Closed aballier closed 5 years ago
Yes, we don't have a regression test suite for the moment. It would be nice though, and is tracked here: https://github.com/rdicosmo/parmap/issues/39 If you can show there is a bug on 32bits with some simple reproducible method, please open another issue.
While tests work really fine on amd64, I'm trying to run them on arm (I've also seen similar reports for x86), and they fail:
This one looks like a bug in ocaml itself:
Reducing nData even more makes this one pass:
This one seems to work but is quite heavy (1h 20 mins to run):
And this one fails:
If I reduce a bit the data used by changing the line to:
(remove one zero)
then it works but is also rather long (3 hours):
All the other tests not mentioned here pass properly.
Now, my question is: Is it possible to reduce a bit the data used by the tests like that? Do they remain meaningful ?
My goal by running the tests is to ensure that they provide a good coverage of parmap, ensuring that if they pass that means parmap will work properly. It seems to me that those tests are more about getting accurate speedup results. Can I skip some of those and keep the same level of safety wrt my goal ? If yes, which ones ?