Re: Issue with markers in isolation tester? Or not?

Michael Paquier <michael(at)paquier(dot)xyz> writes:
> On Tue, Jan 14, 2025 at 11:48:28AM -0800, Noah Misch wrote:
>> I misunderstood, and I was mistaken to see this as a bug fix. The
>> isolationtester is acting per its definition, and this would be a definition
>> change. Do others have opinions on the merits of today's definition vs. the
>> proposed definition?

> I agree that Michail's case with the handling of the markers is kind
> of strange in the case he has reported. Anyway, do you think that it
> would be a good idea to change that knowing for how long we've relied
> on isolationtester to do things the way they are?

We've changed it before, so I'm not totally averse to considering
more change. But I'm not happy about the lack of any analysis in
this thread of the impact on existing test cases. In particular,
it seems to me that this proposal amounts to reducing the number
of states that can be reached during an isolation test. So it
seems to me that means strictly less coverage than we had before.
I'd prefer to start by asking why the proposed new test can't be
made stable without such a restriction.

(I don't mean to minimize the question of whether we need better
documentation about this behavior.)

regards, tom lane