Re: [Gnu-arch-users] library-add w/o ancestor checking?

[John A Meinel]

Zenaan Harkness wrote:
> > In particular, this revlib is 100% empty, so it's not just slow to
> > search for an ancestor in the library, it's completely pointless. ;)
>
>
> If it's empty why is it slow ??

I believe that is a "bug" in the ancestry search code. Basically it goes 
to the revision you want and says
        does this version exist?
                NO
        Who is your parent?
                my-category--branch--version--patch-149
        Does my-category--branch--version--patch-149 exist?
                NO
        Who is the parent?
                my-category--branch--version--patch-148
        Does ...

The problem is that it has to poll all of the remote server until it 
finds something that goes across an archive, or something like that.

I can confirm that if you have a large project with lots of patches 
(>100) and an empty revlib, the discovery phase can take a very long time.

I'm not sure how this could be improved, as the only way to get the 
entire inheritance heirarchy is to query each one for it's parent, and 
then you need to check that against the local revlib.

I would probably argue that it should "timeout" after a certain number 
of steps. But maybe some of this shows up in Aaron Bentley's 
backbuilder, and the discussion about keeping certain delta's cached 
instead of just patches and cacherevs. As far as I know, nothing came 
out of that other than a lot of discussion.

John
=:->

signature.asc
Description: OpenPGP digital signature