You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
They must be executed serially (they cannot be executed in parallel). This is problematic because it makes tests take a long time, and are exceedingly brittle. If a programmer makes a change in one module, they may spend a long time fixing tests to ensure that they all pass given that a failure can take 20+ minutes to appear. Similarly, they are quite brittle, as creation tests change the read tests.
They only assess the minimum of functionality that BHIMA offers. We do not really test balances - either of $$ or stock - because if a developer changes or adds a test upstream, they will be fixing tests forever downstream.
They do not represent a live system due to how few records are in our test database. In the real hospital, we would have thousands of transactions. The tests therefore do not reflect the reality they are supposed to be testing and we still have to do manual follow-ups to find performance regressions.
In light of all this, I propose that we rewrite our tests to be based on a data snapshot we pull from a hospital that has been using BHIMA long term. We would anonymize personal information (patient names, personalized account names, employees, etc), and take a two-year snapshot that we can manually massage to ensure it doesn't have abherrant data (like negative stock balances). Then we could transition all our tests do perform the read tests first, and the create tests on a subsequent fiscal year.
I think this would be enough to give us more confidence in our tests and ensure that we catch when our data structures change.
Note that this would need to happen after #7024 is merged for best results.
The text was updated successfully, but these errors were encountered:
Currently, our tests have three big limitations:
In light of all this, I propose that we rewrite our tests to be based on a data snapshot we pull from a hospital that has been using BHIMA long term. We would anonymize personal information (patient names, personalized account names, employees, etc), and take a two-year snapshot that we can manually massage to ensure it doesn't have abherrant data (like negative stock balances). Then we could transition all our tests do perform the read tests first, and the create tests on a subsequent fiscal year.
I think this would be enough to give us more confidence in our tests and ensure that we catch when our data structures change.
Note that this would need to happen after #7024 is merged for best results.
The text was updated successfully, but these errors were encountered: