RFR: 8373654: Test sources/TestNoNull should only run once

[David Holmes]

On Mon, 15 Dec 2025 06:52:49 GMT, David Holmes <dholmes at openjdk.org> wrote:

> This test is checking for backsliding on the use of NULL in the hotspot source files. The test only needs to be run once per change but it is defined as a regular tier1 test which then gets run for each os-arch combination, plus product and debug, and then again with a bunch of flags in tier3. This is just a waste of resources. At a minimum we should make the test flagless so that it doesn't run in tier3, but we can also restrict it to a single platform.
> 
> Before the change the test ran 27 times in tier1-3 in our CI. After it runs once. 
> 
> Thanks

Thanks for taking a look @shipilev !

> linux and amd64 -- as someone who re-runs this test on my M1 Mac to reproduce CI failures, I disagree!

Reality is that for every changeset integrated only one entity in the world actually needs to run this test. But I can drop the platform restriction - 4 runs is still better than 27.

> You might want to do this for other sources/ tests?

I'll take a look at what they do. I only tripped over this one when a stray file caused it to fail.

-------------

PR Comment: https://git.openjdk.org/jdk/pull/28818#issuecomment-3657554161