Comments on: Two minor thoughts

By: Neil

Neil — Sun, 22 Feb 2009 17:02:05 +0000

Well, I think this sort of makes Abram’s point: if you have to read the manual each time, maybe the tool isn’t as intuitive as it should be.

By: Jakub Narębski

Jakub Narębski — Sun, 22 Feb 2009 02:52:29 +0000

@Abram:

“I checked out an older version of a repo”… and didn’t pay atention to the message from git, hmm…?

“Eventually I found git-lost-found”… no need for such a low level tool. Ordinary “git checkout -b new branch name” should be enough, and if you lost a comit, there is always reflog: “git reflog HEAD”.

In short: read the manual first, please…

By: Neil

Neil — Mon, 16 Feb 2009 20:58:14 +0000

All true. I have a feeling that if your workflow is similar to Linux, it will work for you. If not, or you haven’t got a good sense for the workflow, you’ll be in trouble. Really, for working on my small projects, SVN will do as well.

By: Abram

Abram — Mon, 16 Feb 2009 20:37:39 +0000

My problem with GIT is that the model is dirty and unclear.

GIT is real deal software and it relies on an underlying model, but when I think I’ve learned the model I find I haven’t. Or the maintainers have not given me an interface to do so.

Here’s one example. I checked out an older version of a repo and I commited. Where does that commit go? Turns out it was on an non-existent branch, so I couldn’t check it out without the exact commit ID, I couldn’t do anything with it. It was effectively lost. So then I tried to apply my knowledge of GIT, first I searched for commands which would let me query the children of a commit. Nope. Then I searched for commands which let me search for nodes with a certain parent. Nope.

Eventually I found git-lost-found and recovered it.

I didn’t have this problem in DARCS http://darcs.net/manual/node9.html
Which was built ground up on a formal model of patching. Surprisingly the creator is not some formal models/methods buff, but a physicist. I don’t use darcs much anymore because it is rather slow.