Re: The new Codeville merge algorithm

Walter Landry wrote:

> Bram Cohen wrote:
> > Once we have a full ordering, merge becomes surprisingly
> > easy. Simply combine line states, and pull out all lines from the
> > list.
>
> Does this mean that you only have minimal context?  That is, if you have
>
>   A
>  / \
> B   C
>
> A=abc
> B=aXbc
> C=abYc
>
> then you get a clean merge of
>
> aXbYc
>
> That seems troubling, especially if line "b" is a blank line.

It's minimal context for matching, but not for resolution, which is as it
should be, because the match actually has some history information which
make single-line matches not dangerous.

What would be dangerous is that sort of match on a resolution, so if a
user started with abc and ended with aXbYc and b was just a blank line, it
would be very dangerous to match the b's. With the old non-precise merge
algorithm the same two-way diff was used for both match and resolution,
so the b's wouldn't match, but I think the greater aggressiveness of the
match which uses the history is actually a good thing. One could require
match sections be two lines long though, that's an easy modification.

-Bram