On Tue, Nov 27, 2007 at 04:45:22PM +0000, jamie wrote:
On Tue, 2007-11-27 at 17:35 +0100, Pierre Habouzit wrote:On Sat, Nov 24, 2007 at 02:33:12PM +0000, jamie wrote:On Sat, 2007-11-24 at 12:39 +0100, Pierre Habouzit wrote:On sam, nov 24, 2007 at 11:31:34 +0000, Pierre Habouzit wrote:Package: tracker Severity: normal Tags: patchwe can do this for tracker 0.7 - for 0.6.* its not backwards compatible with qdbm indexes and I dont wanna force a reindex unitl 0.7 jamieWell that's the whole point of the patch I sent, it convert databases on the fly. Given that qdbm/tc deal with something like 10^6 inserts every one or 2 seconds, it's likely that no user will even notice it. This conversion code could be kept for a few versions, until every user is likely not to have any old index anymore, and then the qdbm dependency could be dropped altogether. In debian terms, that'd be great to have the conversion code in lenny, and to drop it for lenny+1.sure But we will force a reindex in 0.7 anyhow as we will use the xesam metadata so may as well drop qdbm at that point if the replacement is superior
I see. Well, the replacement (If you meant tokyocabinet) has a very near API, and has even some nice in-place search and access to keys and values that can make everything even faster. For us (debian) the big issue is that qdbm doesn't support LFS, which is something we want to eradicate. One very nasty side effect, is that if you link a project to a library that has LFS support and qdbm, very very odd things happen, and it's not fixeable.
I will be releaseing 0.6.4 shortly and the one after that will be 0.7 so hopefully in a month or two you can
Okay, that sounds like a plan :) Thanks for your time. -- ÂOÂ Pierre Habouzit ÂÂO madcoder debian org OOO http://www.madism.org
Attachment:
pgp44BWmwk6QI.pgp
Description: PGP signature