In the past we have used a specially programemd mode in emacs to record changelogs in source files. Since using CVS for source code maintenance, we also record messages using CVS. We have now discovered that CVS can automaitcally insert those messages in the file.
This mechanism might render the existing emacs-mode logs obsolete

The CVS documentation says

The CVS documentation mentions a few problems associated with the auto-insertion $Log$ approach.

  1. While the file is in our system, we can get the log info using cvs log (aprs in our case). Once the file is exported the information might be useless anyhow.

    NOT true in our case - files are often moved back and forth, even between developers with an interest in log info.

  2. RCS is not good at handling $Log$ info when a branch is merged onto a main track. It may result in conflicts.

    NOT relevant misbehaviour in our case, as we don't use branches.

  3. People would be able to amend log info that is inserted into the file, when they are editing the file. People might tend to fix typos or even factual information. As the CVS internal log is not updated, it might result in inconsistencies.

    COULD be a problem, but discipline is required not to amend log info once inserted ion the file.
    If the log info gets inserted at the end of the file, we might resist the urge to fix spelling mistakes etc.

My own opinion

Reasons FOR keeping emacs mode logs and CVS logs seperate

I quite like the format that is used by the emacs mode. The messages entered have been sparse in the past, but people have been better recently about inserting concise and descriptive message at the file header. These messages often contain important developer information, like function x moved to y, or variable a renamed b etc. whereas the cvs logs should be more verbose and of interest to people that haven't been involved in the development of that particular module (like me, when I'm writing release notes from the cvs logs).

Changeover problems

One of the problems is that all logs up to now are in cvs and not in the files, although a script could pull out the cvs logs and add them to *EACH and EVERY* file, as if they'd been inserted by cvs in the first place.

The other problem, is the fact that the emacs headers contain vital information that might not have been logged in the cvs messages. Merging them might take up a lot of time.

I wouldn't agree with a gradual changeover, that leaves the emacs logs at the top, but now continues with the cvs insertion. Acedb suffers too much already from gradual changeovers that have never been completed and were never done 'all the way'.

Reasons AGAINST emacs logs and FOR cvs logs only

The format of the CVS log insertions is quite nice also. This is an example :

/*
 * $Log: filename.c,v $
 * Revision 1.5  1998/10/14  09:17:07  fw
 * did something

 *
 * Revision 1.4  1998/10/14  09:12:58  mieg
 * small changes bla bla
 *
 * Revision 1.3  1996/05/09  15:16:47  rd
 * Fixed function xx()
 *
....
*/
  1. They take up more space, but when inserted at the end they would be managable.
  2. People might think more about what they put in the log, as it will be added to the file itself (i.e. it becomes more visible) not just some anonymous system that has to be prompted to show the logs.
  3. Not everybody uses emacs or has this mode file available. Using cvs logs would be truly cross-platform & cross-editor. (I'm thinking of rbrusk and windows here).
  4. If the file is simply renamed, CVS truncates the log at the point when the renaming took place and just refers to the old name where the rest of the log is. But the log insertion in the file would continue in a linear fashion and it would not suffer from that.
    It would also make 'aprs' or 'cvs log' absolete, because one can just 'more' the file or open it in emacs (whilst one is editing it anyway) to get access to the logs.
  5. Once the changeover has been completed (as described above - no half-baked jobs please). All the logs would be at the bottom, every file could contain a proper header at the top, and go straight to the code. This would make the files look a lot cleaner when browsing through code.
    We also wouldn't need the /**** end of file ****/ lines at the end of each file. The end would be marked by the start of the log section.

Fred's Summary

I am in favour of a changeover, but it certainly required a lot of work. Most of that work would be to merge emacs log with cvs logs, which might be tediuos in some cases.

While we do that cleaning up the headers of some source files would be nice, as some are badly cluttered with stuff that should be thought about again and/or move to private header files.

During the changeover, we would also have to touch *EVERY* file in the CVS system, and might gather a better idea of what is there and how to pull it together better or how to modularise it better.

If we agree, that the CVS log mechanism should be adopted, we only have to problem to decide whether it is worth investing the time to do the changeover properly. I think we'd all benefit from it. Changelogs are a good idea, so we should make best use of them.