I gave preference to the email address with the larger number of
commits.

To find duplicates, I used the script
```bash
for i in $(git shortlog -nse  | sed s'/^\s*[0-9]*\s*//g' | grep -o '^[^<]*' | sed s'/\s*$//g' | sed s'/ /,/g'); do if [ $(git shortlog -nse | sed s'/ /,/g' | grep -c "$i") -gt 1 ]; then git shortlog -nse | grep "$(echo "$i" | sed s'/,/ /g')"; fi; done
```

I ran `git shortlog -nse | sed s'/^[ 0-9\t]*//g' | sort | less` and
manually inspected the result for duplicates.

Then I used
```bash
git shortlog -nse | sed s'/^[ 0-9\t]*//g' | sort | sed s'/<.*>//g' | uniq -c | grep -v '^ *1 '
```
to check that there were no remaining duplicates.

Of the three "anonymous" accounts, one ("coq") was used by multiple authors
(in particular Jacek Chrząszcz who developed the module system), while the
other two are each the committer of a single auto-generated commit so we rename
them to cvs2svn and serpyc-bot.