In PSv3+, with its unified handling of scalars and collections, any object - even $null - should have a .Count property (and, with the exception of $null, should support indexing with [0]).
Any occurrence of an object not supporting the above should be considered a bug; for instance:
Using this intrinsic (engine-supplied) .Count property unexpectedly fails when Set-StrictMode-Version 2 or higher is in effect, which is a long-standing bug reported in GitHub issue #2798, still present as of PowerShell 7.2 (whereas a type-native .Count property, such as on an array, can safely be accessed).
([pscustomobject] instances not playing by these rules was a known bug, fixed in 2017).
Since I don't know if said bug is related to the [Microsoft.Management.Infrastructure.CimInstance#ROOT/Microsoft/Windows/Storage/MSFT_Disk] instances that Get-Disk outputs, and since Get-Disk - at least currently - is only available in Windows PowerShell, I encourage you to file a separate bug on uservoice.com.
Use of array-subexpression operator @(...) is only necessary:
Generally, if you do need to ensure that something is an array, use @(...) rather than [Array] ... / [object[]] ... - @() is PowerShell-idiomatic, more concise, and syntactically easier.
That said, given that @() technically creates a (shallow) copy of an existing array, you may prefer [Array] when dealing with potentially large arrays.
Additionally, @(...) and [Array] ... are not generally equivalent, as PetSerAl's helpful examples in a comment on the question demonstrate; to adapt one of his examples:
@($null) returns a single-item array whose one and only element is $null, whereas [Array] $null has no effect (stays $null).
This behavior of @() is consistent with its purpose (see below): since $null is not an array, @() wraps it in one (resulting in a [System.Object[]] instance with $null as the only element).
In PetSerAl's other examples, @()'s behavior with New-Object-created arrays and collections - may be surprising - see below.
The purpose of @(...) and how it works:
The purpose of @(), the array-subexpression operator, is, loosely speaking, to ensure that the result of an expression/command is treated as an array, even if it happens to be a scalar (single object).:
@(...) collects an enclosed command's output as-is / an enclosed expression's enumerated output in an - always new - [object[]] array, even if there's only a single output object.
@(...) is never needed for array literals (in v5.1+ it is optimized away) - use of ,, the array constructor operator by itself is generally sufficient, e.g., 'foo', 'bar' instead of @('foo', 'bar') - but it is useful in the following cases:
to create an empty array: @())
to create a single-element array - e.g. @('foo') - which is easier to read than the unary form of , that would otherwise be required - e.g. , 'foo'
for syntactic convenience: to spread what is conceptually an array literal across multiple lines without having to use , to separate the elements and without having to enclose commands in (...); e.g.:
@(
'one'
Write-Output two
)
Pitfalls:
@() is not an array constructor, but a "guarantor": therefore, @(@(1,2)) does not create a nested array:
@(@(1, 2)) is effectively the same as @(1, 2) (and just 1, 2). In fact, each additional @() is an expensive no-op, because it simply creates a copy of the array output by the previous one.- Use the unary form of
, the array constructor operator, to construct nested arrays:
, (1, 2)
$null is considered a single object by @(), and therefore results in a single-element array with element $null:
Command calls that output a single array as a whole result in a nested array:
@(Write-Output -NoEnumerate 1, 2).Count is 1
In an expression, wrapping a collection of any type in @() enumerates it and invariably collects the elements in a (new) [object[]] array:
@([System.Collections.ArrayList] (1, 2)).GetType().Name returns 'Object[]'
Read on for more detailed information, if needed.
Details:
@() behaves as follows: Tip of the hat to PetSerAl for his extensive help.
In PSv5.1+ (Windows PowerShell 5.1 and PowerShell [Core] 6+), using an expression that directly constructs an array using ,, the array constructor operator, optimizes @() away:
E.g., @(1, 2) is the same as just 1, 2, and @(, 1) is the same as just , 1.
In the case of an array constructed with just , - which yields a System.Object[] array - this optimization is helpful, because it saves the unnecessary step of first unrolling that array and then repackaging it (see below).
Presumably, this optimization was prompted by the widespread and previously inefficient practice of using @( ..., ..., ...) to construct arrays, stemming from the mistaken belief that @() is needed to construct an array.
However, in Windows PowerShell v5.1 only, the optimization is unexpectedly also applied when constructing an array with a specific type using a cast, such as [int[]] (the behavior has been corrected in PowerShell [Core] 6+ and older Windows PowerShell versions are not affected); e.g.,
@([int[]] (1, 2)).GetType().Name yields Int32[]. This is the only situation in which @() returns something other than System.Object[], and assuming that it always does can lead to unexpected errors and side effects; e.g.:
@([int[]] (1, 2))[-1] = 'foo' breaks.
$a = [int[]] (1, 2); $b = @([int[]] $a) unexpectedly doesn't create a new array - see this GitHub issue.
Otherwise: If the (first) statement inside @(...) is an expression that happens to be a collection, that collection is enumerated; a command's (typically one-by-one streaming) output is collected as-is; in either case the resulting count of objects determines the behavior:
