To summarise the discussions I had with Pie=
rre(Pinaraf) and on the main channel (which also involved Cyrille), there a=
re two types of versioning:
- the basic one which would simply store=
a complete "odf file" of each version in a sub-directory. This i=
s how OOo is doing it. The advantage is that it does not really impact the =
rest of the suite. The disadvantages (to my mind) are that first redundant =
information is stored (and in the case of OOo implementation, if I remember=
properly, as compressed zip inside the zipped file), and second that it do=
es not allow workflow like version control system allows (revert commits, c=
ommits history, ...).
- the (to my mind) better system would be linked with the change traking sy=
stem. It could store which set of changes belong to which version. This wou=
ld allow more complex workflow of the versioning system (closer to what ver=
sion control allows). This is however far too complicated for a GSoC and mo=
re importantly, there are unfortunatly, today, some blocking points to such=
a system.
=A0 The first comes from the odf spec itself: tracking a format change is r=
educed to tracking that a format change took place. This means that on chan=
ging for example italic from false to true, the affected fragment will be t=
agged with a "format changed region" but the information about wh=
at change actually took place (italic, was false, is true) is not saved in =
the odf spec. This is a must have if we want to implement more than the bas=
ic versioning. I already have some ideas on how we could get around the pro=
blem, but the ideal solution would be an evolution of the spec.
=A0 The second limitation is the change tracking itself. As the versioning =
would store set of tracked changes, the change tracking system must be ther=
e (and complete) first. When I say complete, this means not only text chang=
es are tracked, but also frames, images,... As previously mentioned, this i=
s what I will be working on mainly when the freeze is over, because that is=
for me one of the building bricks of further evolutions towards a "co=
laborative framework".
Ultimatly I think that the two are not necessarily mutually exclusive. =
If you take an analogy with archiving, the basic system could be used to do=
the "snapshot backup", whereas the change-tracking-based system =
could do the "incremental backup". Or one could also think of the=
basic system as a way to do a "snapshot branch".
Pierre(St)