This is a program to convert a CVS repository to Subversion.
Bob Miller <kbob@jogger-egg.com> is working on it.

-----

Current status (October 1, 2000):
	
	cvs2svn.pl can read the whole CVS repository and generate a
	sorted list of commits, using almost the same heuristic as
	Karl's cvs2cl.pl.  See Commit::accepts_rev for the heuristic.

	The commits right now are simply of the form:

		commit
			path/to/file1 1.23
			path/to/file2 1.34

	The big task that remains is to produce an XML vdelta that
	subversion proper (not sure which part) can parse and process.

	There's a bug in the Rcs package which mixes up log messages
	and diffs.  This means (A) it's not coalescing commits right,
	and (B) it's wasting a lot of memory storing a large fraction
	of the repository's diffs.

	Yes, performance is a problem.  See timings below.

	Packaging is not done:
		cvs2svn.pl is too big.
		Rcs.pm is in CPAN, is not in this distribution.
		CheckedClass.pm is duplicated in cvs2svn.pl,
		because I was too lazy to set up @INC.
		Makefile should invoke tests.
		No documentation.

	cvs2svn.pl needs about 1,000 more comments than it has.

-----

Some timing info:

	Test system is Pentium II 450 MHz, 440BX chipset, 192 Mb RAM,
	Linux 2.2.16, single 7200 RPM IDE disk, Ultra-ATA/33.

gnats-cvs (small):
	55 directories, 647 files, 3165 file revisions
	6 unique authors, 10 unique tags, 314 unique log messages
	55 directories, 647 files, 3165 revisions
	used 4440 Kbytes
	13.250u 1.950s 0:18.10 83.9%    0+0k 0+0io 3264pf+0w


gcc-cvs (medium):
	310 directories, 9825 files, 96686 revisions
	used 172936 kB
	2825.160u 92.610s 51:06.30 95.1%        0+0k 0+0io 155728pf+21120w

NetBSD-cvs (large):
	5724 directories, 65225 files, 760990 revisions
	used over 500 Mb.
	Did not finish running overnight.