Definition of "the Git repository"

Hi there,

raising this on the back of a discussion over at the Software Carpentry lesson about Git,

    https://github.com/swcarpentry/git-novice/issues/810

I used the book to justify my claim that it is the .git directory that is the repository, but I do have to concede that the way that the text in section 2.1 of the book reads, does suggest that one can refer to the working directory PLUS the .git directory as a "repository" as well as being able to refer to the .git directory alone as the "repository".

In the way I think of it

git init

initialises a Git repository, however, the only thing that changes as a result is that a .git directory has been created, ergo, the .git directory is the repository.

Furthermore, the fact that one can take the .git directory, move it to a new directory and start using it there (very much a nice feature) also suggests to me that it is the .git directory that is the repository, as distict from a working directory, under Git control because of the existence of a repository within it.

Interested to hear any thoughts around the semantics here, Kevin Buckley

On Fri, Jun 25 2021, Kevin Buckley wrote:

I think the right answer to this is that there is no right answer, if you read gitglossary(7) you'll find it mostly backs your argument, but you'll also find mentions in git's own documentation that say things like "a subdirectory of your repository" when referring to the repository's working tree.

More importantly it seems like this is documentation for novices, I think it's generally more important to get important notions like their historical data being stored in that .git thingy than it is to nail down every concept involved in that with 100% accuracy.

On 25/06/2021 03:44, Kevin Buckley wrote:

Thinking out loud, and without discussing various places where "Git repository" might be described one way or the other, a "repository" is a place where *something* is *stored*.

"Source code repository" would thus be a place where source code is stored, possibly with some metadata (current version, last change, etc.), but not necessarily the whole (versioned) history. For storage purpose alone, Git's own working tree could be then considered a repository in its own right (source code repository, if it contains source code, but it could contain other stuff as well, in addition or standalone). But as soon as you start working in it it's not really (only) a storage anymore (so not a repository), but a working area. It's more of a conceptual thing.

So if we strictly speak of "Git repository", I think it should be a place where Git keeps (stores) your (committed) work, alongside its own (meta)data - and that is the ".git" directory, indeed. Seems simple enough :)

One place where the confusion might be extended is the notion of "bare repository" for ".git" directory alone (without the working tree), which should then imply ".git" + working tree is in fact "a repository"... which it is, but bare with me - pun intended :)

As Git is mainly used to version artifacts being more or less actively worked on (changing, that is), one needs a working area in order to do the actual work, thus we have a working tree happily and conveniently provided by Git by default, as part of *working with* a Git repository.

As you said, having ".git" directory alone is enough to recreate the contents of the working tree, where if you would have the working tree alone, even if that could be considered to be "a repository" (for storage, not work), you would definitely not have "a Git repository" (no ".git" directory).

Also, when you work with a remote Git repository, it's only the committed stuff you can work with - what's inside ".git". You have no idea of contents of a working tree (and ideally not knowing if one even exists, though that's not always the case, like if you try to push to a checked out branch).

For some additional understanding, I guess we can compare "repository" with "archive", possibly being a more familiar concept - you can store source code somewhere, and that's your archive. You don't work on it in there, as then it would not be an archive anymore, but you keep it as a backup which you can always retrieve if needed.

If you use ZIP to compress your archive, it's then a "ZIP archive". 
And for the most of the time these two would in fact be interchangeable 
- you can have one or the other, being able to recreate one from the 
other (unlike with ".git" and working tree).

BUT, if you add some additional (meta)data to your ZIP archive - like password, description, etc. - then the two are not interchangeable anymore, "ZIP archive" not being the same as "archive", not being able to be recreated from it.

To conclude - Git's concept of a "working tree" alone could be "a repository" (used for storage, not working), but it's not "a Git repository" without the ".git" directory in it. On the other hand, while "Git repository" must have a ".git" directory, it can have a "working tree", too - but it doesn't need to (called "bare [Git] repository" in such a case), as it can be completely recreated from ".git" directory alone, being a mere convenience in order to be able to do the actual (development) work (and not required for "Git repository" to be a "repository").

Finally, as "Git repository" could be referred to as only a "repository" for brevity (which it is, in general), it's important to notice the latter might be ambiguous (as it does not imply "Git repository" in particular), thus using "Git repository" when being in the clear is paramount, indicating that you're interested in ".git" directory precisely (and possibly, but not necessarily the working tree as well). If interested in ".git" directory alone, "bare Git repository" is the most precise term.

