> Thomas,
>
> You're interpretation of the manual is correct. This is a problem in the
> code, documented in APAR IC35203, with fixes targeted in for 5.1.6 and
> 5.2.0.
>
> If you are running a client somewhere between 5.1.5.0 and 5.1.5.5, then
> try the latest patch levels, as they should include the fix. This was
> originally fixed in Windows patch level 5.1.5.6, and since patches are
> cumulative (even for other platforms), it should be fixed in the latest
> patch levels -- even if the README file doesn't mention them.

We are probably going to be stuck with this bug for some time to come.
We have already invested a significant effort in getting to 5.1.5.0,
including not only software upgrades as such but also testing and
change tracking. I don't think I am going to be able to sell the idea
of redoing all of that work because of IBM's ineptitude.

The documentation quoted in my original posting provides something just
short of a ready-made definition of a test case. The same documentation
provides completely clear criteria for recognizing the actual behavior
as being incorrect. As far as I can tell, the incorrect behavior is
completely reproducable. I don't understand how any sort of serious
effort at product testing could miss such an easy target.