http://www.slony.info/bugzilla/show_bug.cgi?id=67





--- Comment #4 from Adam Buraczewski <aburacze at gmail.com>  2008-12-15 16:20:42 ---
Hi Christopher,

Thanks a lot for taking the problem up.

(In reply to comment #3)
> The above test *tries* to exercise the bug, by repeatedly adding and dropping
> columns on table5.
> 
> It doesn't succeed at exposing the wrong behaviour.

In the meantime, could you please have a look at the bug #18 in the Bugzilla?
It looks the patch attached there tried already to cope with the problem, but
somehow has not been applied, although the idea of trimming the attkind string
was accepted and is used in Slony-I 2.0.0. This patch looks almost exactly as
the one I attached to bug #67 (I swear I have not seen bug #18 before ;)

> Adam, can you suggest anything else about your pattern of alterations that
> would induce the problem?

Well, I will think about it tomorrow (it's 1am in my time zone). The problem is
that to expose this bug there should be some "garbage" containing letter 'k'
immediately after the byte 0 ending the string. This would cause to check if
the column is null and raise an error if it is ("old key column is NULL on
UPDATE/DELETE") -- in our test environment it happened randomly. I am thinking
about modifying the pg_trigger.tgargs column to achieve this, but I am afraid
it will not work. As I said before, I will check it tomorrow.

> I will be more than pleased to fix the bug, but I'd be *really* happy if we
> have a test that can reliably induce the condition that exposes the problem, as
> that means we can be *really* confident that it is dealt with.

I second the idea.

Regards,
Adam Buraczewski


-- 
Configure bugmail: http://www.slony.info/bugzilla/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.
You are the assignee for the bug.