-
Notifications
You must be signed in to change notification settings - Fork 672
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Add option to throw when passing expect
a normal Promise
#2207
Comments
Hi @andrewbranch, Yes, this behavior can cause a false positive result. I like the first approach. I think we could throw a warning for await t
.expect(getResult())
.ok({ allowUnawaitedPromises: true //false by default }) @AndreyBelym @AlexanderMoskovkin what do you think? |
Hi folks, any thoughts on this? Should I get started, or wait for @AndreyBelym and @AlexanderMoskovkin to weigh in? Thanks for your time! |
@kirovboris when a warning is thrown, how does it affect the running test? Regarding your suggestion to add a flag ( |
@adrianaisemberg t.expect(callAsyncFn()).eql(smth); i.e. compare |
Hi @andrewbranch, I feel your proposal is awesome! Let's do it =)
I vote for it. And I think it's actual not only for the I'm not sure about your second proposal. If you do it the TypeScript compiler will throw an error case regardless whether do you use the
@adrianaisemberg, yep, @kirovboris means throw an error, not a warning.
This option is too specific and as @andrewbranch mentioned above |
Great! I’ll get started! I feel like I have a good lead on where to begin the implementation, but I'm having trouble navigating the test structure. @AlexanderMoskovkin, where is the right place to test this new feature? |
I guess it should be functional tests, you can add them to the existent assertions tests. To see how it works take a look at the test-functional-local gulp task. It runs mocha tests that run testcafe tests inside =) Each functional test works in the following way: mocha triggers a testcafe test, testcafe runs the test and returns some report, mocha checks the report. Additionally it should be one more unit test that will check the formatting of the new created error (see here). |
One more question: how would you detect Promises that aren’t ClientFunctionResultPromises? I noticed that ClientFunctionResultPromise is implemented with the polyfill Pinkie, so |
Are you saying I should make a new subclass of error for this? What should the base class be? Do you have ideas of what info should be included alongside the error message? |
I guess it's ok.
In you PR you implement exactly what do I mean |
This thread has been automatically locked since it is closed and there has not been any recent activity. Please open a new issue for related bugs or feature requests. We recommend you ask TestCafe API, usage and configuration inquiries on StackOverflow. |
Are you requesting a feature or reporting a bug?
Feature request, sort of. Also, I’m willing to do the work myself.
What is the current behavior?
Internally, TestCafe uses a class called ClientFunctionResultPromise that extends Promise to allow ClientFunctions to be retried in the Smart Assertion Query Mechanism. Most properties on Selector instances return these. It's super useful to pass these to
t.expect
so they get retried until they pass, or a timeout passes, e.g.However, this can also lead to confusion, because if you pass a normal Promise to
t.expect()
, it will not be awaited or retried; it will use the Promise object itself as the value being asserted. And it’s very easy to fall into this trap, because chaining a.then
on a ClientFunctionResultPromise returns a normal Promise:This looks fairly similar, but the assertion always passes, even if the thing is never visible, because the Promise object is truthy. So the big problem is that the mistake of passing a normal Promise to
t.expect()
is really harmful with.ok()
. For any other assertion, you’re likely to catch your mistake, but with.ok()
, it’s very easy to miswrite an assertion that passes when it shouldn’t.To make matters worse for those of us who are using TypeScript, the type definitions for TestCafe don’t expose ClientFunctionResultPromise; instead, it types every ClientFunctionResultPromise as a normal Promise. So these two similar-looking examples above actually look identical to the type system, making this mistake impossible to lint for or guard against with static typings.
What is the expected behavior?
Two proposals:
It seems like passing a normal Promise to
t.expect()
is probably a mistake >99% of the time. Can we somehow add a configuration option that would causet.expect(normalPromise)
to throw an error? If that’s too presumptive, what about makingt.expect(normalPromise).ok()
throw an error? It could perhaps be bypassed with something like.ok({ allowUnawaitedPromises: true })
?If the type definitions exposed
ClientFunctionResultPromise
, we could at least lint for this, or write a type-safe wrapper fort.expect
or something. Only catch with this is that because TypeScript uses structural typing rather than nominal typing,ClientFunctionResultPromise
would look identical toPromise
unless the added methods were made public on the type, and they look like they’re meant to be private. There are some possible hacks around this, but they’d still surface awkwardly in the type system. I think my first proposal is better.Would you accept a PR for either of these ideas? If the former sounds reasonable, what are your thoughts on an API for it?
The text was updated successfully, but these errors were encountered: