Re: [patch] concurrency fixes for the fuse daemon

[Hans Petter Jansson <hpj novell com>]

On Mon, 2008-04-21 at 09:58 -0400, David Zeuthen wrote:
> On Sun, 2008-04-20 at 19:16 -0500, Hans Petter Jansson wrote:

> > That leaves the question of how we detect when an open file gets
> > renamed. The shared table lets us redirect subsequent I/O on the open
> > file descriptor to the renamed file, emulating Unix semantics. There are
> > also special cases for unlink and ftruncate.
> > 
> > I'd rather get rid of that code if it's unnecessary, though. Does GVFS
> > handle all that for us internally?

> I think so. I mean, once you open a file with gio, you should only keep
> a reference to the GInputStream or GOutputStream and that should stay
> valid even if the underlying file changes (e.g. is renamed or unlinked).
> Of course this might be backend specific but I think backends are
> supposed to work that way too and if they don't work it's probably just
> a bug with that backend. So if this is true, which I think it may be, we
> should be able to get rid of that code. Thoughts?

It's definitely worth a try, but should be tested well with a range of
backends.

-- 
Hans Petter