On Jul 15, 2012, at 9:32 AM, Daniel Shahaf wrote:
> Daniel Shahaf wrote on Tue, Jul 10, 2012 at 09:46:55 +0100:
>> Pending rmuir@'s feedback, then, I'll go ahead and edit the revision
>> file in-place. No change should be needed to any of the svn mirrors.
>
> I have now made the following edit: s/modify-file/ add-file /
>
> svnsync has successfully synced r1356317. (As I write this, the
> post-commit fs processing is still running; 'current' on harmonia (the
> mirror) was updated at 13:24:29 UTC.) I do not plan to edit
> rep-cache.db to remove references to the rep of the file in question.
My mirrors are syncing fine now, thanks!
> I am not opening an svn bug yet as there is no evidence that the bug was
> in svn (as opposed to, say, a single bit flip in svn_fs_path_change_kind_t).
I'm inclined to agree. A bit flip is an entirely plausible explanation for
something this odd.
Somewhat related: is this a FreeBSD box? ports/sysutils/mcelog is useful for
getting info on any ECC errors that might have occurred. Is the repo living on
ZFS? Don't suppose you've got a non-standard vfs.zfs.txg.timeout (greater than
5 seconds?) set? That could have exacerbated the situation.
Trent.
