tools: add number-isnan eslint rule

[maclover7]

Checklist

• make -j4 test (UNIX), or vcbuild test (Windows) passes
• tests and/or benchmarks are included
• commit message follows commit guidelines

Affected core subsystem(s)

tools, test

[apapirovski]

This might be a bit finicky since it would catch const isNaN = Number.isNaN; isNaN(val); too? But maybe that's not a real problem... 😆

[Trott]

No objections, but one question and one observation.

Question: What's the reason for avoiding the isNaN() global?

Observation: I think this can probably be implemented using no-restricted-syntax rather than a custom rule.

[maclover7]

@Trott

Question: What's the reason for avoiding the isNaN() global?

For personal reasons, I think it's more clear to use Number.isNaN. Also, several Code and Learn PRs have been submitted to move towards Number.isNaN, so I figured that people had come to consensus that it was a good idea.

Observation: I think this can probably be implemented using no-restricted-syntax rather than a custom rule.

Interesting -- refactored down to use no-restricted-syntax

[apapirovski]

Unfortunately this won't work as it overrides all the no-restricted-syntax rules declared within .eslintrc.yaml in the root directory.

[Trott]

Unfortunately this won't work as it overrides all the no-restricted-syntax rules declared within .eslintrc.yaml in the root directory.

Oh, yeah, I assumed it was a lint rule we'd want for the entire code base...

[apapirovski]

Oh, yeah, I assumed it was a lint rule we'd want for the entire code base...

Yeah, I think we could do that? I don't see any reason why we would be stricter about this in test than lib.

[Trott]

Are we sure we want to necessarily encourage people to change isNaN() to Number.isNan(). They behave differently.

isNaN('a string is not a nubmer'); //true
Number.isNaN('a string is not a number'); // false

[apapirovski]

Number.isNaN should be the new standard way of doing this as per ES2015. There's weird coercion in isNaN.

[maclover7]

updated @apapirovski @Trott

For context, I was initially hesitant to add the rule to lib/, since we are generally performance sensitive, but I don't think the change will massively alter performance. Should benchmarks be run anyways?

[Trott]

@maclover7 Sorry to make this tedious, but this is what I'd recommend:

• Revert to your initial custom rule and apply it just to test
• If you wish, in a separate PR, propose changing isNaN() to Number.isNaN() in lib and prepare to run benchmarks and have a fair amount of conversation about it. :-D
• If the above bullet point suggestion lands, then use the restricted-syntax thing to replace your custom rule.

[Trott]

(And yeah, had I not said anything, first bullet point would have happened and all of this would have been avoided. Sorry!)

[not-an-aardvark]

I think the typeof n !== 'number' check is redundant if Number.isNaN is being used.

[not-an-aardvark]

I think the typeof historySize !== 'number' check is redundant if Number.isNaN is being used.

[maclover7]

updated @Trott, PTAL

[maclover7]

CI: https://ci.nodejs.org/job/node-test-pull-request/12184/

[maclover7]

Will land tomorrow unless any objections

[maclover7]

Landed in e554bc8