Login | Register
My pages Projects Community openCollabNet

Debunking BitMover's Subversion Comparison

Summary

BitMover, Inc. offers a comparison between BitKeeper (their main product) and Subversion, at

http://www.bitkeeper.com/Comparisons.Subversion.html

BitMover's comparison contains various false statements about Subversion, and some misleading statements. Below, we analyze BitMover's page and set the record straight about Subversion. The entire relevant content of the BitMover page is quoted below, as it looked on 2 July 2004. We've omitted only extraneous things like navigation links and the endorsement quote at the top of the page.

Please note that this response is strictly about Subversion. We have no complaints about BitKeeper as a piece of software; we've heard mainly good things about it, though we can't try it ourselves because of its license situation.

BitMover's text below is quoted in red italics where we feel it's in error, and green italics otherwise. Our annotations are in regular black text.

The Response

Subversion is a new system which is supposed to replace CVS. Unfortunately, Subversion shares many of CVS' problems and introduced some of its own problems:

Subversion uses a binary file format for your revision control data and metadata and if that format gets corrupted you are out of luck, your whole team comes to a halt.

That bit about the binary format is classic FUD. It's quite true that Subversion uses a binary file format, but think about it: BitMover is basically saying that databases are a bad idea, since virtually all modern databases use a binary format. Sure, if anything happens to those files, you're out of luck, assuming you have no backup. So, does this mean no one should ever use a database?

Anyone who runs, say, Oracle, MySQL, Postgres, or another database system should be able to see the silliness here right away. Furthermore, it's clearly the case that if BitKeeper's data storage format got corrupted severely enough, your whole team would come to a halt too, at least for as long as it took to get everything fixed, restored from backup, whatever. Not so different from Subversion, really.

There are database management issues to administering a Subversion repository, of course, issues which CVS does not have. If this is BitMover's complaint, then they should say so, rather than making a meaninglessly broad generalization. (Note that Subversion now offers a non-database back end, for those who don't want the overhead of administering a database. This is a fairly recent development, and it's likely that BitMover simply hadn't heard about it yet when they wrote their page.)

Subversion has a single repository model, i.e., client/server. Each work area is clear text only which means no revision control in the work area during development.

This is basically true, though note that this way seems to work fine for a lot of people. Strictly speaking, Subversion has exactly one level of revision control in the working copy, since there are pristine base texts of every file present. One can locally view one's changes, revert them, or save them and then revert, all without a network connection.

No staging areas to protect the main source tree. With Subversion, everyone checks into the same place and if someone breaks the tree, it's broken for everyone. With BitKeeper, you can put a staging area between each group of developers and the main integration tree, thereby protecting the main tree from bad checkins. Anyone who has lived through a change that broke the build can see the value of staging areas.

False and FUD.

Having a staging area is a matter of how the project organizes its repository and moderates its commits. The fact that some projects choose to use their main development trunk as their staging area, and release from branches, is a matter of policy, not of technical limitations. If you want to insert an extra level of staging area, Subversion supports that just fine.

Now, there might be a kernel of a real complaint here, which is that Subversion doesn't have first-class changeset swapping, and therefore merges from one branch to another (say, from staging to live) don't preserve as much history as they could. However, BitMover's text goes far beyond that simple, circumscribed complaint. And since they make the merge complaint separately later anyway, they either thought they were saying something different here, or they're making the same claim twice, for the sake of inflating their page.

Subversion loses information every time there is parallel development because you are forced to merge before you check in if someone else checked in first. The state of your workspace before the merge is lost forever. Another way to say this is that if there is N-way parallel development, Subversion loses N-1 events.

As worded, this is not quite true. But probably what they meant to say was "you are forced to merge before you check in if someone else checked in changes to the same files first." Then it would be true.

So how important is it? It depends whom you ask. BitMover probably has in mind the "star merge" work style that has developed in the Linux family of OSs, where there is never any expectation that all workspaces will converge onto a single canonical form. It's not about how far behind you are, but about how different you wish to remain; less about merging than about cherry-picking. There are Subversion users who ask for this. We usually point them to SVK, which is based on Subversion and supports this style of development, or to Arch, an open-source revision control system unrelated to Subversion. But some others do not find this a compelling feature; see Greg Hudson's explanation why not.

Merging in Subversion is no better than CVS, i.e., primitive at best.

This is basically true. Subversion's merging interface is better than CVS's, but the underlying operation leads to the same result as in CVS. Better merge support is on Subversion's long-term todo list.

Branch management in Subversion is a nightmare.

This is false, at least insofar as it means anything at all. Without more technical substance to respond to, it's hard to know what to do with such a claim; they don't say anything more concrete than "nightmare", unfortunately.

Branch management is actually one of Subversion's strengths. In Subversion, branches are first class, versioned objects, just like regular paths. They can be copied, renamed, deleted, and resurrected — and people do these things all the time. Perhaps BitMover was really talking about merging? But again, they make the merging complaint elsewhere, so this item must presumably be about something else.

Subversion has no integrity checker which means files can be silently corrupted and you will never know until you try and roll backwards.

False. Subversion ships with an integrity checker (svnadmin verify). It also attaches a checksum to every revision of every file. It verifies these checksums automatically at every opportunity where it wouldn't have a serious impact on performance; furthermore, it offers you the ability to paranoidly check integrity even more often, if you want to, via the abovementioned command. So we really don't know what they're trying to say here.

Subversion has only weak rename support, that's something that is inherent in all centralized systems.

It's true that Subversion has weak rename support (work is under way to fix this). We don't see any reason why this is inherent in centralized systems, or even why it has anything to do with centralization or lack of same. It sounds as if BitMover is trying to imply that Subversion, due to its design, simply can't have good rename support. That's the more serious charge here, and we believe it's totally bogus.

Conclusion:

What's really going on is that BitKeeper has just one or two solid complaints to make: "Subversion has only primitive merge capability", and possibly "Subversion doesn't do decentralized versioning."

While these are true, it seems like BitMover is attempting to say each one multiple times, to have a longer itemized list of deficiencies in Subversion. If we ever write up our own comparison between Subversion and BitKeeper, we'll avoid that sort of inflation. Unfortunately, we can't easily make such a comparison, because BitMover, Inc changed its license to prevent free software developers who work on competing products (such as Subversion) from using BitKeeper.

BitKeeper's Licensing Situation:

Before they changed their license, anyone could use BitKeeper for a free software project (with a few minor caveats). Although BitKeeper was not "free" or "open source" in any important sense, since BitMover still retained strong intellectual property rights over the code, it did mean that free software developers could try it out and learn from it.

Then BitMover added a clause to their license, ending this availability for Subversion developers and others:

    (d)  Notwithstanding any other terms in this License, this
         License is not available to You if You and/or your
         employer develop, produce, sell, and/or resell a
         product which contains substantially similar capabil-
         ities of the BitKeeper Software, or, in the reason-
         able opinion of BitMover, competes with the BitKeeper
         Software.

(We'd include a reference to the full license here, but it appears one must fill out a download form even to get to the license's text, and that's a little more trouble than we're willing to go through.)

Larry McVoy, founder of BitMover, confirmed that the new clause prohibits Subversion developers from trying BitKeeper at no charge. Of course, Subversion developers are free to purchase a license, but that is not a realistic option for most of them.

You can read more about the BitKeeper license controversy at these links:

This is why we cannot make our own page comparing Subversion and BitKeeper. Of course, BitMover employees are free to download and install Subversion any time, and we encourage them to do so.