!!!FEATURE: Centralize error handling

[grebaldi]

refs: #3119
prelude to: #3773

The @neos-project/neos-ui-error package had been introduced in #3759. During the discussion of that PR, @mhsdesign suggested it would be a good idea to centralize other error-handling mechanisms in that package (namely the high-level ErrorBoundary, see: #3759 (comment)).

As I was working on #3773, I realized that I would need to decouple FlashMessages from the redux store and expose a different mechanism to use them. So I decided to take the centralization step first.

Bottom-line, this PR moves the following concepts into @neos-project/neos-ui-error:

• The high-level <ErrorBoundary/>-component (introduced in BUGFIX: Add primitive Root ErrorBoundary around whole UI #3485)
  • Catches errors that may occur during React rendering (anywhere in the component tree)
• The terminateDueToFatalInitializationError-function (introduced in !!!FEATURE: Reform foundation and boot process of the UI #3682)
  • Catches errors that may occur before any React component can be rendered
• FlashMessages (have been there since ever)
  • Are used for handling anticipated, user-facing errors in various places

The @neos-project/neos-ui-error package now exposes a showFlashMessage function that can be used to display FlashMessages from anywhere in the system. The state of FlashMessages is now decoupled from the redux store. It is handled in a global, singleton Observable instead.

To achieve this, I introduced a new state handling mechanism via the @neos-project/framework-observable package using the approach I had outlined in #3331.

Thus far, I have implemented the createObservable and createState mechnisms with additional tests and documentation. Later on, this package will allow us to replace redux opportunistically where needed.

This change is breaking, because I have removed the now obsolete parts of the redux store that were related to flash messages. It is unlikely that third parties are relying on the parts I've removed, since they are implementation details related to the read-side (So unless a plugin needs to know what flash messages are currently displayed - which I doubt -, this should be safe).

The only thing still intact is the UI.FlashMessages.add action, because plugins may actually use it and currently have no other way to display flash messages. I have implemented a saga to redirect those action calls to the new mechanism.

My plan is to expose showFlashMessage publicly via a global JS API, that I will introduce in #3773. In that PR I will then deprecate the intermediate saga and the UI.FlashMessages.add action.

[mhsdesign]

Well done ❤️ especially thanks for keeping actionTypes.UI.FlashMessages.ADD in tact as we promote this as part of a guide "how to use neos ui redux" ^^ https://docs.neos.io/guide/advanced/extending-the-user-interface/react-extensibility-api#sagas

[mhsdesign]

i do not quite grasp this timeout here and also was wondering if its ideal that we delegate this timeout logic to this nested component instead of handling it in a higher level ... but please dont see this as critique but rather me trying to learn the tricks ;)

[grebaldi]

I've abstained from refactoring this stuff. This is only copied over from the original implementation and I agree that it's a bit smelly, but I didn't want to do more than strictly necessary to achieve the architectural goals.

[mhsdesign]

Perfect i oversaw that. In that case im even more grateful that its just moved as this should simplify the review :)

[mhsdesign]

thank you for providing extensive documentation for these framework parts.

I think we already partially discussed the idea of the "The Client-side framework" from #3331:

@neos-project/framework-observable

The @neos-project/framework-observable package provides a very simple implementation of the Observable-pattern (see: https://github.com/tc39/proposal-observable).

Furthermore, based on Observables, it establishes two key primitives for use in the UI:

State

A State observable represents a value that changes over time. It can be asked about its current value and any subscriber will receive the current value immediately.

It also provides an update method, that allows to push the next value calculated from the current value.

Channel

A Channel observable represents a sort of message bus. A subscriber will receive all messages that have been published to the Channel since the subscription was established.

It provides a publish method to push messages onto the channel.

@neos-project/framework-react-hooks

The @neos-project/framework-react-hooks package contains a bunch of essential React hooks to allow binding state to React components.

Most notably it contains the useLatestValueFrom hook, that allows to bind Observables to components.

And for the DTO serialisation and de-serialisation (@neos-project/framework-schema) you elaborated why it makes sense to provide a custom implementation instead of using a npm library: #3331 (comment)

Regarding observers and state handling i remember having talked to you regarding using rx-js vs a custom implementation like @neos-project/framework-observable and i think the reasons for a custom implementation were customisability (limited feature scope we only need to implement what we need), smaller bundle size, and that Observable will hopefully come to native ECMA script which we definitely want to profit from if available and this can be ensured by having it all under control.
But im a little fuzzy on the details and dont quite remember it all. So it would be great to discuss this to fill in the gaps. In the end this adds complexity to our codebase and thus the benefits and downsides should be carefully considered. When we decide for this library an outlined discussion would also help future maintainers to understand our decision ;)

EDIT:
I just skimmed over the implementation and its a super slim layer so +1 from my side ;) And we could consider this dicussion closed ^^ The documentation and tests are more than the code :D

[grebaldi]

You've summarized it perfectly :)

[markusguenther]

This PR is now in sync with the latest 9.0 changes. Thanks for all the work @grebaldi ❤️

[kitsunet]

LGTM

[mhsdesign]

As discussed with christian and markus we like this change very much thanks :)

We have to keep in mind that moving back from redux of course means that the store is not easily debuggable via browser extension and that plugins cannot hook into the process. This means we likely end up with more explicit defined extension points instead of one big store. But for the ui is this exactly what wee need. Small components like flashmessage especially profit from that as there is absolutely no need for extensibiliity.