Re: TEST JJJ

[Paul Edwards]

"Pierre" <pceteaudatmacif.fr@netcourrier.com> wrote in message 
b86193eb.0310170632.6dc15f58@posting.google.com">news:b86193eb.0310170632.6dc15f58@posting.google.com...
> >As well as doing III, I've got a new test, JJJ, which I am
> >expecting to fail 0 out of 150.
>
> No.
> It fails (with a script):
> Fisrt 1 out 3 . (2 Ok, 1 fails, 2 Ok, 1 fails ... and so on) 28 times
> (28*3=84 tests)
> then 66 correct logins

Well this is certainly strange.  With a 2 second delay in server
and 5 second delay in client, then 3 * 2 > 5 would explain a
pattern, but I don't see any reason for the server delays to
accumulate in such a fashion.

Nor do I see any reason for it to then suddenly work as expected.

However, the good news is that to resolve this problem, IBM
only needs the ability to consistently reproduce the problem,
and that they can do with TEST III.  The fact that there are
potentially additional bugs will not interfere with the resolution
of this particular one.

I will start a new thread called "AIX shutdown bug" with the
next thing to do, which is report the reproducible problem
to IBM.  I am 99% sure it is a bug in either the AIX I/O
libraries (or libraries that come with your compiler), or with
the AIX operating system.  ie nothing to do with CVS.  Only
IBM can tell you which one it is, as the source code to the
fstat() function is required at this point.

Note that that patch that fails 150 times out of 150 times, is
an AIX problem.  On a properly behaving system, like Linux,
those sleep() commands cause no problem whatsoever, as
expected.  ie I am allowed to put in sleep commands, they
should be innocuous.  If the program can't handle sleeps, it
is a bug in either the program or the operating system/libraries.
In this case, it is the latter.

BFN.  Paul.