Style/ArrayCoercion does not work with Time objects

[Aqualon]

The Style/ArrayCoercion cop introduced in #8295 has some unexpected auto-correct behavior if dealing with Time objects.

Expected behavior

Style/ArrayCoercion auto-correct should not break code if dealing with Time objects.

Actual behavior

Style/ArrayCoercion auto-correct breaks if the wrapped variable is a Time object.

Steps to reproduce the problem

We have code like this to wrap an Time object called inserted_at in an Array

inserted_at = [inserted_at] unless inserted_at.is_a? Array

This was replaced by auto-correct with

inserted_at = Array(inserted_at)

This leads to an array with each part of the Time object as value

inserted_at = Time.now
2020-07-14 14:54:33.339436 +0200
Array(inserted_at)
[33, 54, 14, 14, 7, 2020, 2, 196, true, "CEST"]

I don't know why this behavior happens, but it seems just calling Array on some object, my lead to unexpected breaks.

RuboCop version

0.88.0 (using Parser 2.7.1.4, rubocop-ast 0.1.0, running on ruby 2.7.1 x86_64-darwin19)

[Aqualon]

https://ruby-doc.org/core-2.7.1/Kernel.html#method-i-Array describes Array

First tries to call to_ary on arg, then to_a. If arg does not respond to to_ary or to_a, returns an Array of length 1 containing arg.

The Time class provides to_a: https://ruby-doc.org/core-2.7.1/Time.html#method-i-to_a

So I assume for any object providing a to_a method, there may be unexpected behavior, because instead of [my_object] you could get whatever my_object returns from its to_a method.

[marcandre]

Indeed. I think the equivalent of Array(foo) is [*foo].

[bbatsov]

Well, I guess it's safe to say that this cop is unsafe. Still, most of the time it does improve the code a bit.

[Aqualon]

I'm not sure I'd call it an improvement if it may change the code to something, that behaves differently than before.

[marcandre]

See also #8391

[dvandersluis]

I think this should probably be closed now that #8792 is merged?

[marcandre]

Indeed 👍