Shlomi Fish - Evolution of a Revision Control User

Last Updated: 30-April-2005

Genesis

When you are a DOS programmer in the late 80s and early 90s, people don't tell you about UNIX or about Source Control. Even when you start programming in Windows and then in UNIX, not everybody use it. So you use poor man's methods of cp myfile myfile.bak or myfile.txt.1 myfile.txt.2 or copying everything to a different directory. Stuff like that.

Those methods worked for me, and I rarely lost too much code. But still they were kludgy. By and by I started hearing buzzwords like RCS or CVS floating around. When working on the IGLU server, I was told that we can keep different copies of the files by an obscure "rcs" command. Also, I received an account on the KDE CVS repository where I uploaded some code I had. But I only had a vague idea of what revision control was and what it was good for.

I remember reading the Joel Test and was surprised that Joel Spolsky necessitated the use of source control and actually said CVS is fine in comparison to commercial offerings. This made me intrigued of what I had done wrong so far.

The Pleasant CVS Experience

When I worked on Freecell Solver, I kept a tar.gz containing every single minor digit on my hard-disk. I have 295 of them here. But for the IP-Noise Simulator project my partner and I took for the Technion, I decided to use CVS. So, we registered a project at BerliOS and started using it. The CVS import command gave us some trouble as we did not really know how to use it, but eventually we were right on track, and were amazed at the convenience.

I worked on the project from School and from home and kept getting changes back and forth through the server. Wonderful. I also used CVS for my second Technion project (a Seminars Management web-interface), and was quite happy.

The BitKeeper Years

When I started working on LM-Solve, I used CVS (naturally) and quickly ran into some of its limitations like no file renames, copies, etc. I had previously attended a lecture by Moshe Zadka about CVS (half of which I did not understand or it was still too advanced for me), where he said a project called Subversion will address all the limitations of CVS. I looked at Subversion and even joined the mailing list but could not get a public hosting for it.

I looked for something better and found BitKeeper, which was free for open-source developers and it was clear on the site that they had a hosting service. I read an old LWN editorial about it and was quite impressed with Larry McVoy's, BitKeeper architect's attitude towards his product and the free software developers community. (As it turned out, its licensing has changed a lot since the editorial had been written).

I contacted BitMover through the E-mails I found at the site, and got to ask them a few questions. They were polite and explained to me what to do to set everything up. As it turned out, I should have used the bitkeeper-users mailing list instead, as that specific address was geared towards paying customers.

In any case, after a period of neglect, I set up a repository and started hacking. The only thing I can say about BitKeeper is that it is wonderful. It feels very slick and is very slick. One works with a copy of the repository, so there isn't a constant client-server traffic, and its offers a lot of other power. I was very much impressed.

I ran into a few problems for which I consulted the mailing list or was able to solve on my own and I recall that I was once able to help a newbie out myself. I also discovered the BitKeeper source code was no longer available online for all to download, which did not bother me much, as I was hoping I could eventually win the vendor's trust to get it myself.

In any case, I liked BitKeeper so much, that I even thought about eventually converting Freecell Solver (that was converted to CVS a few months back) to BitKeeper and BKbits.

No BitKeeper for You!

This all changed when Larry McVoy announced a change in the BitKeeper license that required all the contents of the free users' repositories to be made available online. I did not think too much of it at the time, but then I realized it was not something I could agree with ideologically. The BitKeeper license forced me to use the most recent version, even if it had a different license. So BitKeeper was taken right out of my hands.

I sent a letter to bitkeeper-users titled "Larry - You Lost Me". Larry calmly responded that he was sorry to hear that, but such discussions were off topic their. This caused an entire chain of events whose end results were:

1. My public repositories were deleted without my explicit consent.
2. Larry became angry several times.
3. I started the "Better SCM" initiative, in order to empower the other open-source alternatives that were better than CVS (but perhaps not as good as BitKeeper)
4. I decided that I wished to contribute to one source control project or another.

I liked BitKeeper and still would love to use it, but was appalled by Mr. McVoy's attitude towards free users. Note that this was before the fiasco with the non-compete clause was Slashdotted. (Note that I still respect Larry as an engineer and a software vendor who is trying to sell BitKeeper to software houses.)

So I was looking for something else.

A brief encounter with Aegis

When I was first directed to the Aegis homepage by a user who read my Advogato diary where I had said I was looking for something better than CVS, I could not understand half of the features that were presented there. But after I used BitKeeper, all those features made sense, and I was impressed by its capabilities.

I joined the Aegis mailing list introduced myself as a BitKeeper refuge and asked for interactive tutoring at a timezone close to Israel. The first things I found two people who could help me, one of them was Mark Veltzer, who was living in Israel and used Aegis.

He tutored me through the IRC for two days, in a session of several hours on setting it up and getting a change in. From the transcript I could gather the process for the other stuff, and I could continue work on it. However, I noticed the Aegis process was an overkill for the small projects I wanted to manage with it and for my programming habits.

Being interested in other alternatives as well I also installed Arch and Subversion.

Subversion - The Silver Bullet, that was not made of Gold

I started using Subversion, and immediately saw that it was very nice to use, a bit like CVS. It felt very slick and was very slick. Setting it up was a bit hard, with having to set Berkeley DB 4.0 and Apache 2.0 beforehand, but afterwards it worked nicely. Upgrading turned out to be slightly easier, with the ability to check out the latest copy out of the repository, build it and install it.

As opposed to BitKeeper, Subversion is not distributed, but this feature is planned. Still, it is very stable and usable, (with a few quirks that can be avoided with some expertise), so I have used it ever since. I still maintain Freecell Solver with CVS, because I don't have Subversion hosting. Nevertheless, I can probably become a junior partner in a co-oped server hosted at an Israeli ISP, where I can set up Subversion hosting. I'll have to pay some money for it, but I guess there's no such thing as a free lunch (at least until the hosting services start supporting Subversion as well, which I hope will happen soon).

Verdict

I use revision control even at home for such things as doing my homework (in LaTeX, C, or Scheme). I used CVS for this, but now converted to using Subversion. Aegis seems very powerful, reliable and nice. Still, it suffers from some childhood diseases that make it unsuitable for me: its operation is complicated and tedious, and it is a file-system only system (which can only be pseudo-networked over NFS) with many non-Win32 UNIXisms. Subversion, on the other hand, is easy and fun and just works.

I have started contributing to Subversion and one of my patches has already got accepted. I plan to contribute more. I still miss BitKeeper, and wish I could use it. If Larry McVoy is planning a land grab on a market dominated by CVS, he'll have to make sure the license is acceptable by the free software community. I am not a free software "fanatic", yet even I could not cope with the attitude and license change.

Larry may be confident that there is not an open-source alternative that is up-to-par with BitKeeper at present. But it does not mean there will never be one. Small teams of open source hackers, were known to create products that surpass in quality those of competing commercial companies who seemingly had much more personnel and resources on their hands. (e.g: gcc, GNU Emacs, The GIMP, KDE and GNOME, GNU/Linux itself, Perl/Python/PHP, Apache, Zope) While I cannot guarantee that the same will happen to BitKeeper (which at the moment deviates very much from the Bazaar style of development), it is a possibility if a few dedicated hackers populate the competing projects.

And even if this will not be the case, I am certain BitKeeper will not become the de-facto standard of the Open-Source World that CVS is today. It's like saying that Microsoft IIS+ASP will replace Apache and PHP or Perl soon.