Regards, Buga

On 25/06/2021 03:44, Kevin Buckley wrote:

Another way to look at it could be that you possibly have a repository already, and `git init` only makes it a Git repository... ;)

Joking aside, for a looong train of thoughts, see that other reply[1].

Regards, Buga

[1]: https://lore.kernel.org/git/9099ab09-f6d0-f1e2-1b8e-a55d22a903c5@gmail.com/

Kevin Buckley wrote:

That's not necessarily true. Another way to look at it is that the Git repository lives inside the .git directory, but not everything inside that directory is necessarily part of the repository.

I do have plenty of stuff that isn't part of the repository inside the .git directory.

The easiest way to disprove that hypothesis is finding something inside .git that isn't part of the repository, and that's easily done:

  .git/index

Hi Kevin,

On 25/06/2021 02:44, Kevin Buckley wrote:

In general, the Git semantics are confusing.

There is the generic, the conceptual and the implementation, all of the same term which has to be understood in context (which again uses terms with the same multi-way context..). This leapfrogging from concept to implementation and back again causes lots of learner confusion.

You have already seen that a source directory can become a repository by being initialised, and that the primary artefacts are in the .git sub-directory. One can also include in the generic 'repository' the various special .git* files that are [user] added to the main source directory.

But it gets worse. In the .git directory there is the 'objects' directory which actually holds all the _content_ of the 'repository', each object named by its hash value. This object store is a superset of those objects that form the versioned repository structure (the 'DAG'), and other parts of Git, such as the staging area (Index) contents, and other temporary copies of 'stuff'.

Meanwhile, to make the repository structure work, there are 'branch' pointers (see 'ref/heads' [0]) to the specific object hashes that provide the _starting point_ for each branch's linked list of commits and their content.

The use of the write-once unique object hash names [1] is part of that implementation 'trickery' that allows Git to work so well, but helps confuse those who think the object store is synonymous with the repository (e.g. all the new learners..).

In summary, everyone is right, as long as they are clear about the context. Which they rarely are.

Philip

[0] the 'ref name's are just strings that conveniently look just like posix paths ... [1] the current hash is the 160 bit sha1, so roughly 1 : 10^24  chance of a collision - unique enough;-)

On 2021/06/26 04:48, Philip Oakley wrote:

I don't believe it does get worse, indeed, I am not convinced that it is as bad as your observation on the semantics might suggest.

Everything within the .git directory "belongs", in my way of thinking, to the "repository", that is, the directory that gets created when git is (init)ialised.

For me, the 'objects", the 'ref/heads', the "staging area' and the like, also lie within the repository.

As for any anciliary files, that control how the git commands actually process any data in the "working directory" beyond the defaults, some of them, eg the global commands, will typically exist outside of the working directory: below one's home directory, and, of course, there are the system-wide files, typically below /etc.

Given then, that we can have Git-related files, for any given working directory, below /etc, below the user's home directory, and within the working directory itself, perhaps it is the whole "computer" which should be referred to as "the repository"?

Furthermore, even in the case of anciliary files typically found within the working directory, and I'm thinking of .gitignore as the obvious example, even the directives in that can be overriden on the command line, as well as complemented by files outside of the working directory.

I am tempted to say that even if the definition of a "repository" can't be agreed on, or rather, easily determined by inspection, then at least the corresponding working directory can be, however, the fact that one has access to EnvVars and corresponding command line arguments such as --git-dir=<path>, --work-tree=<path>, would suggest that it could be harder for the novice to get their head around things, however, as it is typicaly clear what the working directory is (the top level directory containing the files you are working on, under the control of Git), calling that same directory the "repository" seems, to me anyway, to add to any confusion, that a Git novice may have.

Given the can of worms that this question has opened up, although I'm getting the feeling that it has long since been opened, I still maintain that it would be less confusing, at least for a novice, for the working directory to be identified and for a previously non-existent directory, that gets created by the 'git init', to be referred to as the repository, so as to make a distinction.

Kevin

Kevin Buckley wrote:

Does it?

Suppose you have three directories, each with exactly the same contents in their corresponding .git directory, the only difference is the .git/index file:

 a) No .git/index file at all
 b) The .git/index file doesn't have anything staged
 c) The .git/index file contains some staged changes

Do you really consider them three different repositories?

In my mind the staging area is where you put stuff in preparation for the commit. The commit is part of the repository, the staging area isn't.

On 2021/06/28 11:24, Felipe Contreras wrote:

I think I do consider them as different, yes, but in the sense that, because the contents of any working directory can change in isolation to the others, they have become different instances (perhaps clones?) of the same repository.

