Added note about multiple Reacts, since this error commonly implies that.

[jimfb]

Added note about multiple Reacts, since this error commonly implies that.

[jimfb]

@spicyj you have context on this diff. For those following along at home, we've seen a bunch of github issues where people are confused, which means it's probably confusing a bunch of people who don't report it to us. We want to point users in the right direction with a helpful/timely message.

Lint failure is from a different file, unrelated to change. Should be fixed independently.

[zpao]

(lint is some false positive thanks to build caching, cleared and retriggered - I hit it on master too)

[maoziliang]

How about add a magic token in the global object. Use the token to determine whether a past React had been loaded. And then print a warning ?

[jimfb]

@maoziliang We tried that, but it ended up warning people in unit tests and we had to revert the diff (#3646).

@zpao or @spicyj ready to merge?

[jimfb]

Ping.

[jimfb]

@zpao or @spicyj Is there a reason we are not moving this forward, or should I keep pinging the thread? Soon I'll get as good at this as pingbot!

[sophiebits]

I want to criticize the too-long wording but haven't gotten around to it yet…

[jimfb]

@spicyj Comment overall or specifically my addition?

[sophiebits]

Mostly it's just really long now.

[syranide]

@jimfb An idea would be to add a property (not attribute) to React root nodes with a per-React instance random value. Then for all the various asserts that can be affected by multiple Reacts we first check whether the property of the root node of that React tree has the value we expect, if it doesn't we know there's a different instance present.

[jimfb]

@spicyj Better? It is now 4 lines instead of 7 lines :P.

@syranide That's not a bad idea. If the root node is a web component, it might respond badly to having a random property set on it, but maybe we skip the check for web components? It's worth looking into, but not worth blocking this fix. This is a trivial mitigation that will catch the most common case.

[sophiebits]

https, period at the end of the sentence

[sophiebits]

Okay, I guess this is fine. Can you shorten the remove message too? (Not sure I've ever seen it though since you'd always hit the add invariant first in practice.)

[jimfb]

@spicyj Yeah, done. Is it even possible to hit that remove path? Seems like you'd need to add the ref in order to remove it, so it would be impossible to hit.