Pavel Sanda wrote:
Vincent van Ravesteijn - TNW wrote:
I'd be happy to get feedback from you (and other experts on this issue),
its very good somebody detached from the past flames works on this and i hope
we won't lose next developer :) it would be indeed better to dissect the feature
into smaller chunks so the focus stay unchanged, but i'm not sure there was
ever any agreement on use cases, which would be imho the first thing.
Right. But part of the problem with the previous attempt was precisely
that it proceeded, in trunk, in small pieces. There was an initial
implementation, consisting of several patches over a long period, and
then that one was partly reverted, which led to another long sequence of
patches. I don't think anyone but Bo was able to follow what he was
doing, and, as Bo said in his message, it's clear from the need for a
massive reversion that even he didn't know exactly where he was going.
So when people did finally start to pay attention, and when problems
started to arise, everyone had to do a lot of catching up, which made
things difficult all around.
So, while I do agree that there is probably work that could be done at
an infrastructure level before any big decisions have to be made, I'm
also wary of anyone's starting off in a direction that might presume
those questions had been answered one way rather than another. Which is
to say that, just perhaps, a discussion of use cases and the like would
make some sense. Different implementations make more or less sense
depending on which use cases one has in mind. Some need per-file
embedding, some don't. Having per-file embedding (as opposed to what,
say, OOo has, which is total bundling) is obviously going to be a lot
more complicated, since total bundling is just the simple case of
per-file embedding. So we only want to go that way if we're convinced
that's an IMPORTANT use case, not just a theoretically possible one. And
there, I have no strong opinion. I'd like to see a reasoned discussion
of the matter, if that's possible. And one that is independent of any
issue about reversibility, unbundling, and the like, which is quite a
separate issue.
Richard