I vote for the second one. We can hide the details in the test helpers. JunXu Chen <[email protected]> 于2020年10月16日周五 上午9:08写道：
> Hi, community, > > what do you think ? > > On Fri, Oct 16, 2020 at 9:05 AM JunXu Chen <[email protected]> wrote: > > > Recently, Vinci and I are refactoring the `manager api` for Apache APISIX > > dashboard, to remove the dependency on MySQL, and that will write data to > > ETCD directly. > > > > We have completed the main work. > > > > Yesterday I tried to use the test cases in the `t/admin` dir of Apache > > APISIX to test the refactored `manager api`. > > Many use cases failed because they tested the response status code and > > response body. > > The status code can be maintained the same, but the response body is > > difficult, especially the error message, because a lot of this > information > > comes from third-party libraries, such as `jsonschme`. > > > > I think it’s time to discuss how to develop and test between `manager > api` > > and APISIX > > > > The main issue is how to test to ensure that the data is correct between > > APISIX and `manager api`. > > Especially in the development process, when the data schema needs to be > > changed. > > > > There are currently two ideas. > > > > 1. Share test cases > > This method guarantees the correctness of the data to the greatest > extent, > > but it is more complicated (we need to make a lot of modifications to the > > old use cases, and the two projects need to pull each other's code for > > cooperation testing) and the two projects are relatively coupled. > > > > 2. Agreed on a schema and test separately > > This method is relatively simple, and the two projects can also be > > decoupled. But it is not so restrictive. > > >
