arch

» see gnu.org/software/gnu-arch as it lived, on the wayback machine

~ the obit ~

The version control a computer engineering student ran into in the mid 2000s was CVS, and then Subversion once a department got round to upgrading. Arch was not on that list anywhere. It belonged to a small group of people on mailing lists who had already worked out that the central server was the problem, several years before the rest of us noticed there was a problem to have.

Thomas Lord started it in 2001 as a collection of shell scripts meant to be a better CVS. The command was tla, for Tom Lord's Arch, and the design decisions inside read now like a list of things everyone takes for granted. Changes were tracked as changesets rather than file revisions, and commits were atomic. Renaming a file kept its history because files carried identities of their own, and branching happened across archives, which is to say across machines, because there was no single place the history had to live.

It became an official GNU project in 2003. That is the high point, and it is worth saying plainly that arch got there first. Distributed version control existed and worked and was under the GPL while the rest of the industry was still arguing about whether locking a file was a feature.

It was also famously unpleasant to use. Archive names were long and formal, the commands took real effort to learn, and the documentation assumed you had already accepted the model before you arrived. The people who stuck with it were evangelical. Everybody else bounced off, and the bouncing off is most of this story.

~ the rap sheet ~

~ the forks ~

Arch forked twice and both forks were bad-tempered. Walter Landry took the code in a different direction as ArX after a serious disagreement about where the project should be heading. Canonical produced Baz, and Lord was sharply critical of how that one had been announced.

In august 2005 Lord resigned as maintainer and suggested that Baz should become the main arch project. Canonical then abandoned Baz in favor of Bazaar, which was not a fork of arch at all, and scrapped the Baz 1.5 release in 2006. Andy Tai picked up maintenance of tla in october 2005 and kept it going for a while longer.

~ the postmortem ~

April 2005 is the month that settled it, and arch had nothing to do with why. Linux lost its license to use BitKeeper, and within weeks there were two new distributed version control systems written by people who needed one immediately. Git and Mercurial both arrived that spring. Both were faster than tla and much easier to explain to somebody who had not already agreed with you.

The last arch release was 1.3.5 on 20 july 2006. In march 2008 Tai said it was no longer under active development. By 2009 the official GNU status was deprecation with security fixes only, and the project page now states it has been decommissioned and superseded by GNU Bazaar.

~ the inheritance ~

Everything arch argued for won. History belongs on your own machine and branching is cheap. A commit is a changeset with an identity rather than a mark against a file. Nobody uses tla to get any of it.

GNU named Bazaar as the replacement and Bazaar's own activity has since wound down as well, while the world went to git without needing the argument made to it. Lord's position is that the case had already been proven and then lost on ergonomics rather than on principle, which is a specific and quite bitter way to be right.

@}-->--

              )
             (_)
            .-'-.
            |   |
            |   |
            |   |
            |   |
          __|   |__   .-.
       .-'  |   |  `-:   :
      :     `---'     :-'
       `-._       _.-'
           '""""""

« previous  ·  random  ·  next »