[williamdclt]

I agree with the author that it’s almost never what you want. But I agree with you that it’s the reality of the platform, ignoring it will cause its own problems.

I’d surface the footgun rather than trying to pretend it’s not there: isNonArrayObject and isObjectOrArray, or something like that

[7bit]

Absolutely agree. I also hate that empty arrays are true, which is different from other languages. But I agree that it's better to face the reality of the language than create a function that evaluates [] to false. It trains you a bad habitnand some day that will cause you to introduce a bug.

[williamdclt]

> I also hate that empty arrays are true, which is different from other languages

I don’t mind that actually! I don’t think I have much use cases for “empty array is semantically different from non-empty”. Usually I find null/undefined are better choices, an empty array is just a normal array, I don’t expect it to be handled differently

[7bit]

What do you think about empty strings being falsy in most languages including js?

Null/undefined is a better choice, but there's many occasions where you do not have the power of choice. For example with document.querySelectorAll, which returns an empty array if nothing is found. The simple thing to do is to just check for it's length or just iterate over it's nodes, but still. I prefer empty arrays being falsy.

Just to clarify, I'm not saying one is better than the other. I just prefer how it works in other languages like Python. But I still would rather work with the JS language properties, than import a library that changes how I test for empty arrays.

[williamdclt]

> What do you think about empty strings being falsy in most languages including js?

I don’t love it! I think it’s inconsistent with empty arrays being truthy.

In practice, I see a lot more mishandling of empty strings than empty arrays (eg “!myvar” catching null and empty strings when it’s not what was meant) which hints that it’s not the right thing to do: an empty string is a valid string, whether its semantically different from non-empty entirely depends on the context and shouldn’t be baked into the langage.

Generally, I think implicit casts are not a great idea and I’d rather they didn’t exist (one of the few things I think Go got right).

> For example with document.querySelectorAll, which returns an empty array if nothing is found

Yes I think it’s doing the right thing. It’s indeed returning the list of elements, whether there’s 0 elements or more it’s still a valid list. As you say, you don’t usually need to handle it any differently from a non-empty array (just iterate), so I don’t think it should cast to a different value.

However if for example you were passing something like a whitelist as an argument, I’d expect [] to mean “there is a whitelist, allow only these items (ie allow nothing)” and null to mean “no whitelist, everything allowed”. These two things are semantically different, I think it makes sense to cast to different values.

I do see your point and it’s definitely at least partly a matter of taste. Again id rather there was no implicit conversion at all, removing the risk of mishandling altogether!

[nothrabannosir]

Akshually, implicitly casting any non Boolean type to true or false is no better than implicitly casting “0” to 0. :) B-)

[7bit]

No better based on what criteria? In what language? This is typical in Python and done even by core-devs. It is how the language was designed and it was designed to support that.

[nothrabannosir]

In the "akshually" sense. "Zen of programming" (or "I'm smart because I use the word monoid unironically").

"If is an anti pattern", "null values are an anti pattern", "`string' is an anti-type", "util libraries are an anti pattern", etc.

And of course "boolean is an anti type" but let's not get into that. :D

(Nor into the value of "idiomatic python" as an argument on healthy programming habits....)

[moritzwarhier]

Lodash also has "isPlainObject". Not sure if that's as performant as using Array.isArray or even simpler trickery (see below), but it seems that this fixes it.

"isObject" is just not well-defined for other cases that might be interesting, in my opinion. Not just "null" or arrays.

Functions can have properties (although they don't have the object prototype or primitive type).

Class instances behave differently compared to plain objects when used with libraries or other code that inspects the shape or prototype of objects, or wants to serialize them to JSON.

If you deal with unknown foreign values that are expected to be a JSON-serializable value, you could go with something like

  isObject = (o) => !!o && typeof o === "object" && !Array.isArray(o)

But it does not deal with all the non-JSON objects, e.g. Map or other class instances that are not even JSON-serializable by default.

When data comes from parsing unknown JSON, the check above should be enough.

In other cases, the program should already know what is being passed into a function. No matter if through discipline or with the aid of an additional type syntax.

For library code that needs to reflect on unknown data at runtime, I think it's worth looking at Vue2's hilarious solution, probably a very good and performant choice, no matter how icky it might look:

  https://github.com/vuejs/vue/blob/547a64e9b93d24ca5927f653710b5734fa909673/src/util/lang.js#L293

Since the string representations of built-ins have long ago become part of the spec, it seems that there is no real issue with this!

[dcsan]

your option is a bit more verbose but definitely more clear. confusing the underlying definitions of the language itself will lead to problems later.

[PUSH_AX]

Will it? Will it really?

I’ve been writing JS for so long I’ve forgotten all these language quirks, I feel like it’s fair for most people, these language choices are kind of meaningless in day to day, what’s meaningful is a function returning things that will make sense to most people. Or at least have two functions, languageStrictIsObject()

[williamdclt]

Yeah, in practice I would only encounter the case of arrays being objects when doing something highly polymorphic (eg reading some JSON that can be anything), which likely would have test cases for every type anyway.

But when I’m writing a reusable lib and I can’t perfectly abstract a gotcha, I design the API so that the dev _has_ to make an explicit decision rather than defaulting on their behalf (defaults are evil). So, isObjectOrArray and isObjectNonArray: you have to think about which applies. Another example is sorting: does it sort in-place or immutably? Make it explicit in the function name. Does a function expect a sorted array or will it sort itself? Make it explicit in the function name. And maybe provide both variants.