> If this conversion looks good to you and Werner I'll add a commit to
> it describing the conversion process and install it at Savannah.
It looks good, thanks a lot! However, there are some issues.
1. The commit entry for 1.10 is
groff before CVS: release 1.09
and should obviously be
groff before CVS: release 1.10
2. Look at the commits from 2000-02-06, starting at 10:34:01
(385c8b0e8cd19436ee3ba12d5d8970e27e0b0dac), ending at 10:40:10
(58d43cfe3ec21b13fe3f75501b081bca92f295a3). This is the initial
import of groff into the Savannah CVS. Can this be handled as
single git commit?
I'm quite sure that I did this as a single CVS commit which took
those six minutes due to a slow modem connection.
There are many more such series of commits, and I think such series
should be handled as a single commit in git, too. Can you improve
your tool to handle that gracefully? Perhaps unifying commits that
have a small time delta (relative to the next commit) and either a
missing log message, or a log message but no change.
3. The commit bd3759da506a6fc989f9766313f2d38d66e4e1e1 (`Forgot to
commit this file.') exhibits a bug in the conversion: The affected
file (tmac.doc.old) was not part of the first CVS import, but
present in the 1.15 tarball. It seems that the step
version 1.15 -> first CVS commit
didn't properly remove files not in this first CVS commit.
4. For those changes which intentionally lack a commit message, I
suggest that the name of the changed file(s) is used to synthesize
one, up to a certain string length. Example for
8bf10b34a1f339b2c69a683e08ad4edd91accc3d:
Update `ChangeLog'.
We might use the word `Update' for the `ChangeLog' file, and
`Change' for everything else.
Werner
