What we need from a version control system is
- archive all current and old files/pages.
- protect each user's files. When you modify somebody else's file you get a new file of your own instead; a branch.
- support tools to make it easier to see what's going on and merge branches.
The FPO should be designed to allow users to use version control with a minimum of fuss. It should take what you give it, and give you what you ask for.
Albert van der Horst suggested that a source control system must be absolutely mature, bug-free, and simple. Good idea. Do that part in Forth and there might be bugs, scalability issues, and very slow acceptance.
- Albert suggested that CVS fits that and RCS even more so.
- Charles Turner suggested darcs.
- Paul Bennett seconded RCS.
- Andrea Kochenburger and John Passaniti suggested Subversion.
- Trey Boudreau and John Passaniti suggested Trac.
- Ian Osgood suggested git.
- Mark Humphries suggested Aegis, which combines version control and testing.
Subversion is not mature. It has a lot of features. It may have overcome some limitations of CVS.
Darcs is less than 2 years old. It is organised along precisely the lines I want. It might be considered a competitor in some ways.
Trac does many of the things FPO ought to do. It's a competitor with a head start.
RCS is considerably simpler than CVS. Its latest version is 11 years old. Since 2000 there have been a handful of bugs reported in the Unix version:
- problem with more than two locks on one file.
- possible problem with locks on multiple files with the same name.
- possible problems from end-of-line from mixed DOS and Unix files.
- problem with symbolic links, RCS destroys the link when it should destry the file.
- occasional segmentation faults when deleting version 1.1 of a file.
- there's an error that sometimes runs diff even when the file contents are identical. There's a patch for that one.
Under Windows there's a report that occasionally important lines are truncated and merged.
Most of the Unix problems look avoidable. The segfault problem leaves behind a couple of files that must be deleted before RCS will run.
These appear to be rare problems, but if we use RCS in a way that's unusual we may run into problems that nobody else has reported. And there's the issue of Windows RCS.
