Summary

  • The Begin3 catalogue template asked for a package’s final location, not the incoming directory where its author had just dropped it.
  • Archive keepers used names and versions to place and replace software. Those useful updates also changed what the surviving catalogue could tell historians.

The instructions allowed a guess. They just ruled out one particularly tempting answer. When filling in a Linux Software Map record, a programmer had to give the software’s eventual home. The incoming directory was where it had arrived; it was going to move. Putting that address on the card was rather like telling dinner guests to meet you at the loading bay.

This small piece of administrative comedy survives in the Begin3 template reproduced in MetaLab’s October 1999 research report. The form accompanied contributions to a Linux archive that had begun at SunSITE in 1992. It supplied a title, version, description, locations and people to contact. Writing a program earned you another writing assignment.

There was a practical reason. On slow connections, downloading an intriguing package simply to discover what it did was an expensive form of browsing. Jeff Kopmanis devised the Linux Software Map, drawing on a 1992 proposal for describing anonymous FTP archives. A small record could travel ahead of a large file and tell a prospective user whether the journey was worthwhile.

The format itself needed caretakers. Lars Wirzenius took over from Kopmanis and introduced Begin3 on 4 August 1994. Among its improvements was one entry for all the files belonging to a package, along with less awkward multiline descriptions. Aaron Schrab took over LSM keeping in October 1996. Even the machinery for describing software had a succession problem to solve.

At MetaLab, contributors deposited the program and its LSM together. Archive staff inspected the incoming area daily with keeper, a tool written by Eric Raymond and modified by Miles Efron. The archivist reviewed the description, put the files in their proper place and used the package name and version to replace older material. The record then went onward to the maintained LSM list, which other sites mirrored. An upload had arrived; a librarian still had work to do.

The template made that work unusually visible. Authors were asked to keep a package’s title stable between versions so updates could be matched. The original author and the maintainer of a Linux port had separate fields. The main location specified a host, directory and files. These boxes connected different kinds of knowledge: what the thing did, which release it was, where it belonged and whose inbox should receive a question.

The surviving ibiblio search form shows the reader’s side of this bargain. It offers fields, terms and weights, as well as a choice of headline field. A contributor’s little description could become a reader’s search control. The form is an interface preserved on the site; the 1999 report supplies the historical account of the archive’s operation.

Scale made tidy labels useful. In a snapshot taken on 21 September 1999, the researchers counted 4,455 LSM records covering 20,633 files after excluding the distributions and documentation directories. One package could contain several files. The record was the handle by which that bundle could be found and managed.

People remained capable of defeating a perfectly respectable box. The researchers found unusable dates, including “Thursday”. More subtly, even a correctly formatted date recorded when the entry had last changed. New versions replaced old LSMs, so a recent date might belong to a thoroughly old project that had just been refreshed.

That gave the catalogue a particular memory. It was good at pointing toward the software now on the shelf. To reconstruct every arrival, a historian needed something else. The same diligent update that made tomorrow’s download easier could rub yesterday’s date off the card.