Messaging patches

[Jeremy Allison]

On Tue, Apr 29, 2014 at 02:23:26PM -0700, Jeremy Allison wrote:
> 
> I think this patchset might have exposed
> an interesting bug within the code.
> 
> I *love* this stuff :-).

Oooh Oooh Oooh ! It's an ordering bug
in test_mux_open() exposed by the messaging
change (told you I *love* this stuff... :-).

test_mux_open does the following:

open file "open.dat" 		-> fnum1
open file "open.dat" again	-> fnum2

open file open.dat async... (async1)
open file open.dat async again (async2)

close fnum1
close fnum2

Then expect to get the following returns:

<---- (async1) (open OK fnum3).

wait for timeout...

<---- (async2) return SHARING VIOLATION.

However, with the messaging change
we return the two async messages in
the order async2 followed by async1,
*not* the order the test code is expecting !

This is perfectly allowable of course,
just not what the test code is expecting
(and I'd guess not what Windows does).

Of course we shouldn't end up hitting
an assert when the test tries to tear
down the test directory paths (which
is what is happening) but that's the
root cause of the test fail.

More as I work on it !

Jeremy.