guberti opened a new pull request, #12530: URL: https://github.com/apache/tvm/pull/12530
#12207 went a long way to improving microTVM code reuse, but introduced an unintentional? change. Previously, the full `pytest` names of microTVM tests included the value of `board`, prepended to any paremeterization. For example, `test_rpc_large_array` would have the name: ``` tests.micro.zephyr.test_zephyr.test_rpc_large_array[nucleo_f746zg-(4*1024)] ``` This behavior was useful, as it meant that exported Junit XML files would correctly record which board the test was run on. However, after #12207 the names of tests stopped including this information: ``` tests.micro.zephyr.test_zephyr.test_rpc_large_array[(4*1024)] ``` This prevented Junit from differentiating between the tests when they were run on different boards, and instead just overwrote results on previous boards. Additionally, we never made this distinction for `common` tests that were parameterized by `platform`, so tests like `tests.micro.common.test_autotune.test_kws_autotune_workflow` would have one of the platforms overwrite the other (with whichever one finished last being the one that was ultimately reported). --- This PR uses the `pytest_generate_tests` hook to fix this issue. Now, the `board` parameter is included in the test name as before, and the `platform` parameter will be included for any tests that use that fixture (e.g. common tests). -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
