Mistake #22 "nil vs empty slices" regarding memory allocation of empty slices

[denpeshkov]

Item 22 says:

One of the main differences between a nil and an empty slice regards allocations. Initializing a nil slice doesn’t require any allocation, which isn’t the case for an empty slice.

Correct me if I am wrong, but AFAIK both slices will have no memory allocations.

Slices are represented in memory as 3 words, defined by the slice header:

type slice struct {
	array unsafe.Pointer
	len   int
	cap   int
}

A nil slice is actually a zero value of the slice header:

slice {
	array: nil,
	len:   0,
	cap:   0,
}

An empty slice is just a slice with len and cap equal to 0 and an array pointing to 'zero-sized' array:

slice {
	array: unsafe.Pointer(&zerobase),
	len:   0,
	cap:   0,
}

The value of the array is the address of the runtime.zerobase, the base address for all 0-byte allocations

[wureny]

The empty slice itself doesn't need memory allocation, but when Initializing it, we also need to initialize the slice header, which contains a pointer(array) that needs memory allocations. zerobase is the base address for all 0-byte allocations, but itself is not 0.

[denpeshkov]

@wureny

I am not really proficient in Assembly, but looking at the code, the only difference is which value gets stored in AX register.

Assembler for nil slice:

XORL	AX, AX // zeroing
XORL	BX, BX // zeroing
MOVQ	BX, CX	

Assembler for empty slice:

LEAQ	runtime.zerobase(SB), AX // take address
XORL	BX, BX // zeroing
MOVQ	BX, CX

Compiled for GOOS=linux GOARCH=AMD64

[teivah]

As said in #74 (comment)