[darcs-devel] [issue904] record fails on sshfs (sshfs has no atomic_create)

[Trent Buck]

Trent Buck <trentbuck at gmail.com> added the comment:

The strict requirements of roundup's Subject header field parser are annoying.

----- Forwarded message from Miklos Szeredi <miklos at szeredi.hu> -----

To: trentbuck at gmail.com
CC: fuse-sshfs at lists.sourceforge.net
Subject: Re: [sshfs] [issue904] record fails on sshfs (sshfs has no
	atomic_create)
From: Miklos Szeredi <miklos at szeredi.hu>
Date: Sun, 25 Jan 2009 20:32:00 +0100

On Sun, 25 Jan 2009, trentbuck at gmail.com (Trent W. Buck wrote:
> Nathaniel W Filardo <nwf at cs.jhu.edu> writes:
> >>> open(O_CREAT| O_EXCL) is inherently an atomic operation [...]
> >>> AFAICT, sshfs understands open(O_CREAT|O_EXCL) just fine and carries
> >>> the O_EXCL bit across the SFTP transport faithfully[...]
> 
> Would a developer here please confirm the above?

It has been true for a while now (since about 2006).  The requrements
for atomic create+open are:

  linux kernel >= 2.6.15
  libfuse >= 2.5
  sshfs >= 1.3

> 
> Such confirmation is necessary before Darcs developers will remove a
> paranoia setting that prevents Darcs from working over sshfs:
> 
> > Hello fuse-sshfs.  I'm attempting to get a patch included into
> > darcs to make it fall back to open(O_CREAT|O_EXCL) locking
> > mechanisms when link() returns ENOSYS as it does on sshfs.  While
> > it certainly appears that this operation is safe, the darcs
> > developers would like confirmation that this is the case.
> 
> It seems Nathaniel didn't receive a reply.  Perhaps this list is for
> users, and the developer list is elsewhere?

It seems the previous posting didn't make it to the list.  It may have
been discarded accidentally while moderating for spam.

Thanks,
Miklos

----- End forwarded message -----

----------
nosy: +dmitry.kurochkin, simon, thorkilnaur

__________________________________
Darcs bug tracker <bugs at darcs.net>
<http://bugs.darcs.net/issue904>
__________________________________