digitalmars.D - Is GC smart enough not to reallocate?
- MMJones (16/16) Jun 16 2016 Suppose one has something like
- Steven Schveighoffer (11/25) Jun 16 2016 If you reassign x, the compiler does not know enough context to assume
- John Colvin (36/52) Jun 16 2016 To prevent confusion, here's a related example:
Suppose one has something like class foo { int[] x; void bar() { x = []; } } Does the GC trash the "cache" when calling bar or does it realize that it can use the same memory for x and essentially just shortens the array? Is it equivalent to setting length = 0? I'm a bit worried that setting a managed array to [] might cause a completely new reallocation, which is unnecessary and undesirable.
Jun 16 2016
On 6/16/16 9:54 AM, MMJones wrote:Suppose one has something like class foo { int[] x; void bar() { x = []; } } Does the GC trash the "cache" when calling bar or does it realize that it can use the same memory for x and essentially just shortens the array?If you reassign x, the compiler does not know enough context to assume nothing else has a reference to x's old data. So no, it would not re-use that same data.Is it equivalent to setting length = 0?Even this is not going to overwrite the data. You'd need to do: x.length = 0; x.assumeSafeAppend;I'm a bit worried that setting a managed array to [] might cause a completely new reallocation, which is unnecessary and undesirable.Use assumeSafeAppend when you need to do this. BTW, x = [] is equivalent to x = null. So this is most certainly going to cause a new allocation on the next append. -Steve
Jun 16 2016
On Thursday, 16 June 2016 at 13:54:11 UTC, MMJones wrote:Suppose one has something like class foo { int[] x; void bar() { x = []; } } Does the GC trash the "cache" when calling bar or does it realize that it can use the same memory for x and essentially just shortens the array? Is it equivalent to setting length = 0? I'm a bit worried that setting a managed array to [] might cause a completely new reallocation, which is unnecessary and undesirable.To prevent confusion, here's a related example: void foo() { int[] x = [1,2,3]; x = [4]; } in theory, the first allocation (for [1,2,3]) could be avoided. It wouldn't be the GC doing it though, it would just be the optimiser eliminating the redundant initialisation of x. However: class C { int[] x; this() { x = [3,2,1]; } void foo() { x = [0]; } } auto bar() { auto c = new C; auto localX = c.x; c.foo(); return localX; } the initialisation of c.x is no longer redundant, because the memory is referenced by localX, so a new allocation is necessary in foo. P.S. remember that D's arrays/slices aren't "managed" as such. Only the memory backing them is managed, and then only if the memory was allocated by the GC.
Jun 16 2016