Exactly. The git data model is intuitive and easy to grasp after a short amount of time. The porcelain is poorly designed, but part of that is due to the flexibility of primitives underneath, and the desire to support arbitrary workflows.

Contrast with svn which attempts a very clean porcelain interface with a completely muddled data model underneath. The conflation of repositories, directories and branches in svn makes it impossible for it ever achieve 20% of git's functionality simply because things are so poorly defined.

After using git for 6 months I understood it better than I did about svn in the previous 5 years. I would prefer a better porcelain, but given that software development is my full time job and that I can use git for all software development regardless of the language, I'm happy to commit a bit of muscle memory to git's idiosyncrasies.

> the desire to support arbitrary workflows.

Now I am laughing and laughing bitterly. git supports one workflow -- the massively decentralized one. To this day you can't have a simple workflow with git, the one that cvs/svn supported and practically all small project would benefit from, the one that bzr calls a bound branch.

git , I believe , is the textbook case of what the opposite of a user friendly UI is. commands have switches which change the command so fundamentally it should be another command. Which noone wants 'cos there are like 140 commands already. switches which across commands do the same thing but are named differently. The same command doing wildly disparate things without any indication of what's happening -- try git checkout file, and guess what the state of file will be. It might from the staging area or it might come HEAD if it wasn't staged. Nuts.

> "To this day you can't have a simple workflow with git, the one that cvs/svn supported and practically all small project would benefit from"

What? Of course you can. I've worked on teams that did it. You are talking total rubbish.

I'm a happy git user, but tell me how this is supposed to work: there's me and one other guy working on a project. There are three logical branches: trunk, his branch, and my branch. In SVN there would be five source trees: those three on the server, his working copy and my working copy. But in git we end up with fifteen: the three on the server, my branch on my machine, his branch on my machine, master on my machine, the remote-tracking copies of those three on my machine, and the same again on his machine. All of which could be different.

How do I reduce that complexity? I always pull and never fetch, which helps slightly, but only slightly; pull seems to fetch other branches, so it's still possible to have my copy of a branch end up behind my remote-tracking copy of that branch. There's no command analogous to pull for "add and commit and push", so that's always a second step to possibly forget (I don't want to rely on an alias as I work on a number of different machines). Most problematic at the moment is that there's no way to tell the difference between an up-to-date branch and a non-remote-tracking branch, so I sometimes delete branches that I haven't fully pushed, because I forgot to make them remote-tracking, so they didn't show up as behind when I "git status"ed.

Set up a repo on a server somewhere and declare it to be the "central repo". You and the other developer pull/push from that repo only. In other words, ignore some of git's capabilities and treating git like SVN. Just because you can pull from your teammates, does not mean that you have to.

I've been on a team that transitioned from SVN to Perforce, the again to Git, keeping the same workflow all the way through.

It isn't the way that I prefer using git, but it works perfectly well.

> Set up a repo on a server somewhere and declare it to be the "central repo"

That's the situation I described. We still have the problems I mentioned: sometimes we commit without pushing (particularly because we sometimes didn't set a branch to be remote-tracking and didn't realize this), and sometimes our branches are behind our copy-of-the-remote-branch because we pulled different branches (which leads to bogus merges in the history).

Committing without pushing is not a problem (unless you intended to push but forgot to?) It is not a concept that necessarily exists in traditional centralized version control systems, but the fundamental problem still exists. In SVN "forgetting to push" is just called "forgetting to commit".

If your team member forgot to push and you put out new changes, that is a problem for him to resolve. If you forgot to push, and your team member put out new changes, that is a problem for you to resolve. Workflow wise, this all works the same as it does with any other centralized workflow.

If you forget to check for updates... well that is something that happens in other centralized schemes as well. You figure it out when you go to push and it fails, you correct it, then you are good to go.

> Committing without pushing is not a problem (unless you intended to push but forgot to?) It is not a concept that necessarily exists in traditional centralized version control systems, but the fundamental problem still exists. In SVN "forgetting to push" is just called "forgetting to commit".

Sure, but you hit the problem twice as often in git, because you have to do twice as many things.

> If you forget to check for updates... well that is something that happens in other centralized schemes as well. You figure it out when you go to push and it fails, you correct it, then you are good to go.

In SVN that doesn't show up as a merge in the history.

> "Sure, but you hit the problem twice as often in git, because you have to do twice as many things."

Well no, I don't...

If this really is a frequent problem for you, then you might want to consider adding a note to the end of git-commit's output to remind you to push, or even just aliasing git-commit to push by default. I would recommend that you instead learn how to use git, but failing that...

> "In SVN that doesn't show up as a merge in the history."

If you don't want to resolve those situations with a merge, then don't resolve those situations with a merge... Rebasing exists for a reason.

In SVN, that shows up as a conflict which has to be merged by hand. 'git rebase' and/or 'git mergetool' are nicer.

> In SVN that doesn't show up as a merge in the history.

You might like "git pull --rebase"

Don't pull origin/master directly into local feature/test branch. Pull into local master then rebase local branch onto it.

This goes against all the advice I've seen elsewhere; we push our feature branches all the time, and sometimes pull from each other's if we're working in overlapping areas, so we need to not rebase them.

If that's what you want to do, that is fine, but you are no longer using a centralized process if you do it that way.

Git allows you to follow a centralized process perfectly fine. However if you choose to not follow a centralized process, it will not force you to.

Is there a reason you aren't using disposable feature branches?

I don't track my other developers' feature branches locally unless I need to view them.

You're right, thinking about it - most of the time I wouldn't have a copy of my coworker's branch (though we do sometimes pull changes from each other's feature branches - indeed that's supposed to be the big advantage of git, no?)

> It turns out however that Git is really easy to grasp.

Please write something to explain it. I must be pretty dumb, because I don't get it, even after reading about ten explanatory texts.

Since you have already read ten things about it and don't understand it, it would help if you told us what you do not understand so that we do not waste our time rehashing things that you have already not understood.

That's a very good question. I hadn't really tried to pinpoint what exactly I don't understand about it until now, but here's what some thinking uncovered:

Branching/merging/committing is pretty straightforward. The problem is that some commands seem to be very convoluted. For example, why does reset do four different things depending on whether it's soft or hard or plain? I keep having to Google to find how I can revert my latest commit.

Another thing I have trouble with is obscure failures. Obviously this isn't something I can just learn, but there are times when git fails for a reason I don't understand...

Ah yeah. The "why" with a lot of the porcelain stuff probably tends to be an unsatisfying "Because somebody didn't think it through very much a few years ago." git-reset would probably be better if it defaulted to --soft, and possibly left the other options to other commands.

Here http://www.sbf5.com/~cduan/technical/git/ this is the only way to understand git.

That's a great explanation, thank you!

Same here, whenever I attempt to give it another chance, sooner or later I break down and revert to hg-git. I recently discovered EasyGit [1] and it seems promising from the docs. There's also gitless [2] that was covered in HN a few days ago.

[1] https://people.gnome.org/~newren/eg/

[2] http://people.csail.mit.edu/sperezde/gitless/

And we all know what runs through porcelain before it runs through plumbing.