In-place extension of arrays only for certain alignment?

[Salih Dincer]

On Tuesday, 16 August 2022 at 18:11:54 UTC, Ali Çehreli wrote:
> ```d
>     version (good) {
>       // Dropping front elements equaling 2048 bytes works.
>       // (Likely a GC magic constant.)
>
>       enum magic = 2048;
>       enum elementsPerPage = magic / S.sizeof;
>
>       if (arr.length == elementsPerPage) {
>         arr = arr[elementsPerPage..$];
>       }
>     }
> ```

First I tried it on an older version (2.0.87).  And with 
version(good) I got interesting results:
```d
      (length == 1, capacity: 2031 -> 6127)
      i = 4079, 8175

      (length == 2, capacity: 1015 -> 3063)
      i = 2039, 4087, 6135, 8183

      (length == 3, null)
      i > 10.000 = ?

      (length == 4, capacity: 507 -> 1531)
      i = 1019, 2043, 3067, 4091, 5115, 6139, 7163, 8187, 9211

      (length == 5, 6, 7..., null)
      i > 10.000 = ?
```

For some reason the capacity changed frequently when data.length 
== 4 .  And I didn't see a result in 3, 5, 6, 7 since I tested 
less than 10k loops. I got more interesting results with the 
extra version:

```d
version(extra) arr.length++; /*

2087
291 elements, capacity: 582 -> 1167, ptr: 7F1306346010
1169 elements, capacity: 2338 -> 4093, ptr: 7F1306346010
2339 elements, capacity: 4678 -> 8189, ptr: 7F1306346010
4387 elements, capacity: 8774 -> 14626, ptr: 7F1306346010
7313 elements, capacity: 14626 -> 23403, ptr: 7F1306346010

Process finished.
*/
```

version(neither) is concurrent with i:

```d
       (length == 1, capacity: i)
       i = 4079, 8175..16367

       (length == 2)
       i = 2039, 4087, 8183..14327

       (length == 3, capacity: i)
       i = 1359, 2725, 5455, 9551..16378

       (length == 4, capacity: i)
       i = 1019, 2043, 4091, 7163..12283

       (length == 5, capacity: i)
       i = 815, 1635, 3273, 5731, 9827..16380
```

SDB at 79