[darcs-devel] How could merge-by-value reduce changes of identity in rebase

[Ben Franksen]

Am 23.11.18 um 08:44 schrieb Ganesh Sittampalam:
> On 20/11/2018 12:27, Ben Franksen wrote:
> 
>> Ganesh Sittampalam wrote: 
>>> This change might also make your desired behaviour for rebase 
>>> feasible, where patches only change identity lazily, but we'd have 
>>> to think it through and I'm again nervous :-)
>>
>> Interesting. I don't see the connection yet. Could you explain in more
>> detail? (I guess this doesn't strictly belong here and should be
>> discussed separately.)
> 
> I think you've suggested before that patches should only change identity
> when being unsuspended if there are no fixups blocking them, or
> something along those lines. One of the reasons not to do that at the
> moment is that it would allow the identity of a patch's dependencies to
> change without the patch's own identity changing, which would no longer
> be an issue here.

Okay, I see. I always wondered why you think rebasing needs to change
identity regardless of whether we edit the patch itself or not. I think
I understand that part now.

> However, I think it'd also allow changes in the prim patches that were
> depended on (the "near duplicates" concept I mentioned in a comment on
> patch1758), so it would be more radical still.

Hm, yes. I guess to get the full thing here i.e. replace dependencies
whenever they establish the same context, we'd need a modified prim
patch format where hunks are always minimal i.e. insert or delete a
single line. Or some other mechanism that allows Darcs to see that two
different hunks aren't that different after all. (And yes, this is very
hand-wavey at the moment.)