Fix some Cognitive Language test errors

[heaths]

Fixes errors introduced in #35228

[heaths]

/azp run net - cognitivelanguage - tests

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).

[heaths]

@annelo-msft @KrzysztofCwalina this is a good example of what not having models might lead to. In part, though, it's because we don't validate inputs for our test framework and I had a typo I copied/pasted in several tests. The other issue was that an object was nested one level too deep - all things that models over anonymous types or dynamic types would've prevented.

[annelo-msft]

Thanks, @heaths, it's a good call-out, and I think a known risk of hand-authoring JSON (i.e. not using compiled model types) using any method.

I'm curious, did these tests pass before? How did you discover the issue?

[heaths]

Thanks, @heaths, it's a good call-out, and I think a known risk of hand-authoring JSON (i.e. not using compiled model types) using any method.

I'm curious, did these tests pass before? How did you discover the issue?

The recorded tests pass because we don't validate requests - only responses. The live tests failed because I missed an "e" on the end of projectName and misplaced the task object one level too deep on another request.

[annelo-msft]

The recorded tests pass because we don't validate requests - only responses. The live tests failed because I missed an "e" on the end of projectName and misplaced the task object one level too deep on another request.

That's good feedback. We may want to recommend in our docs that end-to-end testing is a best practice to avoid this kind of error. I'll make a note on the docs PR.

[azure-sdk]

API change check

API changes are not detected in this pull request.

[heaths]

That's good feedback. We may want to recommend in our docs that end-to-end testing is a best practice to avoid this kind of error. I'll make a note on the docs PR.

We don't currently recommend they have end-to-end testing? Or do you mean that they run live tests prior to submitting initial tests? To record they would've had to. This is a case where we don't check the request like we do the responses. Perhaps we could by recording our own byte count for a quick test and drill into body matching for requests. /cc @JoshLove-msft @christothes

[JoshLove-msft]

Perhaps we could by recording our own byte count for a quick test and drill into body matching for requests. /cc @JoshLove-msft @christothes

Can you clarify why we are not matching requests here? Is it because it is disabled for this specific library? Request body matching is on by default.

[JoshLove-msft]

Perhaps we could by recording our own byte count for a quick test and drill into body matching for requests. /cc @JoshLove-msft @christothes

Can you clarify why we are not matching requests here? Is it because it is disabled for this specific library? Request body matching is on by default.

Ah we disabled body comparisons because of the old "doubles mismatch on different runtimes" issue. Here is the recommended workaround - we should probably add this to the Test Framework docs. If only someone had an issue assigned to them instructing that they do this 😆 (that would be me)