Re-assigning NaN is seen as a value change

[Legioth]

• Check if updating to the latest version resolves the issue

Environment

• I am using @preact/signals-core
• I am using @preact/signals
• I am using @preact/signals-react
  • I understand usage changed in v2, and I've followed the React Integration instructions

Describe the bug
In JavaScript, NaN === NaN evaluates to false. A consequence of this is that a signal with the value NaN will trigger an update if the value is again assigned as NaN. In the worst case, this might lead to an infinite loop with a two-way UI binding if the receiving end is also treating NaN in the same way.

To Reproduce

const x = signal(0);
effect(() => console.log(x.value));
x.value = NaN;
x.value = x.value;

This logs NaN twice even though it's expected that the last row shouldn't lead to running the effect again.

Expected behavior
Expected that the dirty checking logic has a special case for re-assigning NaN

[rschristian]

I don't want to say this won't be supported here, but I'd really, really recommend against setting NaN in any state value. It's essentially an unhandled error, keeping it around is usually a mistake. Better to fallback to -1 or gracefully handle it in a similar form.

[Legioth]

Just for context, I discovered through through doing parseInt(fieldValue) for an empty string and then wasted some debugging time looking at other things before remembering that NaN != NaN. More graceful handling in the signal would have helped me avoid an infinite loop which would in turn have made it easier for me to realize that I had an unexpected "number" in the system.

[rschristian]

You should really never call parseInt without having a fallback though is my point, which is more of a thing for a linter to solve (I seem to recall there being a rule for this to, but not positive).

I believe the intention here is to still ship ES5, so Object.is() is likely out, and anything else would probably hurt perf slightly. We also don't support it in Preact for that reason.

---

Again, not saying we won't, but I at least would prefer not to. It's a bad pattern to pass NaN around, IMO.

[dmail]

I agree it's more the job of a linter and it's a bad pattern to have NaN.
But this is an easy mistake to make. I think if preact signals can help human devs to catch errors and gain time it should.