Closed altheaden closed 1 year ago
I tested this by running the PR suite (which includes the inertial gravity wave test) against a baseline. Also, this fix allows the test to pass in my task parallel branch, so the dependencies seem to be working as expected.
@xylar I made the change you suggested and ran the tests again, this time making sure to run the viz step, and everything passed against a baseline.
Then, I ran the same change in the task parallel branch, and that went smoothly, except that it took a much longer time to run than usual. I don't have any reason to suspect that it's an issue with this change, rather than an issue with that branch and/or perlmutter (which sometimes executes very slowly).
If you think that we can assume that that behavior is likely to be isolated to issues in the parallel branch, then I think it's safe to say that this PR is good to go. That is my thought, but let me know what you think.
@altheaden, I agree with you that the slow execution is a separate issue, probably to do with running in task parallel, that we will address in a separate PR. This fix looks good and your testing sounds fine to me.
The output files for the init step aren't being added to the step's outputs attribute, which is causing issues when running in task parallel.
Checklist
Testing
comment in the PR documents testing used to verify the changes