Thread overview
How to flatten N-dimensional array?
May 24
9il
May 23
I have tried to implement a simple flatten function for multidimensional arrays with recursive templates but got stuck. Then I googled a little and stumped into complex https://rosettacode.org/wiki/Flatten_a_list#D implementation which requires your arrays to be either TreeList or Algebraic. That is, it does not work out-of-the-box on something like int[][][].

I'd like to clarify a couple of questions first.

How come Phobos doesn't have "flatten" function for arrays?

Is there an implementation of flatten that works out-of-the-box on N-dim arrays outside of Phobos? Excluding "flattened" from mir.ndslice since it works on Slices.
May 23
On 5/23/20 11:15 AM, Pavel Shkadzko wrote:> I have tried to implement a simple flatten function for multidimensional

> I'd like to clarify a couple of questions first.
>
> How come Phobos doesn't have "flatten" function for arrays?

We call in 'joiner'.

I wrote something like this:

import std.stdio;
import std.algorithm;
import std.range;

int value = 0;

auto makeNdim(size_t N)(size_t length)
if (N == 1) {
  auto result = iota(value, value + length).array;
  value += length;
  return result;
}

auto makeNdim(size_t N)(size_t length)
if (N > 1) {
  return iota(N).map!(n => makeNdim!(N - 1)(length)).array;
}

auto flatten(R)(R range)
if (!isInputRange!(ElementType!R)) {
  return range.joiner;
}

auto flatten(R)(R range)
if (isInputRange!(ElementType!R)) {
  return range.map!(r => r.joiner).joiner;
}

void main() {
  auto a = makeNdim!3(5);
  writefln!"Original : %s"(a);
  writefln!"Flattened: %s"(a.flatten);
}

Output:

Original : [[[0, 1, 2, 3, 4], [5, 6, 7, 8, 9]], [[10, 11, 12, 13, 14], [15, 16, 17, 18, 19]], [[20, 21, 22, 23, 24], [25, 26, 27, 28, 29]]]
Flattened: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29]

Ali

May 24
On Saturday, 23 May 2020 at 19:59:30 UTC, Ali Çehreli wrote:
> On 5/23/20 11:15 AM, Pavel Shkadzko wrote:> I have tried to implement a simple flatten function for multidimensional
>
> [...]

Thank you, I was lacking practical examples for templates with "if" constructs, ehh.
May 24
On Saturday, 23 May 2020 at 18:15:32 UTC, Pavel Shkadzko wrote:
> I have tried to implement a simple flatten function for multidimensional arrays with recursive templates but got stuck. Then I googled a little and stumped into complex https://rosettacode.org/wiki/Flatten_a_list#D implementation which requires your arrays to be either TreeList or Algebraic. That is, it does not work out-of-the-box on something like int[][][].
>
> I'd like to clarify a couple of questions first.
>
> How come Phobos doesn't have "flatten" function for arrays?
>
> Is there an implementation of flatten that works out-of-the-box on N-dim arrays outside of Phobos? Excluding "flattened" from mir.ndslice since it works on Slices.

If the common nd-array isn't jugged (a parallelotop), you can use fuse function.

----------
/+dub.sdl:
dependency "mir-algorithm" version="~>3.8.12"
+/
import std.stdio: writeln;
import mir.ndslice;

void main() {
    auto arr =
    [[[0, 1, 2, 3, 4],
      [5, 6, 7, 8, 9]],
     [[10, 11, 12, 13, 14],
      [15, 16, 17, 18, 19]],
     [[20, 21, 22, 23, 24],
      [25, 26, 27, 28, 29]]];
    auto flatten = arr.fuse.field;

    static assert(is(typeof(flatten) == int[]));
   	assert(flatten == 30.iota);
}
----------

It performs exactly one allocation.
May 24
On Sunday, 24 May 2020 at 11:21:00 UTC, 9il wrote:
> On Saturday, 23 May 2020 at 18:15:32 UTC, Pavel Shkadzko wrote:
>> [...]
>
> If the common nd-array isn't jugged (a parallelotop), you can use fuse function.
>
> ----------
> /+dub.sdl:
> dependency "mir-algorithm" version="~>3.8.12"
> +/
> import std.stdio: writeln;
> import mir.ndslice;
>
> void main() {
>     auto arr =
>     [[[0, 1, 2, 3, 4],
>       [5, 6, 7, 8, 9]],
>      [[10, 11, 12, 13, 14],
>       [15, 16, 17, 18, 19]],
>      [[20, 21, 22, 23, 24],
>       [25, 26, 27, 28, 29]]];
>     auto flatten = arr.fuse.field;
>
>     static assert(is(typeof(flatten) == int[]));
>    	assert(flatten == 30.iota);
> }
> ----------
>
> It performs exactly one allocation.

Yep, I know about Mir fuse. I was more wondering about the absence of flatten in Phobos. But on dlang forum some people claimed that when it comes to anything multidimensional, use Mir.
May 24
On 5/24/20 2:37 AM, Pavel Shkadzko wrote:
> On Saturday, 23 May 2020 at 19:59:30 UTC, Ali Çehreli wrote:
>> On 5/23/20 11:15 AM, Pavel Shkadzko wrote:> I have tried to implement a simple flatten function for multidimensional
>>
>> [...]
> 
> Thank you, I was lacking practical examples for templates with "if" constructs, ehh.

Template constraints are great but 'static if' can be more useful as it allows custom error messages:

auto makeNdim(size_t N)(size_t length) {
  static if (N == 1) {
    auto result = iota(value, value + length).array;
    value += length;
    return result;

  } else static if (N > 1) {
    return iota(N).map!(n => makeNdim!(N - 1)(length)).array;

  } else {
    static assert(false, "N cannot be 0.");
  }
}

Ali