Replacing svn with git/Mercurial

Git is the way forward! Would love to see that replace the SVN on here in the future especially if it's hard work to maintain it.

 

Seconding a vote for git.

 

I want to put for the case to replace Subversion with Mercurial:

With the platform diversity of the folks of Sonic Retro, I'd probably say that Mercurial is a better option, since it has been ported to many platforms and handles line endings a bit more gracefully than git does.

Mercurial is much easier to get a handle on, thanks to the trove of user guides available for it. For example, Hg Init is a tutorial site aimed at helping new VCS users and SVN users understand and get used to Mercurial. There's also a tutorial on Mercurial's official wiki, and a free official book on Mercurial.

TortoiseHg, after being rewritten to use Qt instead of GTK+, integrates much more cleanly into Windows Explorer than previous versions, and is very good at handling Unicode as well, since Qt has native UTF-8 support. And of course, TortoiseHg has a GNOME/Nautilus module, which no other TortoiseVCS variant has.

Mercurial requires no daemon program running in the background for the server repository to work, only that the repository configuration files are set properly. Mercurial will also shrink the total size of the repository many times over, because of the way the revlog works. It is also very easy to make a complete SVN import into Mercurial and preserve the entire repository history.

Mercurial is also quite fast on most platforms it runs on, whereas git unfortunately is not very fast on Windows because it relies heavily on UNIX style forking, which is emulated using either MSYS POSIX library or Cygwin POSIX library. Also, while git has gotten easier to use, it is still more cryptic and confusing than Mercurial.

Mercurial has been very well tested, with Mozilla/Mozdev, OpenJDK (Java), OpenOffice, OpenIndiana (OpenSolaris fork), Python, WiX, Enano CMS, and others relying on it for their projects, and having excellent reliability. Many of these projects chose Mercurial because it works very well in nearly all environments they develop under, be it Windows, Linux, Mac OS X, Haiku OS, or something else altogether!

Mercurial also has two free Visual Studio integration add-ins available: VisualHG (for MSVS 2005/2008/2010), and HgSccPackage (for MSVS 2008/2010). Integration with Eclipse, NetBeans, and other IDEs are available as well.

Some available web interfaces for Mercurial repositories are: phpHgAdmin, hgweb (included in Mercurial), TracMercurial, etc.

Maven, Redmine, and JIRA already include Mercurial support built-in.

 

I don't care what we move to, but two things:

1) I don't know if we should keep the current folder-based system or split every project into its own repository. Would there be any advantages to either way? I know that if we go the latter, then everyone would only have what they need...
2) What about tying commits to users? Or is that not going to happen?

Then you have the problem of sharing code, though I don't know how to resolve that except keep everything together.

 

Out of curiosity, is there anything in git or Mercurial that would render them immune to the recent issue(s) we experienced with SVN? If yes, then I fully support the move, but I don't care which VCS is used (git or Mercurial). If git and Mercurial would fare no better, I don't care which VCS is used (git, Mercurial or SVN), and will use whatever is available.

 

Simple: they're decentralized, so you have a copy of the repository that you can privately commit and then can push up to the server acting as the source. I'm not sure how conflicts are resolved, though...

Actually, there is no way we can continue to have the everything-in-one-asm-file approach if we go this route without edit conflicts up the ass if people decide to seriously work on these things. Sorry.

 

Jumping on the Mercurial bandwagon here. Whilst SVN has been a great advantage to many users here (and I'm not just covering the Retro SVN Project), Hg seems to be the better option compared to sticking with SVN, or jumping to Git. Plus, we need to have a system that can be used by many people, not just some.

 

In Mercurial, merging is usually done with "hg merge" and it will attempt to automatically merge files unless some edit differs at the same spot in the file being merged. Then Mercurial forces YOU to deal with it instead of having it try to do it and fail and make a horrible mess of things. Git handles it the same way. So does Bzr, and pretty much every other modern VCS developed in the last ten years.

It's bad development practice to keep everything in one ASM file, so you shouldn't be doing that anyway.

I'm not sure about git, but Mercurial would fare better because hgweb doesn't actually write humongous cache data all the time. And since there's no daemon running to keep a repository working in Mercurial, there isn't the overhead of another server caching in commits and writing them to the repository. This should also be true for git, but I'm not positive.

It might not be too difficult. One thing I've found that might help is something called SCM-Manager. It seems to support total user management for repos from the web interface. Its user auth system is extensible, and a plugin could be written to use the Retro user database and allow to specify which users are allowed to log in and push commits.

 

We should be polling the people that actually make commits to the SVN (myself, Andlabs, FraGag, Flamewing, Hivebrain, occasionally some others), because everybody else can use the web frontend if they just want to get the latest disassemblies.

Personally, I don't care much which way we go.

 

Really? What do you mean by complicated? The minimal I can think of is

Code (Text):

1. git init # new repo
2. OR
3. git pull # grab new changes from a server
4.  
5. git add <files>
6. git status # prints info
7. git commit -a # not sure what the -a means...
8. git push # push up to a server
9.  
10. # branching
11. git branch <newbranch>
12. git checkout <branchname>

 

I agree. As a C++ programmer, I have gotten used to the "1 class per source file" paradigm (even though I don't apply it ruthlessly), but this is harder to implement in an ASM context. In my own hack, I split up the s2.asm in a manner similar to the S1 disassembly, down to the splitting into multiple files of Sonic's and Tails' code; but I know that the latter is not well liked.

I would vote for restructuring the S1 and S2 disassemblies into a folder structure similar to the S3 disassembly, but also splitting object code into those same folders as their art, but not splitting (say) Sonic's code as is done in the current S1 disassembly.

FYI, I am putting my vote for "change", but I still don't care about which gets selected.

 

I was planning on doing this as an experiment (so I'll be hosting this change rather than putting it on the svn) to see how people would like it but didn't get to it this weekend =P Maybe tomorrow?

 

As for Mercurial, the equivalent commands are:

Code (Text):

1. hg init # new repo
2. OR
3. hg pull # grab new changes from a server
4.  
5. hg add <files> # add files for versioning
6. hg remove <files> # remove files for versioning
7. hg addremove # Removes files marked as missing from versioning and adds any new files not already marked for versioning
8. hg status # prints info about working repository
9. hg commit -m <commit message> # -m switch lets you add a descriptive message about the commit without making a text editor appear
10. hg push # push up to a server
11.  
12. # branching
13. hg branch <newbranch> # creates a branch from top commit
14. hg pull # Pull from server like normal to check out code
15. hg update -C <branchname> # allows you to switch between branches

And if you think Mercurial (and/or git, by extension) is more complex than Subversion, read this.