Rule idea: no imprecise number literals

Please describe what the rule should do:

Flag number literals which do not precisely represent an IEEE754 double. Have an option to allow number literals that round to Infinity, such as 2e308, a somewhat common pattern to avoid referring to the Infinity global.

What category of rule is this? (place an “X” next to just one item)

• Enforces code style
• Warns about a potential error
• Suggests an alternate way of doing something
• Other (please specify:)

Provide 2-3 code examples that this rule will warn about:

0.1234567890123456789

12345678901234567890

12345678.901234567890

0xFFFFFFFFFFFFFFF

Why should this rule be included in ESLint (instead of a plugin)?

It’s a possible error that can affect all JavaScript users and can lead to logic bugs that are very hard to notice.

[not-an-aardvark]

In practice, what’s the advantage of referring to 0.10000000000000000555 rather than 0.1? Does the system become more mathematically sound (e.g. by avoiding unexpected rounding errors)?

To me, 0.1 seems more readable than 0.10000000000000000555 because the former makes it clear that the literal is supposed to represent the mathematical value of 1/10, to the extent that floating-point precision can allow.

[rightfold]

The rule should probably display the exact value of the literal, so that it can be substituted for by the programmer.