Re: The new Codeville merge algorithm

Dear diary, on Thu, May 05, 2005 at 07:20:38PM CEST, I got a letter
where Bram Cohen <bram-V2j9jqiNNzWqrLZZDAjWCg@xxxxxxxxxxxxxxxx> told me that...
> Two lines will never be allowed to appear in different orders in different
> revisions (that would nearly guarantee an ambiguous clean merge). This
> leads to an interesting case. If there's an ancestral version which is ab
> (we tend to use strings where each character represents a line when giving
> examples) and one person makes aXb and another makes aYb, is aXYb
> possible? Obviously we can't have both aXYb and aYXb allowed, but one or
> the other doesn't cause problems. In ther interests of overall simplicity,
> we've decided to make a full ordering of all lines. The full ordering, as
> I'll explain in a second, makes merging a lot easier, and that's the
> motivation for doing it. Allowing aXYb is bonus.

This sounds quite unsettling. I'm rather conservative on merges, and
this sounds dangerous on its own. There is just a huge amount of cases
when this decision is wrong. Let's take these perl snippets:

        $foo = "abcd";          $foo = "abcd";
        if ($foo eq 'abcd')     if ($foo eq "abcd")
                { exit }                { exit }

Then if I understand you correctly, the merge would generate

        $foo = "abcd";
        if ($foo eq 'abcd')
        if ($foo eq "abcd")
                { exit }

I think you are just being way too liberal here, by allowing aXYb. It
would be much more reasonable to just disallow this case and make it a
conflict.

-- 
                                Petr "Pasky" Baudis
Stuff: http://pasky.or.cz/
<Espy> be careful, some twit might quote you out of context..