From mboxrd@z Thu Jan 1 00:00:00 1970 From: Mark Fasheh Subject: Re: [PATCH 0/3] btrfs: extended inode refs Date: Thu, 12 Apr 2012 09:19:03 -0700 Message-ID: <20120412161903.GP17950@wotan.suse.de> References: <1333656543-4843-1-git-send-email-mfasheh@suse.de> <4F7E0AF9.7070305@suse.de> <4F858312.6090905@jan-o-sch.net> <20120412161113.GM29506@shiny> Reply-To: Mark Fasheh Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii To: Chris Mason , Jan Schmidt , Jeff Mahoney , linux-btrfs@vger.kernel.org, Josef Bacik Return-path: In-Reply-To: <20120412161113.GM29506@shiny> List-ID: On Thu, Apr 12, 2012 at 12:11:13PM -0400, Chris Mason wrote: > On Wed, Apr 11, 2012 at 03:11:46PM +0200, Jan Schmidt wrote: > > Hi Jeff, > > > > > >> An alternative solution to dealing with collisions could be to > > >> emulate the dir-item insertion code - specifically something like > > >> insert_with_overflow() which will stuff multiple items under one > > >> key. I tend to prefer the idea of > > > > > > I vote for this option. > > [ Big patch series, thanks Mark! ] > > I prefer the insert_with_overflow because it makes the deletion case > less complex. If we handle collisions with bits in the offset, we have > to search around in the tree to find the key that was actually used to > insert the item. > > The insert_with_overflow code uses just one key, at the cost of having > to search around inside the item. Yeah this actually turns out to be a bit easier to code as well. I'm taking this approach. --Mark -- Mark Fasheh