We could define a selector using `JQuery` [type selectors](https://api.jquery.com/category/selectors/) with a string like `'.gf-form-input.login-form-input'` but that would be brittle as style changes occur frequently. Furthermore there is nothing that signals to future developers that this input is part of an e2e test.
At Grafana, we use `aria-label` as our preferred way of defining selectors instead of `data-*` attributes. This also aids in accessibility.
Let's add a descriptive `aria-label` to our simple example.
Now that we added the `aria-label` we suddenly get more information about this particular field. It's an input field that represents a username, but there it's still not really signaling that it's part of an e2e test.
The next step is to create a `Page` representation in our e2e test framework to glue the test with the real implementation using the `pageFactory` function. For that function we can supply a `url` and `selectors` like in the example below:
The next step is to add the `Login` page to the exported const `Pages` in `packages/grafana-e2e/src/pages/index.ts` so that it appears when we type `e2e.pages` in our IDE.
Now that we have a `Page` called `Login` in our `Pages` const we can use that to add a selector in our html like shown below and now this really signals to future developers that it is part of an e2e test.
- The `url` property is used whenever we call the `visit` function and is equivalent to the Cypress function [cy.visit()](https://docs.cypress.io/api/commands/visit.html#Syntax).
> Best practice after calling `visit` is to always call `should` on a selector to prevent flaky tests when you try to access an element that isn't ready. For more information, refer to [Commands vs. assertions](https://docs.cypress.io/guides/core-concepts/retry-ability.html#Commands-vs-assertions).
- Any defined selector in the `selectors` property can be accessed from the `Login` page by invoking it. This is equivalent to the result of the Cypress function [cy.get(...)](https://docs.cypress.io/api/commands/get.html#Syntax).
Let's take a look at an example that uses the same `selector` for multiple items in a list for instance. In this example app we have a list of data sources that we want to click on during an e2e test.
You might have noticed that instead of a simple `string` as the `selector`, we're using a `function` that takes a string parameter as an argument and returns a formatted string using the argument.
Just as before we need to add the `DataSources` page to the exported const `Pages` in `packages/grafana-e2e/src/pages/index.ts`.
The next step is to use the `dataSources` selector function as in our example below:
Now we can write our test. The one thing that differs from the `Basic example` is that we pass in which data source we want to click on as an argument to the selector function:
> Best practice after calling `visit` is to always call `should` on a selector to prevent flaky tests when you try to access an element that isn't ready. For more information, refer to [Commands vs. assertions](https://docs.cypress.io/guides/core-concepts/retry-ability.html#Commands-vs-assertions).
The most common error with PhantomJs image rendering is when a PR introduces an import that has functionality that's not supported by PhantomJs. To quickly identify which new import causes this you can use a tool like `es-check`.
There is no easy or comprehensive way to debug PhantomJS smoke test (image rendering) failures. However, PhantomJS exposes remote debugging interface which can give you a sense of what is going wrong in the smoke test. Before performing the steps described below make sure your local Grafana instance is running:
