Documentation: Git as archive

February 10th, 2022 Tagged with: technical, qualitative

This post is the first in a series on documentation. They’re my attempts to navigate one set of junctures between tech tools and thinking in the humanities.


The official and commonly-repeated summary perspective on Git is that it is a “distributed version control system.” That’s absolutely true, but it’s also not meaningful unless you have prior experience with version control, and especially unhelpful to humanists and non-tech people.

So this post is an attempt to understand Git another way, as a tool for creating, managing, and sharing historical archives simultaneously along the events of actual history. Though normally a tool for managing software projects, I feel that Git’s features enable us to maintain historical archives in ways traditional tools do not.

What is Git?#🠑

First, a tiny bit on background information on Git, to give the rest of this post some context.

A project in Git is a single folder, called a repository, which can have as many subfolders inside of it as wanted. A project develops or grows with more content added or deleted in files, and when a sufficient amount of changes to a Git project have been made to form a cohesive logical unit, the project manager should commit these changes with a message. This stores the changes you’ve made into timestamped, annotated, discrete steps.1

A project can contain many different concurrent versions through Git’s branches, which can be split and merged with other branches in many different ways. A branch is born by a splitting another branch, and begins with the exact same history as that origin branch. Committing to the new branch makes that branch diverge from its origin and embark on a new history. The origin branch and the new branch can be switched back to with ease, allowing you to work a bit on a new version but switch back to the main version with no loss to the changes you’ve made.

Projects managed through Git can be shared with others through a platform like Github or Gitlab. Other people can download the copy of the project and its history by cloning it. Because of this pattern of having all collaborators keep a comprehensive copy of a project’s history, Git is known as a distributed system.2. Once collaborators make changes, they can commit directly to the repository, or submit pull or merge requests on Github or Gitlab that others can review and then merge.

The last thing I’ll say in this section is that Git is incredibly popular. Some 90% of the world’s software engineers say that it’s their version control system of choice3, and it’s pretty much a requirement to know your way around it if you are a programmer.

That’s about all of the Git knowledge this post requires, but more can definitely be read through thousands of other sources online — the Gitlab guide to Git and the official Git book are great options.

Curated histories#🠑

The essential feature that makes Git a useful tool for archival creation is its concept of a commit.

There are lots of software out there that people are more familiar with, that keep track of previous versions of files and allow you to roll back to them. Google Drive, Dropbox, Box, — really any enterprise tool for file storage can do this. The difference between these tools and Git is threefold:

  1. Git forces you to pick a finite number of points that can be returned to, rather than simply keeping track of every possible previous point in time
  2. Git forces you to annotate these points with specific messages
  3. Git forces you to take snapshots of the entire project’s state, rather than any specific file

To someone who is just looking for backups, these are limitations, because it requires you to do extra work and creates a discrete number of states instead of noting every single file change comprehensively. But to a historian or an archivist, these are features. Git forces you to give extra meaning to your changes, by selectively marking events to record, by annotating these changes, by recording information not about a change to a single file but synthesizing information about many changes about the project as a whole.

Meaning is theoretically lost, because you may make many changes to a file that are lost if you erase them without attaching them to a commit. But meaning is gained through all of the above strategies. That’s the essential task of a historian: to take the infinite amount of events that have happened in the past and record them in a way that is useful and meaningful.

Multiplicitous histories#🠑

The other defining nature of Git, especially in contrast to other version-control systems, is its inherently distributed nature. In a few different ways elaborated below, this distributed system encourages a perspective on history that allows many histories to exist at once — what I see as mulitplicitous histories.

The first and most general aspect comes from its concept of branches. A Git project with branches means that it has several concurrent histories of the same project. Traditional archives don’t provide a method or format to keep track of these parallel strands of events as different versions of a historical process. That role is often left to the realm of archival perusal, not archival creation.4

The second aspect of its distributed nature comes from the use of Git across different machines, none of which Git privileges over another. When building a software project, many people could have different ideas for a new feature, and develop them independently across their own laptops. There is no single source of truth that Git might privilege by letting only one branch overwrite another branch’s idea.

