Hi Ralf:

In message <20171013060722.4ntvgkuw42hcpup2@runtux.com>,
Ralf Schlatterbeck writes:
>Ralf Schlatterbeck added the comment:
>On Fri, Oct 13, 2017 at 12:02:43AM +0000, John Rouillard wrote:
>> 
>> Ralf, I am not seeing this with a working checkout of
>> 5017c3422334 which is the tip.
>I have an application that has a "sign&send" button that calls a custom
>web action. This does a commit(). Then several auditors are called which
>should prevent the commit from going through. I will have to check the
>exact path where this happens, I think I'm doing some actions on other
>classes in the process.
>The effect is that on the next sign&send I'm getting an assertion
>violation because one change that shouldn't have made it through the
>Reject was made.
>
>I need to investigate this further, I've created the issue so that it
>doesn't get lost. I feared it would not be reproduceable with a standard
>setup...
>
>I've already investigated to the point I'm sure this is a change in the
>database (not just browser state). To get back to the expected state I
>had to change a flag in the DB using psql...

Hmm, what back end DB and what isolation level?