RFR: 8323529: Relativize test image dependencies in microbenchmarks

[Erik Joelsson]

On Thu, 11 Jan 2024 14:17:39 GMT, Erik Joelsson <erikj at openjdk.org> wrote:

>> In principle, I think tests should be executed from their corresponding test-support directory. (I think there is some JBS issue for this that's been around for some while.) The rationale is that this is a good directory for any arbitrary output generated from the test, for files that are just dumped in the current directory. (This happens from time to time with tests.)
>> 
>> But if executing from the test image improves the code quality of the tests, I am not going to argue heavily against it.
>
>> In principle, I think tests should be executed from their corresponding test-support directory. (I think there is some JBS issue for this that's been around for some while.) The rationale is that this is a good directory for any arbitrary output generated from the test, for files that are just dumped in the current directory. (This happens from time to time with tests.)
> 
> Magnus does bring up a good point. The root of the test image is a very bad place to accidentally drop arbitrary files when run in our distributed test system, but we don't really run the micros there, at least not like this, do we?

> If the test image dir could be relative to the test-support directory (via a symlink, or knowing that `../test/` always points to the test image) then running from out of there could work. As @erikj79 has figured out then the root issue for me here is that it's not feasible to pass the test image directory as a parameter (since the `@Fork` annotation needs compile-time constants), so if it's possible to configure test-support directories to be in arbitrary places then that would become very fragile.
> 
> None of our JMH tests pollutes the current directory as far as I'm aware, though. And if they did the current situation where the micros are run out of the make directory might arguably be worse.

Fair enough. This isn't worse than the current CWD. The test image dir and the support output dir have no guaranteed relationship relative to each other, so we can't rely any relative path between them.

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

PR Comment: https://git.openjdk.org/jdk/pull/17349#issuecomment-1887672959