[RNmobile] Fix problems with undo/redo on Android

[daniloercoli]

This PR fixes a problem where the undo/redo feature was not working on Android if small changes were made to the content.

Actually undo/redo feature was not working fine even if there were lot of changes to the content, but in this case the problem was less noticeable, since the undo/redo feature replaced the content with the wrong text.

The problem was that RichText props were not correctly updated on typing. Then on undo the RichText component already had the same "old" value stored in it, and the Native side was not refreshed at all (The setText method of the AztecWrapper was not invoked).

To test this PR follow the steps in the gb-mobile PR here:wordpress-mobile/gutenberg-mobile#297

[mzorz]

Tested, explanation and code make sense, and was able to confirm it fixes the behavior it intends to fix.
I've seen the e2e tests are failing due to reasons unrelated to this PR.
LGTM feel free to merge as soon as those get fixed

[mzorz]

I noticed while debugging and setting a breakpoint in shouldComponentUpdate() in rich-text that in this branch all RichText components return true in shouldComponentUpdate(), even if no change happened to them (that is, they are not the focused block or the block where changes are happening). It doesn't make sense to change the state of every single block instance in the list when a simple change happened in one block only.

I couldn't think of a solution for it, so leaving the comment here for now.

[daniloercoli]

That's right @mzorz, but the shouldComponentUpdate is only called on those blocks rendered on the screen. Since we're using a FlatList, only few blocks are loaded on the screen/memory at the same time, there should no big performance impact on re-rendering those.
Also consider that eventCount is sent together with the content to the Native side, and used there to block the refresh of the native AztecWrapper component when it's not needed (I guess it could be undefined if you don't modify the block).

I agree with you that a better solution is required, but for the alpha I think it's ok, and well tested since returning true very often :) Just need to verify performance problems, since the whole shouldComponentUpdate is there to avoid those bottlenecks we were seeing in aztec-wrapper.

[daniloercoli]

I did another test this morning to confirm that the native side is only hit (setText on the AztecWrapper called) for the one single block where the changes happens. This is because for the other blocks no props are changed and React doesn't trigger the re-render (render is called React side but does't update the component).

If you want to test this put a breakpoint on the native side setText method, and see that it is only called once on redo/undo.