setNativeProps with style is incompatible with also updating style prop itself

[comp615]

Knocking this out quick before a meeting, but will try to understand more tonight.

The problem
When using setNativeProps to set the style of an object, the explicitly passed style (i.e. the style prop of a View), gets locked to its initial value. If it's changed, it will no longer update after 1-2 render passes. I believe this is because the props get internally "cached", but then it will always clobber the "new" style prop, thereby freezing it.

How to reproduce
Simplified test case: https://codesandbox.io/s/sweet-davinci-f1lut?file=/src/App.js

Steps to reproduce:

1. Click the button, it should expand the square
2. Notice that after a few clicks it stops expanding the square

Expected behavior
The box should continue expanding indefinitely

Environment (include versions). Did this work in previous versions?
Yes, this is a regression in #1939 / 0.15.2

• React Native for Web (version): 0.15.2+

I am happy to submit a PR with tests in a few hours, but I believe reversing the order of style and previous style here will correct it https://github.com/necolas/react-native-web/blob/master/packages/react-native-web/src/modules/usePlatformMethods/index.js#L27

[comp615]

@piaskowyk curious why previous styles is the second arg in the array there instead of the first. I'm assuming there may be some other issue you were working around there.

[necolas]

Damn. I knew that PR was going to cause a regression, I had just lost context since stabilizing that method and there isn't test coverage of all the undefined RN behaviors implemented. So yeah adding tests sounds great

[necolas]

I believe reversing the order of style and previous style here will correct it

That didn't work. But I think I have a fix...or at least I can produce the expected results for both issue test cases. No tests yet

[comp615]

@necolas thanks for looking at this, I ended up in a bunch of meetings today so still haven't gotten further, but looks like you are on top of it. Happy to help test as I can once you have something. FWIW, on twitter.com this manifests as the Image cropper not being zoomable (maybe other manifestations but that's the only one I've found so far)

[necolas]

I'm going to publish a 0.15.x patch that reverts the regression, and we can figure out how to accommodate any changes to the logic in a future minor release.

[necolas]

Reverted in 0.15.7