RFR(T): 8214857: "bad trailing membar" assert failure at memnode.cpp:3220
tobias.hartmann at oracle.com
Thu Dec 6 09:33:27 UTC 2018
On 06.12.18 10:16, Andrew Dinn wrote:
> Which suggest to me that any extra effort we make as a response to this
> goes into the early adoption program. There are clearly going to be lots
> of users out there who will find this sort of issue faster and more
> efficiently than would be achieved by throwing extra jars into the
> current tests.
While I agree that we should offer EA builds as early as possible, it's still a fact that the cost
of fixing a bug is much higher, the wider it spread and the later we catch it (especially for recent
regressions). Just considering the overhead of communication, limited reproducibility due to
restricted access to customer machines/applications and the risk that the issue is not even reported
due to it's intermittent nature (especially relevant for compiler issues). And I'm not even
considering that releasing a buggy prototype for EA makes a bad impression.
So I don't think we should justify less testing by putting that burden on the early adopters, but I
would see this as an orthogonal approach to improving the overall quality of the product.
More information about the hotspot-compiler-dev