The D Programming Language page 161.

[Adam D. Ruppe]

On Wednesday, 13 November 2013 at 08:16:34 UTC, Tor Einar 
Tønnessen wrote:
> void writeln(string a0, int a1, string a2, int[new] a3) {

Around the time the book was written, there was a debate as to if 
we should make dynamic arrays and slices two different types. 
T[new] was the proposed syntax for a dynamic array, leaving T[] 
to be a slice.

The reasoning would be to separate a window into the memory block 
- the slice - from the memory block itself, the array. This is 
the case with static arrays: int[10] != int[], but not with 
dynamic arrays. The advantage of the separation would be to help 
keep track of the array's ownership - appending to a T[new] could 
resize the block, whereas appending to a T[] would be a whole new 
allocation, among other things.

However, minds were changed and T[new] never actually came to be. 
Instead, slice append was changed to address the ownership 
stomping problem, it asks the gc for capacity at the end. The old 
problem is solved, and it didn't break any code by introducing a 
new type... everyone wins, except for the few things written 
about T[new] that weren't updated.

This article goes into the append to slice problem and solution 
in more detail:
http://dlang.org/d-array-article.html


Bottom line, any time you see T[new] in the book, just replace it 
with T[] and it should all work. So int[] a3 ought to compile and 
do the same thing described.

BTW, I don't see this item in the list, but on Andrei's website, 
there's a list of little changes since the printing:
http://erdani.com/index.php?cID=109