Let's say I make two commits, that resulted in the same state of the files in the working directory, but I make them in different order in two of the working directories.

Clearly I need to sync the two different repositories in order to gain a consistency across them, and that suggets, to me, that they should be thought of as different.

Let's say I'd rsync-d one working directory, in which I'd made changes on its branch, with an unchanged one on a different branch (not that I'd recommend doing this without the utmost caution), then although the sync will not have affected the original branch in the unchanged directory, the git internals, stored in the .git repository, would have changed to reflect the new state.

As to the staging area,

again, for me, Git has an understanding of a "staging area" based on its inspection of the state of the working directory and a comparison of that state with what it knows has been committed.

Given that one can stash uncommited changes fron the "staging area", I think of both the "staging area" and various stashes as parts of the repository, in that you can't have a "staging area", unless there's a .git repository there to help instantiate the staging area".

Very much my perception of it all though: as others have pointed out, there does seem to be some wriggle room in how it is percieved.

Kevin

Kevin Buckley wrote:

Yes, but this has nothing to do with .git/index. You are talking about branches and commits. Clearly these are parts of a repository, and nobody objects to that.

Wouldn't it be much easier to say that git is comparing the working directory with HEAD?

Clearly HEAD is part of the repository, and so is the commit that it's pointing to (indirectly or directly).

But that doesn't say anything about the staging area, which is between the two (in my mind): the working directory and the repository.

To try to make it more orthogonal, let's suppose the index file was outside the .git directory. Would you consider then the staging area separate from the repository?

In fact, we don't have to suppose:

  GIT_INDEX_FILE=/tmp/index git checkout @~ -- .

Does that command change the repository in any way?

On 2021/06/28 13:21, Felipe Contreras wrote:

  
I have to admit that I don't know, and that I can't immediately
see where the "repository" would be, in that example. This is
obviously a gap in my understanding: happy to defer to yours.

I do however feel that the fact that we have moved to using examples that override the Git Index file on the command line, in order to define what a "repository" is, just so that we might be able to give a "more correct" definition of the term, to someone completely new to Git, suggests that, as others have already noted in the discussion, it's not easy to be "correct"?

Kevin

Kevin Buckley wrote:

Well, think about it.

No. The repository is what the repository is. The command is merely an aid for *you* to think about it.

If you are trying to define what a swan is, and you think a swan is a white bird, but then I show you a black swan, what do you think you should do with your definition?

What's complicated about this definition?

  A git repository is contained in the .git directory, but it's not the
  same thing. The staging area is usually inside the .git directory too,
  but it's not part of the repository.

Why do you want to make it more complicated than that?

On 28/06/2021 03:24, Kevin Buckley wrote:

Part of my view (of the multiplicity of descriptives within Git) regarding what is in a repository is from comparing the git created content from a clone, that of a `git init`, and a users overall project and .git trees after multiple interactions. Each has different features.

My main point was that there is this creative tension between the different contexts and that beginners should be aware that it's not all cut and dried in the same way that language definitions tend to be.

--

Philip

On Mon, Jun 28, 2021 at 8:24 AM Philip Oakley <philipoakley@iee.email> wrote:

Users should be aware that users are humans, and humans are not consistent. Nothing is as cut-and-dried as it might appear, ever. Or, as Kant put it: "Out of the crooked timber of humanity, no straight thing was ever made." https://en.wikipedia.org/wiki/Crooked_Timber

(If you want to put fine distinctions on things, some parts of a repository are repository-wide, and some parts are work-tree-specific. The git worktree command, when it adds a new work-tree, adds a new index and HEAD and *_HEAD and other work-tree-specific refs, for instance. That said, I still teach that .git mainly contains the repository proper, with the rest being your working tree -- or, once we introduce git worktree, your *main* working tree.)

Chris

Kevin Buckley <Kevin.Buckley@pawsey.org.au> writes:

Well, there is a distinction between "non-bare" and "bare" repository, so it is not exactly _wrong_ per-se to include the working tree portion of a non-bare repository as part of the repository created when you run "git clone", i.e. it is not a crime to say "please run 'git clone' to get a new repository to work in".

Often we do need to single out the things under .git/ in a non-bare repository when explaining certain features of Git, and need to clarify if it is not clear in the context (e.g. "the .git repository proper in the result of such a non-bare 'git clone'").

It is nothing new in inter-human communication.

If there is a use of word "repository" in our documentation where it is not clear if the repository proper or the combination of both working tree and the repository prper is meant, you may want to propose a clarification patch.

Thanks.