Re: Supporting convergence

Martin Pool wrote:

> In weave terms, this seems a bit like allowing for lines to be
> reinserted  following a revision that deletes them.

Exactly. Implicit undo requires even more complicated line states.

> Perhaps there are practical problems in automatically inferring
> reversions, if you're doing it just by looking at text changes?

Uh, yeah, I've been giving examples.

> It is probably common to incidentally change some lines back to their
> previous version, without the whole commit being a reversion.

Umm, yeah. That's 'implicit undo', and like I said, it doeesn't seem to
play nice with convergence.

> It's probably also common to semantically undo a change, but have some
> context or whitespace changes that mean the reversion is not seen as
> such.

If it's not completely identical, as far as the version control system is
concerned, it's not the same.

> Nevertheless njs's idea of trying to solve this for scalars first before
> worrying about text files is probably a good way to proceed.

That whole last post was just about scalars.

-Bram