[PATCH 4 of 4] revlog: support writing generaldelta revlogs

Sune Foldager cryo at cyanite.org
Sat May 7 06:02:54 CDT 2011


On 07-05-2011 11:32, Benoit Boissinot wrote:
> On Sat, May 7, 2011 at 8:52 AM, Sune Foldager<cryo at cyanite.org>  wrote:
>> On 06-05-2011 22:34, Benoit Boissinot wrote:
>>> I used a postorder, which works pretty nicely (similar to shrink-revlog):
>>> http://paste.pocoo.org/show/384245/
>>
>> Alright, a postorder approach is definitely good.. however it doesn't take
>> into account deltaparents. My simple algorithm above is, as it turns out,
>> too simple as well. I think we need the beeeeeeeeest of both worlds*, so a
>> postorder-prefering-deltaparents. I have something in mind :).
>
> See the "# XXX explain reversed" comment, if I remember correctly it
> chooses to explore p2 first which means the chain will favor p1delta.
> I think that's where you should play.

Yes, exactly. For generaldelta, I would choose to favor the non-deltaed 
parent first, if any of them are deltaed, in order to possibly build a 
longer chain. Since GD is currently PD (once pushed, at least), it will 
work just like that.

Sune :)


More information about the Mercurial-devel mailing list