Full Network Layer Stubbing / fetch support

brian-mann commented 6 years ago

Pull Request (in progress): https://github.com/cypress-io/cypress/pull/4176

What users want

To be able to stub any kind of HTTP request including window.fetch
- note: temporary polyfill available https://github.com/cypress-io/cypress/issues/7710 v4.9.0
To blacklist certain domains (google analytics, new relic, intercom)
To have more control over dynamically routing / stubbing requests and responses
To more easily stub graphQL calls

What we do now

We modify the host XMLHttpRequest object which limits us to only stubbing XHR's from the main application frame.

What we need to do

Throw away the entire existing system for stubbing
Move stubbing into the internal Cypress network proxy layer (outside of the browser)
Synchronize the state of the browser tests with the backend (so its aware what needs to be stubbed / routed)
Remove cy.server and instead just use cy.route
Expand the API's for more advanced configuration
Accept blacklistDomains option for cypress.json
As each request comes in, if we expose callback functions we'll need to constantly communicate with the driver for guidance on what to do - whether to proceed - halt, etc. This can get pretty complicated and might have significant performance issues. We'll need to experiment ~~and we may not be able to provide callback functions anymore.~~

Things to consider

onRequest and onResponse currently provide the raw level xhr object. This would no longer be possible. We'd have to serialize the request / response into basic objects. No method calling would be available.
The onError callback could no longer exist
Additionally we hook into the onerror events for XHR's and will automatically fail the test if this is called. This would no longer be possible.
Blacklisting https cannot be dynamic. It would have to be set in cypress.json or environment variables. The reason is that for each blacklisted domain that goes over https, we would have to route that traffic to a self signed certificate. Browser freak out if traffic to https gets assigned one certificate, and then within the same session it gets another cert applied. What this means is that before spawning the browser we have to isolate that traffic and know ahead of time not to route it to the actual server. This enables us to send a single cert to the browser.
We could enable non https hosts to be dynamically blacklisted, but these API's would be async and likely difficult for users to understand / use. We might be able to bake these into cy.route but I'm not sure there's much of a use case here of only dynamically blacklisting non https traffic.
We will no longer be able to tell the difference between an XHR and any other kind of http request. We'd likely have to change the page events to say REQ and REQ STUB instead of XHR and XHR STUB.
It would be very difficult to provide accurate onResponse callback functions. It's possible for the internal server to know when an HTTP request has been completed, but it's not possible to know whether the browser has actually finished processing it. This may or may not be a problem.
How would we handle stubbing binary / streaming responses?
How would we handle stubbing things like images? (This is likely pretty simple)

Bonus Points

It would likely be possible to stub Websocket traffic in the same manner
The problem with this is that most Websocket libraries (ie socket.io) add their own "encoding" pattern on top of each websocket message frame. The ergonomics around stubbing this would be odd. The problem is this is library (and even version of library) specific. This may be a better use case for cy.stub, or perhaps a 3rd party extension

brian-mann commented 6 years ago

Related to:

ericadamski commented 6 years ago

@brian-mann I would be very willing to help out with this issue. This is a big blocker for use when testing our app.

brian-mann commented 6 years ago

@ericadamski could you go into more detail about this? What exactly is a big blocker?

ericadamski commented 6 years ago

@brian-mann making network requests (using a fetch polyfill) and asserting on those requests is important to our integration testing. Currently we are just spying on the window.fetch and doing some simple assertions on the params request. It would be ideal to provide the same functionality as XHRs to completely omit the server and mock responses when needed, understanding that we could stub the fetch as well, but this means stubbing each individual request with a tailored response.

brian-mann commented 6 years ago

You can do this right now using stubs. Albeit it's certainly not as nice as what we did for XHR stubbing, but it is possible and we have several examples of making this work. This is even in a new getting started video we're releasing later this week.

https://github.com/cypress-io/cypress-example-recipes/blob/master/cypress/integration/spy_stub_clock_spec.js

ericadamski commented 6 years ago

Yes we realize it can be done, albeit we would rather wait (or work toward) a more XHR style stubbing, it is just super nice.

brian-mann commented 6 years ago

Yeah it is. Okay just making sure we're on the same page.

The proposal here does a decent job outlining what needs to be done but it will be a significant amount of work.

Likely the next steps just need to be formalizing a plan for the API's themselves. There have been many requests for additional features and we should start there.

ericadamski commented 6 years ago

That sounds good.

Is it worth creating separate issues for some of the requested features for I know almost nothing about graphQL queries 😜

brian-mann commented 6 years ago

They mostly already are other issues. We need to factor in all of them and come up with a set of API's that unifies those different use cases and desired experiences.

ericadamski commented 6 years ago

Right OK, I see what you are saying.

ericadamski commented 6 years ago

@brian-mann side note, has it been thought of moving away from coffeescript and into es6?

paulpruteanu commented 6 years ago

Do you consider this feature in the next release? I reckon it would be highly valuable, for instance when

authentication/ other actions are performed at script level ( Githubissues.
Githubissues is a development platform for aggregating issues.

cypress-io / cypress