cvs2svn fails with “xxx is not a valid ,v file”

Question

I finally found an answer to my question when I wanted to post it! However I'll still post it, including my answer, in case it helps someone else:

When converting from CVS to Subversion cvs2svn failed on some files with the message

"xxx is not a valid ,v file"

What's the problem?

Answer 1

As it turns out CVSNT omits the last 0xa from some files where cvs2svn needs them. This can be easily fixed with the following c# code:

static void Main(string[] args)
{
  foreach (string file in Directory.GetFiles(args[0], "*,v", SearchOption.AllDirectories))
  {
    using (FileStream sin=File.Open(file, FileMode.Open, FileAccess.ReadWrite))
    {
      sin.Position=sin.Length-1;
      if (sin.ReadByte()==0x40)
      {
        Console.WriteLine("fixed "+file);
        sin.WriteByte(0xa);
      }
    }
  }
}

Answer 2

In my case there was corruption in the symbols section of the xxx,v file. The expected format is tag_name:tag_rev, but there were instances of:

• Missing :tag_rev
  e.g. tag_name
  Fixed by deleting the line.
• Multiple tag_name
  e.g. tag_name1:tag_name2:tag_rev
  Fixed by removing the second tag name (which one you remove probably depends on what they are).
• Invalid name/revision delimiter. In my case the invalid character was always z (there is only 1-bit difference between ASCII : and z).
  e.g. tag_nameztag_rev
  Fixed by replacing the z with :.

To help during my investigation I added a print line to cvs2svn_rcsparse\common.py. If parsing the symbols fails, the last tag printed is the cause.

def _parse_admin_symbols(self, token):
while 1:
  tag_name = self.ts.get()
  # WileCau print the token and tag_name
  print 'token=|%s| tag_name=|%s|' % (token, tag_name)
  if tag_name == ';':
    break
  self.ts.match(':')
  tag_rev = self.ts.get()
  self.sink.define_tag(tag_name, tag_rev)

The additional print adds quite a lot of noise to the output so it might be nicer to only print if an exception happens, but this was good enough for my needs.

---

I also found this link which turned out to not be my problem but may help someone else. Credit to Christian Haarmann for documenting it.

http://tigris-scm.10930.n7.nabble.com/suggestions-for-cvs2svn-fix-for-error-quot-myfile-txt-v-is-not-a-valid-v-file-quot-td54240.html

In case the link becomes invalid, the summary is that someone had edited the xxx,v file and their editor had replaced 0x0A (LF) with 0x0D/0x0A (CR/LF), and the additional character caused the parser to think the file was corrupt.

Answer 3

I've also such an error. When I use cvs2git in order to migrate a cvs repository to git, this error occurs for several files. I have detected that there is missing a closing 0x40 (@) at the end of file.

So my solution is:

1. Open the corrupted cvs-history-file e.g. with vim (maybe in binary mode)
2. Add the missing @

If this doesn't fix the problem, then compare the content of the corrupted file with the RCS-file format: rcs_man_page

Answer 4

One way to troubleshoot this is to run rcs log *file,v*, which may provide you some insight.

In my case, I had some files missing @'s, some files missing a semicolon, and the tool I used to import my old repository onto the cvspserver had thrown in an unreferenced version.

Good luck!