On 8/25/2013 1:18 AM, Michael L. Semon wrote:
On 08/22/2013 02:19 PM, Ben Myers wrote:
Gents,
On Thu, Aug 22, 2013 at 11:14:56AM -0500, Geoffrey Wehrman wrote:
On Thu, Aug 22, 2013 at 12:02:26PM +1000, Dave Chinner wrote:
| I'm very, very, very unhappy about how this situation is unfolding.
As am I.
Mark provided some mkfs code to test the v4 feature bit with, and it
worked fine for me.
Given that we are protected by a feature bit, I feel that pulling in the
v4 feature is considerably less risky than what we did in 3.10, with
Dave still cleaning up his mess in -rc6, so go ahead and call me
reckless: I've pulled in both v4 and v5 versions of this code.
Mark, please post your mkfs code ASAP, even though Dave hasn't reposted
his userspace series yet.
Everybody gets his code in and nobody is happy.
-Ben
Mark's v4 dirent patches seem to work on 32-bit x86. I happen to agree
100% with Dave on this issue. However, lacking a dirent test and
xfs_db skills, I threw everything else and the kitchen sink at
v4-dirent XFS and did not find any evidence to back up Dave's argument.
So I'll tip my cap to Mark for his insight on the matter, hoping that
his testing skills are fine as always.
Thanks!
Michael
BTW, are there any xfstests results times for dirent vs. non-dirent
code? For whatever reason, generic/068 is posting noticeably worse
numbers for v4-dirent XFS than for both v4 XFS and v5-dirent CRC XFS.
I'm running the tests myself as well, but it will take a while to run
them for v4 XFS, v4-dirent XFS, v4-dirent-512b-inode XFS, and
v5-dirent-CRC XFS...which kinda speaks to Dave's point about supporting
a matrix of filesystems, but I wasn't meaning it that way...
Thanks again!
Michael
_______________________________________________
xfs mailing list
xfs@xxxxxxxxxxx
http://oss.sgi.com/mailman/listinfo/xfs