There is still a hierarchy of knowledge in Git projects, as all events are incorporated into the structure of the Git index. Other events that are part of the creative process are often not logged to this index — meeting notes, journal entries, good conversations with friends might all help push forth a project, but none of them are incorporated into Git’s history. Even with this hierarchy, Git encourages copying, dispersal of the information it holds with no sense of one “authoritative” copy against other modifications. On the other hand, while there are many archives that could be created of a single historical event or topic, traditional archives encourage no such distribution of a piece of knowledge. There are many map collections in the world, but the Yale Map Collection (or any other single archive) does not allow researchers to clone the archive into their own version that has all the legitimacy of the original in the way that Git encourage.

On documentation, and doing history#🠑

So far I’ve talked a bit abstractly and interchangeably used the words “repository,” “history,” and “archive.” This can be a bit confusing, given history’s events and history’s explanation or recording might be thought of as two separate things. Is Git a tool for historians, or for the actors of history? Is Git for the creation of an archive, or for the creation of history?

The pragmatic answer Git gives (and any historian would scream out) is that these are sort of ill-formed questions, since the two sides of actor and historian aren’t separate things. The actor of history necessarily records and gives meaning to history’s events, just the same as the historian does. In a broader sense, both are obviously the curators of knowledge, and as much as there might be some logistical or circumstantial distinction between the two groups, the archive created has been

This is to me the largest difference between Git and traditional archives; Git recognizes that these two parties are in fact one and the same, whereas traditional archives by structure do not. From the outset, Git forces the actor or subject of history to index and give meaning to events, or in other words peform the duty of a historian. Indexing and giving meaning to archives happens no matter the format; Git is unique in encoding that into its structure and format of the archive.

The sort of elephant in the room with using Git as an archive is that it it’s very limited in scope, recording only the events of discrete file changes. It’s hard to record even the meeting you had yesterday or the conference you attended and the insights you gained; you can only record the files you create after-the-fact. Forcing the roles of actor of history and archivist into one is also not just not useful in most cases: most historian’s day jobs are studying the past, not recording and making meaning of the present,5 and even when they are discussing the present, they are often reflecting on other’s actions from a distance, not consciously serving as an actor of history.

So I’m not advocating for Git as an archive in the sense that we should use it for most the things we use archives for. It just offers no benefit to store pictures from some event last year using Git compared to any other file format storage. Git is only useful as an archive when the process of archival creation is itself an interesting subject of history.

An example: my thesis#🠑

For me, a useful area where I’ve been using Git in this way has been in writing my thesis. I’m writing about the IMF and the Asian Financial Crisis, and as often the case for the dressed-up projects like a senior thesis, I’ve had to present it in a few places, alter the format several times, and develop a coherent body of knowledge from scratch.

Git has been useful mostly in a practical sense, as a project management system. I can workshop an idea a few different times across different branches, ultimately picking the version that speaks to me most. I also use branches for project presentations, making a slideshow once and modifying it a few times as needed for different events.

But I’d like to look back and reflect on my growth, since this is sort of my first big self-guided intellectual project. What was I thinking last month? How did things change from when I hurriedly submitted a project proposal at the beginning of the year? When did I write most, and when did I take a three-month break? These questions refer to a history too; maybe not as consequential as the Asian Financial Crisis, but meaningful nonetheless.

  1. Under the hood, Git saves only the changes you've made between each commit, not the contents of the entire project (which can often be thousands of times greater). With this approach, Git can be an extremely compact way of storing all states of a project.
  2. In other words, by its original design, Git has no single source of truth. One does emerge by the common pattern of having a single place to host the files like Github or Gitlab
  3. Though others exist, like Subversion, Bitkeeper (which Git was designed to replace), and Concurrent Versions System, Git is by far the most popular
  4. This isn't to say that the creation of different strands of history by the archivist is necessarily a good thing. What constitutes a strand of history anyhow, that they can be so cleanly separate into different things by Git? This question is fairly easy for software projects, but unanswerable for history at large.
  5. While acknowledging that the past and present are also not separable, in reality most historians seem to be discussing the past in their day jobs.