Reference data handling

[Jan Blechta]

The scripts in test/regression commit an eventual differences on top of master, ignoring whether a current reference is master HEAD or not. This leads to confusion and does not produce nice diffs. It would be better to commit diffs on top of current reference. This leads to the problem of branch naming. Please suggest any scheme.

[Miklós Homolya]

This leads to the problem of branch naming.

Probably not the kind of idea you are looking for, but Mercurial does not require commits to be accessible from named branches for pushing and pulling.

[Jan Blechta]

I did 15 seconds research on anonymous branches. Seemed that those are considered garbage in git.

[Jan Blechta]

Maybe we could store the data in FFC tree using git-lfs.