Bye Bye SVN, Hello Git
engineering.secondmarket.com
engineering.secondmarket.com
Just because you are an elitist hipster hacker, doesn't mean that this issue will only pertain to staid, conservative companies.
Install TortoiseSVN. Done.
Install TortoiseHg. Oh it needs something else. What's it called? Ok. I need msysGit. I'll get that. Ok, they're all in beta? Whatever I'll just get one. What? It runs on top of Cywin? And on, and on...
Am I missing something here or is it really that convoluted? I work with non-programmers (engineering types) who program and I need it to be easy for them. Any recommendations?
(Just pointing out because TortoiseHg actually is fairly painless to install if you want to use Mercurial on Windows, and the typo might confuse some)
So git is worth switching just for branching/merging superiority? Maybe I don't do it enough, but svn has never let me down. Or maybe I'm not on a large enough dev team?
Guess I'm really just trying to figure out if git is so amazingly better that I really am under a rock (I'm not seeing that), or if it's more of the "if you're not using {latest hip tech} you're not to be taken seriously" kind of thing that we see so much on HN.
(sorry about new account -- can't find original credentials)
I wish I had time to get in all the details about why git is better but I will leave you with two questions to ask your existing version control system:
1. How fast is your version control system?
Unless you have tried git, you will not realize how painfully slow SVN (or any other VCS which has to talk to server) is. 90% of my git operations take less than a second.
Now, you may say that 'speed' is not an issue. Trust me, it is. Once you have a super fast VCS, you will embrace it as your friend in everyday coding instead of using it at the end of the day. It's something similar to what Google thinks that making website/webapp faster brings in more users.
2. Can your version control handle renames?
There are very few version control systems out there which can handle renames. If you dread renaming files because of SVN, you ought to look at git. BTW, not only can git handle renames without a hitch, it can show you code history across file renames.
Again, you may think that 'renaming' is not an issue. But if you believe that 'naming' is very important for your code and you do re-factoring all the time, you don't want your version control system to dictate how you work.
I can imagine why one may think that git is fad. But after using it for 2 years now, I can not use any other version control system. [And for a background, I have used following version control systems for real proejcts: CVS, VSS, Perforce, Surround SCM, SVN, TFS.]
Btw, SVN supports partial checkouts (git doesn't). Which I use regularly on a gigabyte repo.
I use git regularly for all my hobby projects (mainly because 'git init' seems way simpler than the equivalent svn command), and I might even start a company off on git, but I'll never claim that SVN is obviously a bad way to do things.
1) Fast enough? I have tried git. It's pretty cool, but the difference between 5 seconds and 1 second is minimal at best to me. I don't think I've found myself dying for a faster SCM, though I usually make maybe 1 - 5 commits per day. Git on Win32 is meh compared to git on Linux, so that doesn't really convince me any further either. Git is also a PITA to compile on !Linux using !gcc, because heaven forbid we write using ANSI/ISO C.
2) Yes? I can do renames, but strictly speaking it is not an atomic "rename" operation. I don't particularly care about rename history, and I'm not really sure why others do, but I guess if I tried I could make up a use-case for it. Still, it wouldn't be a selling feature for me, that's for sure.
It seems strange to me that you take the time to tell us that "git is better", like we're comparing O(n) to O(log(n)) algorithms. Different workflows, projects, and organization structures have different needs. It seems hopelessly naive to simply assume everyone's needs are best fit by git, and git alone.
How about this: I want to tell, given two revision numbers in the mainline branch, which came first? In SVN, it's pretty simple. In git? Well, first open a terminal...then ...argh. As release manager, it's kind of annoying. Simple use case: is 2aa8aca05357004d4418807c06f53d81517bc629 before or after e8d0555ce401b6acb0f225aa0263e9b72136347d? Did e8d0555ce401b6acb0f225aa0263e9b72136347d make it into the build given that 5b5c73ceaa442bafff08b4470aad3ff276be94e3 was the final commit before the cut-off?
