Good catch. I wonder why the linter doesn't complain.

also calling useEffect on line 349 of Tabs.js with no second argument as well.

This is intended.

So the solution would be?

diff --git a/packages/material-ui/src/Tabs/Tabs.js b/packages/material-ui/src/Tabs/Tabs.js
index b59620fb1..97b0021e4 100644
--- a/packages/material-ui/src/Tabs/Tabs.js
+++ b/packages/material-ui/src/Tabs/Tabs.js
@@ -334,6 +334,7 @@ const Tabs = React.forwardRef(function Tabs(props, ref) {
     debounce(() => {
       updateScrollButtonState();
     }),
+    [],
   );

   React.useEffect(() => {

@cubbie How did you spot it?

I had that warning pop up from Preact. Probably should remove useCallback, because it makes no sense to use this method with an empty array, it will create an additional wrapper which wont increase performance (at least I think? I could be wrong)

@cubbie Awesome to see the library used with Preact :D. I wish React had such warning, how comes they don't 😮. @eps1lon a potential opportunity for a contribution to React? ;)

Regarding the best option, removing vs keeping. I honestly don't know. Either way sounds great. @eps1lon What's your preference?

I think in this instance it will have to be kept because debounce creates a function. And the reason the ESLint plugin doesn't complain is because it only expects to be passed something it can statically analyse: a function (or an arrow function). It can't know what the result of calling debounce would be :)

It can't know what the result of calling debounce would be :)

But in 99.9% of the cases you'll use the second argument anyway. That should be the default. There are use cases to omit them for useMemo/useCallback but in those cases you're on your own and should disable the rule.

I actually can't think of the 0.01% that you'd not want a dependency array when using useMemo/useCallback, but yeah, maybe this discussion can be moved to the React core repo, and we just add []