Delivery

Git at twenty, considered as organizational memory

The first commit landed in April 2005. Two decades on, the repository has quietly become the most honest historian most organizations employ.

Code, Noted1 min readDelivery

The first commit to git landed on April 7, 2005, describing itself as "the information manager from hell". Twenty years later it manages, by any reasonable count, most of the world's source code. The anniversary retrospectives this spring have concentrated on the tool. The more interesting subject is the record.

A version control history is an institutional memory with properties institutions rarely achieve on purpose. It is complete: every change, however small, however embarrassing. It is honest: the commit that introduced the outage is still there, attributed and timestamped, long after the postmortem's action items were quietly abandoned. And it is durable in a way documents are not, because it travels with the thing it describes. Wikis rot beside the systems they document; the history is the system, at every point in its life.

Organizations underuse this record to a degree that would puzzle an archivist. The questions that stall modernization programs, what does this module actually do, why does this check exist, who understood this last, have answers sitting in the log, retrievable in minutes by anyone who has learned to read it. Reading history is a skill; git log is a primary source, and like all primary sources it rewards patience and punishes skimming. The engineer who reads a subsystem's first hundred commits knows things about its intentions that no current document records, because the documents describe what the system became, and the history records what it was for.

I have sat in three separate meetings where a two-hour argument about "why is it built this way" ended when someone finally read the 2011 commit message that explained it. Twice, the explanation still applied.

The practical conclusion is modest and worth institutionalizing: commit messages are letters to the archaeologists, and the archaeologists are your own staff, five years hence. The information manager from hell turned out to be the profession's most reliable historian. It would be mannerly to give it something worth filing.