Re: locking workaround patch

[Michael Meeks <michael ximian com>]

Hi Havoc,

On Thu, 2002-09-12 at 16:36, Havoc Pennington wrote:
> If you say it works, then OK. We should do it that way. 

	We should certainly do it using the same code, since any safe way also
has performance problems; it'd also be nice to propagate it from app to
forked app via command line arguments, to save re-scanning.

> I assumed that ORBit just stuffed the name of the tmpdir in the IOR
> or the like.

	Sure - but since we use the same directory for b-a-s locking /
location, and we'd have to create some sort of safe directory for ORBit2
(without producing reams of tmp dirs) I use the same code for ORBit2,
and b-a-s.

	As I say, there's still a potential race in there; we need a two phase
process:

	<scan>
	create foo-user-random-tst
	<scan>
	rename foo-user-random-tst > foo-user-random

	And then we have quite a well behaved algorithem, that's fully race
safe; currently we can be attacked in theory since we only do a single
scan.

	Anyway, I filed a bug explaining the issues here:

		http://bugzilla.gnome.org/show_bug.cgi?id=93554

	So we don't forget them,

	Regards,

		Michael.

-- 
 mmeeks gnu org  <><, Pseudo Engineer, itinerant idiot