Revisit implementation of `accessibilityRole="button"`

[necolas]

Is your feature request related to a problem? Please describe.

Revisit mapping accessibilityRole="button" to button element.

The current approach uses ARIA roles to achieve similar results due to historical problems in Firefox with flexbox children of a button. But this was fixed in Firefox 63 and the current ESR version is 78, so we may be able to migrate away from this special case and map "button" to an element like we do for several other roles. (FB web still supports FF versions prior to 63 but no one performs visual tests that far back so I don't think we need to be too concerned about it.)

This shouldn't produce any savings in bytes since the code used to implement ARIA-backed buttons is also used by the unstable createElement API and for other ARIA roles that should have similar behavior to buttons (i.e., respond to user interaction).

[doumart]

I was gonna file the same issue. I use react testing library and i use accessibility to test the content of my page. To have a real button tag removes a lot of friction and is more near the reality of pure web development.

[intergalacticspacehighway]

@necolas What should be the case for <View accessibilityRole="button" />?
Should this renderd as <div role="button" /> or <button />.
Got this as a failing test case while implementing the above change.

[georgii-ivanov]

@intergalacticspacehighway what's the problem with <View accessibilityRole="button" /> cuz at this point you'll get button tag and you can put another View/Text element into it as HTML allows you to put something else besides text node.

I think that would be a useful improvement to HTML semantics. Also probably useful for SEO (not sure about that tho)

[jgornick]

We ran into this issue because our a11y rules require semantic HTML rather than using a11y attributes (e.g. aria-*). In our case we needed to produce a <button> when using the <Button> component. However, we learned that once that goes through <TouchableOpacity> to <View> to createElement and finally propsToAccessibilityComponent, we noticed that it wasn't mapping a role of button to a <button> a11y component.

So, we started with @intergalacticspacehighway's PR #1944 implementation, but because not everyone shares our same a11y rules and may not want to map roles to semantic HTML, we short-circuited the <Button> component rendering to use createElement('button', ...). That way others could still use a <View> (<div>) with an a11y role of button.

It brought up a question of whether react-native-web should be prescriptive in converting a11y roles to semantic HTML? There is great value in using semantic HTML, so I understand why it does today. However, that might not fit everyone's need. It would be interesting to see if there is a way for components to opt-in/out of the conversion from a11y role to semantic HTML.

After all this being said and based on current react-native-web behavior to map a11y roles to semantic HTML, I think @intergalacticspacehighway's PR #1944 does what's needed to support our direct need to mapping a11y role button to <button>.

[necolas]

Your accessibility rule requiring a semantic tag doesn't surface a functional problem for users.

[jgornick]

Your accessibility rule requiring a semantic tag doesn't surface a functional problem for users.

Great point.

With that said, then does it make sense to map the button a11y role to the button tag? If so, does #1944 address that change?

Thanks @necolas!

[antoniojps]

What's stopping #1944 from being merged?

According to mozilla best practises regarding this issue.

Where possible, it is recommended to use native HTML buttons (, , , and ) rather than the button role, as native HTML buttons are supported by all user agents and assistive technology and provide keyboard and focus requirements by default, without need for additional customization.

[p4bl1t0]

Also this is stopping the availibity of using features like this: https://developer.mozilla.org/en-US/docs/Web/API/Navigator/share

Because they need "Must be triggered some kind of "user activation""

[necolas]

Also this is stopping the availibity of using features like this: https://developer.mozilla.org/en-US/docs/Web/API/Navigator/share

No it isn't.