On Sun, 14 Feb 2010 17:09:46 -0500, dsimcha <[email protected]> wrote:
== Quote from Andrei Alexandrescu ([email protected])'s
article
Now with the full legitimization of statically-sized array, the
syntactic kludge "new T[10]" is more painful than ever.
Walter and I were talking about changing the semantics of "new T[10]" to
mean "allocate a T[10] object and give me a pointer to it" and make the
syntax "new T[](10)" mean "allocate an array of 10 elements and give me
a handle to it."
I see major disruptions of existing code though. Do you guys have ideas
of a clean migration path?
Andrei
I'd recommend a simple compiler switch where new T[10] doesn't compile.
You turn
this on when you're porting old code. When your old code compiles with
this
switch on, you know it doesn't have any legacy "new T[10]" stuff in it.
You can
then turn this switch off and start using the new meaning of "new T[10]"
if you want.
Also, although I don't care about the breaking of old code (D2 is still
alpha and
warts should be fixed no matter how much it hurts backwards
compatibility), I
don't see what advantage this more verbose syntax offers. Why would
anyone ever
want to allocate a statically sized array on the heap? In these cases,
what's
wrong with just using a dynamic array?
A dynamic array consists of 2 elements, a length and a pointer. If you
never change the length, that's a lot of wasted bytes and passing around
of data if you never change it.
Also, the current interpretation makes generic code more difficult. It
makes statically sized arrays different than all other value types.
Just for that reason, it's worth the change.
If you're going to make the argument that new int[x] is a dynamic array,
then I'd argue that new int[x][y][z] should not have any static arrays
anywhere.
Having new int[x][y][z] make only the z dimension a dynamic array is
madness. I think the proposed fix is the most sane and consistent
possible.
-Steve