I might waste 5 seconds opening a terminal window or tab, 30 seconds finding, copying, and pasting the revision numbers, then a second or two determining the answer. This, to me, is not a productive use of my time when revision numbers simply tell me what I need to know. All of that magical time git just saved me might have just been absorbed or gone into time debt.
Have you ever thought about why do you only have 1-5 commits a day? I do 70 to 100 commits a day. I commit each and every semantic change. I do not commit changes so that I have a backup of my code, I create a commit so that I can track my 'semantic code change'.
Let me give you an example:
Let's say I have a working website. The website has header/footer/content style. I spend one hour to customize my website layout by modifying html/css.
When I am done with header, I commit my HTML/css.
When I am done with footer, I commit my HTML/css.
When I am done with content, I commit my HTML/css.
During my work, I found a bug in CSS which fixes float in IE and I created another commit for this particular bug fix.
The most important point being that each of my commit is 'independent' and represent a meaningful unit of work.
You may ask, why create 4 commits instead of one? The reason is that git will let me use these commits anyway I chose.
It's 2.30pm and my manager stops at my desk and tells me that I have to fix the float but in IE 'RIGHT NOW' before the 3pm meeting. I am only needed to fix the float bug and nothing else because the website has to match with printed material for the meeting. How do you do it? Easy. git will let me 'cherry-pick' my last float bug commit and merge it in to current production branch. After merging the one particular commit, I test everything to make sure nothing is broken and I am all set before the 3pm meeting.
Obviously, this was a simple case. There will be cases where you can not just pick one commit as it depends on other commits. You will have to do more work to get the fix in. Git is the helping hand which can let you play with the commits to get the answer you want. SVN can not do that.
Going back to my point, I think you may not realize that you do far fewer commits than you should because your subconscious brain knows that commits are slow. Once your brain will realize that commit can take less than a second, you will automatically commit more often.
To give a unrelated example: Back in the day when we didn't have access to high speed Internet, we would consult local documentation only because searching online was slower than finding it locally. Now, with Google and high speed Internet, I have stopped downloading documentation locally and rely on Google to find it for me.
'caring about rename history'.
Again, this is dependent on the way you code. I like to re-factor my code a lot. I rename my classes whenever I find a better name. I move my code around in different folders as I find better way to organize my code. And as I re-factor a LOT, I DEFINITELY care about my code history. Git can track renames/folder movement and still show you the file history. You may not care about rename history now but you may care about it in future. With SVN, you have no choice. Git offers you that choice.
Now to your workflow comment. The beauty of git is that it is a 'content tracking' tool. You can use it to FIT any work flow you like. git does not force you to use any particular workflow. You can use it in the exact same way you currently use SVN. Or you can come up with your own workflow which no other team in the world uses. It's all fine.
Finally, the revision number comment. Generally, I haven't had a need to figure out which revision came first based on 'revision numbers' alone. But once again, git has an answer. You can use git describe command. It will give you something like:
"git describe gives you a version description like the following: v2.0-64-g835c907. The v2.0 part is the name of the latest annotated tag preceding the commit, 64 is the number of commits after that, and 835c907 is the abbreviated commit id. It is basically there to identify any revision in an exact and convenient (although technical) way."
From: http://stackoverflow.com/a/1817099/92414
And that's the beauty of git. Git already has a way to address your common concerns and if it doesn't, you have enough meta data to make it work the way you want to.
Once again, I really understand where you are coming from because I was on that side once. But I sincerely believe that git is a much better tool and it will change your life as a developer. It has definitely changed mine.
Please, please, do yourself a favor and at least experiment with git before dismissing it.
But, at the end of the day, I rename files so infrequently that it's hardly a selling point or problem on either system.
Of course you still need some centralization in order to share work with other people. But the more compelling thing about Github is forking, not centralization. With Git you can easily synchronize commits between multiple remote repositories.
I've never had to use svn myself, but it's this aspect that's always sounded positively crippling to my workflow
There's also the benefit of being able to checkout just one folder of the project wherever you want without having to keep the same folder structure as the source project. I understand git is able to do a "sparse" checkout now, but the parent source folder structure is checked as well.
