This repository hosts code that supports the testing infrastructure for the main PyTorch repo. For example, this repo hosts the logic to track disabled tests and slow tests, as well as our continuation integration jobs HUD/dashboard.
The mechanism to disable a test is sometimes "abused" to keep a test disable for a long time without any owner to work on fixing it. Although, this helps mitigate an flaky test well, it has several drawback:
It makes thing harder to keep track, a test that is disabled directly in the code is much easier to see
Ideally, we want to keep the current mechanism but limit it only to recently flaky tests, i.e. within 2 weeks. Anything older than that and has not yet been addressed should be disabled directly in the code. Doing this manually doesn't scale, so we need a better tool for this like a codemod service of sort.
The mechanism to disable a test is sometimes "abused" to keep a test disable for a long time without any owner to work on fixing it. Although, this helps mitigate an flaky test well, it has several drawback:
Ideally, we want to keep the current mechanism but limit it only to recently flaky tests, i.e. within 2 weeks. Anything older than that and has not yet been addressed should be disabled directly in the code. Doing this manually doesn't scale, so we need a better tool for this like a codemod service of sort.
cc @clee2000 @malfet @kit1980