> Do you have to be running a process to use these? The nice thing
> about FSEvents is that (IIUC) you can just subscribe the the
> directory, quit your process, and then comeback later and ask it what
> changed. It might take some selling to convince the devs and the
> users that a daemon is needed for Subversion.
Yes, as far as I can tell you must run a process.
The need to have a daemon (or service, to use Windows terminology) is
exactly why I suggested that this be optional. I expect most users
would want it installed for the better performance, but some might not,
or might want to toggle it on and off.
Re: FSEvents, it's tough for me to read
SEvents/index.html but it does seem as though you're correct about being
able to come back later since the change events are stored persistently.
(Based on my reading, I politely if a little naively disagree with Bob
Archer's subsequent response on this thread.)
But personally, I'd be worried if this went on too long (say the user
adds, rewrites, and deletes files like crazy without touching svn--does
the persistent change information fill the filesystem?). I would rather
have a daemon consuming the FSEvents data whenever possible, or not use
it at all.
One other thing: The wiki at http://en.wikipedia.org/wiki/FSEvents
compared FSEvents to the Linux inotify at
http://en.wikipedia.org/wiki/Inotify. From what I can tell, inotify
operates more similarly to how Windows does--no persistence that I can
No idea about other OS's, but since this functionality isn't something
you can count on everywhere, obviously svn must work without it.
However, with it, a lot of client operations could see tremendous
performance boosts. I even saw a link
(http://markmail.org/message/vmigwu3wfxsy7r7z) where they talked about
FSEvents for Mercurial, so I'm glad to hear that the svn developers are
considering this capability.
To unsubscribe from this discussion, e-mail: [users-unsubscribe_at_subversion.tigris.org].
Received on 2009-03-04 19:08:30 CET