At the end of the day, use what works best for you. If you want to test the waters, use git-svn. You can then work locally and then dcommit your changes when they're ready to go upstream. It'll give you a pretty good approximation of what git can do for you workflow-wise.
Perforce seems to be entrenched in more corporate environments.
Did you experience some of the issues Facebook has?
See:
http://thread.gmane.org/gmane.comp.version-control.git/18977...
Setup 1 exploites this. This would be movie companies were every version of every rendering is in Perforce. On the server you have easily many times even the size of the desktop hd, and users only grab what they need/want. These companies should stick with Perforce.
Setup 2 came about from laziness. Everything was dumped into the repo with no organization. Converting it to git will require a XXXGB repo which while manageable windows can't handle well. You encounter say dozens of copies of the binary sprinkled around inside of the source directory (not even revisioned, just copied with the version included in the file name). Need a copy of every Windows NT CD's stored somewhere? Why not in the src directory!?! Once you cleanup/split this the src repo goes down to a usable size.
After the above basic cleanup (which honestly can/should have been done in svn/perforce anyway.) the src repo can still be a large size because the src isn't the src for one project, but maybe dozens or hundreds of projects that compile to one binary (or some similar setup such as lots of little binaries that are one product). It is then up to you to decide on how you want to proceed. There are a few different approaches with different pros and cons and situation specific. (And discussing them is really a full blog entry not a random hacker news comment).
I'll leave you with my law about repo size: When every developer in a company is committing to the same branch the odds that a commit will break the build increases as more developers are hired.
Edit: simplified
Now I'm freelance I've had the time to put into learning Git and I'm glad I did, but the point is there are still a lot of people using SVN, some of whom will switch to Git in the future, and posts like this can be very useful to them.
How true this is. I spoke with one sysadmin who told me that even if 100% of the developers were using git-svn he would never allow git as the official location to store code on the servers and it must be pushed back to svn. I am not sure how to respond to people like that.
Edit: When pressed for more details it was clear that he was happy with his svn server setup and didn't want to change and have to learn something new. This was not a logical discussion, but an emotional one and as he ran the servers he had the final say.
Resistance to change might be an authority thing - you could just walk around this troll's bridge and see what happens. Try talking to his boss or higher about source control, casually. If his manager asked him to change to GIT he would do it. He might have dismissed you because he believes you're not in a position of power over him or the work required to switch is mundane.
apparently there was enough interest to bump it to the front page.
Thing is, I understand svn. I get it, and I have used it for years and know, almost instinctively now, what will happen when I use the various commands.
Git for me just feels overly confusing. I suppose it's just because I'm used to "the old way" but just take a look online at how many articles that are out there trying to explain how git works. Why are so many needed? It feels like there are many more than there are explaining how svn works.
By the way, I can see the benefits of git and I feel I'm getting the hang of it quite well now, but I still like svn :)
It's not really that hard to learn btw. There are some esoteric things that can be tricky but the Svn workflow can be learned in a day or two tops.
I think that we're still not fully taking advantage of Git because the other members of the team tend to commit and push all the time (so the code is backed up) but that makes rebasing and squashing quite difficult.
Git has also allowed us to work better with maintaining feature branches and bug fix branches which helps us to give individual branches to QA for test and only release those features/fixes that have been QA approved. If a feature isn't ready on time for deployment, it just doesn't get added to the integration branch and it'll go out the next release. That's pretty cool.
Personally, I've settled on using SourceTree as a GUI with a fallback to the command-line for certain operations.
As for squashing or not, I rarely do, and in general only squash commits that are simple typo fixes. This makes bisecting much easier.
Also, and this may or may not be relevant, maybe some people are reluctant to spam the central repository with their own temporary branches. To avoid this, I have everyone in my team set up a personal, backed up repository. Anyone can pull from these personal repos, but only their owners can push to them. When the temporary branches are done, they can be merged.
I like what Paul Stadig had to say on the topic http://paul.stadig.name/2010/12/thou-shalt-not-lie-git-rebas...
Really? How? I've always found Subversion branching to be painless, reliable, fast enough, and merges often go flawlessly, and it's really not been that hard to resolve merge conflicts.