[ajs]

After years of procrastination, I finally got around to finishing off the SVN->Git migration for MegaZeux's source code. The old SVN repository, hosted on SourceForge, is now obsolete, and anybody who had write access to it has had their access removed. Consider sourceforge.net's SVN repository dead (though I may still release tarballs there; we'll see). The repository won't be deleted because its history is better in some ways (described below).

The new source repository is now available from github. I haven't decided whether to use the push or pull model; we'll sort that out when somebody wants to contribute some new code. Please feel free to try setting yourself up with Git; ask me any questions on this thread. I have updated the MSYS documentation for folks using Git on Windows. It includes instructions for installing MSYS/mingw, Git, and checking out and building MegaZeux.

As it stands, the new repository has every GPL version of MZX released, all the way back to 2.51g. The DOS versions can be built in DosBox with Borland Turbo C++ 3.0. I have removed the original PRJ files (they had hard-coded paths from Exo/Koji's machines) and replaced them with Borland-syntax Makefiles. The migration wasn't perfectly smooth as there were lots of issues introduced by the automatic CVS->SVN migration I did some years ago. The git history is more or less a direct import of SVN after 2.81e, but some intermediary versions won't build correctly. Versions prior to 2.81f have been imported as "snapshots" of released source. I stopped tagging releases after 2.81h too, because 2.82's trunk diverged too much from the branch -- I may revisit this later. Non-mainline branches have not yet been imported, but that's also on my TODO list.

You can get a list of tags with "git tag" and switch to a tag with "git checkout v2.81e" (for example).

A full clone of this repository, including all versions from 2.51g to SVN HEAD is just over 5MBs. Happy hacking!

--ajs.

[Exophase]

Why?

[ajs]

Fair question! The biggest reason for the move was to get rid of sourceforge, which has slow and unreliable servers. Operations like "svn log -rN" and "svn diff -rA:B" were very slow, and as of recently, sometimes would time out altogether. It was becoming exasperating to use. This is hardly a recent problem, though; I started the Git repository almost 2 years ago for the same reason.

Why Git? I thought about just moving the SVN server, certainly that would have been easier, but when folks forked the code, for example dsmzx and Revvy's robotic debugger, they'd put their code in other version control systems. I guess the primary motivation was that a) SVN isn't very good and b) it's not very easy to "clone" an SVN repository. Distributed version control's much better at that, so that narrowed my choices. I ultimately picked git because it was familiar to me (I use it at work) and the Windows port was acceptably functional, and constantly improving.

Also, sometimes it's been useful to be able to switch to the old versions to see how a feature was implemented, and it's much handier to do that from version control than with tarballs.

--ajs.

[Wervyn]

From my perspective as an end-user who occasionally pokes at the code, I only really care that the system works. It works, so I'm happy. Hooray!

"git r done"

[Exophase]

Yeah, I don't really care either, since I'll probably just be checking things out. ajs's reasons sound good. I've just had kind of an interest in the git phenomenon since it seems a little vague, and a little pushed as Torvalds Word of God. Not to knock the guy, he's certainly a more capable programmer than most of us, but his tirades can be pretty overblown and arbitrary. Like his C++ hate, and this is coming from someone who almost exclusively programs in C over C++.

[Terryn]

What does this mean for your nightly build system? Will it be much of an issue to adapt to the new repository?

[ajs]

I'll probably fix it tomorrow. The only visible change will be directory names (from SVN rev to date/time + git sha).

--ajs.