GC Scan Range

The Go garbage collector must scan structs for pointers. It scans from offset 0 to the last pointer-containing byte — the GC scan range.

Why field order matters

When pointer fields are scattered after non-pointer fields, the GC scans padding and non-pointer data between them. Grouping pointers at the start reduces the scan range.

Pointer-containing types

These types contain at least one pointer word:

Type	Pointer words	Reason
*T	1	Single pointer
string	1	Pointer to string data + length
[]T	1	Pointer to slice data + len + cap
map[K]V	1	Pointer to hash table
chan T	1	Pointer to channel
func(...)	1	Function pointer
interface{}	2	Type descriptor + data pointer

Example

type Widget struct {
    a bool     // offset 0,  1 byte, no pointer
    b *string  // offset 8,  8 bytes (+7 padding), pointer!
    c int64    // offset 16, 8 bytes, no pointer
    d string   // offset 24, 16 bytes, pointer!
    // GC scan range: 40 bytes (from 0 to end of d)
}

With GC-optimal ordering:

type Widget struct {
    b *string  // offset 0,  8 bytes, pointer!
    d string   // offset 8,  16 bytes, pointer!
    a bool     // offset 24, 1 byte,  no pointer
    c int64    // offset 32, 8 bytes, no pointer
    // GC scan range: 24 bytes (from 0 to end of d)
}

The GC scan range dropped from 40B to 24B — a 40% reduction.