I can't find any algorithm/range in Phobos that can be used to split (eagerly or lazily) a sequence using a binary predicate on adjacent elements as follows

[1,2,3,5,10,11,12,13,20,21,100].splitBy!"a + 1 != b"()

should evaluate to

[[1,2,3], [5], [10,11,12,13], [20,21], [100]]

.

Is there one?

On Tuesday, 17 October 2017 at 13:09:18 UTC, Nordlöw wrote:
> Is there one?

If not, what should we call it?...yet another overload of `splitter()` with a binary predicate and only a single parameter `Range input`?

On Tuesday, 17 October 2017 at 13:25:01 UTC, Nordlöw wrote:
> On Tuesday, 17 October 2017 at 13:09:18 UTC, Nordlöw wrote:
>> Is there one?
>
> If not, what should we call it?...yet another overload of `splitter()` with a binary predicate and only a single parameter `Range input`?

Would probably need to use findAdjacent to get it to work
https://dlang.org/phobos/std_algorithm_searching.html#findAdjacent

So splitterAdjacent?

On Tuesday, 17 October 2017 at 13:09:18 UTC, Nordlöw wrote:
> I can't find any algorithm/range in Phobos that can be used to split (eagerly or lazily) a sequence using a binary predicate on adjacent elements as follows
>
> [1,2,3,5,10,11,12,13,20,21,100].splitBy!"a + 1 != b"()
>
> should evaluate to
>
> [[1,2,3], [5], [10,11,12,13], [20,21], [100]]
>
> .
>
> Is there one?

Try this:

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

import std.typecons;

auto splitBy(alias F, R)(R range)
{
	
	auto tmp = range
		.map!(x => tuple(x, 0))
		.cumulativeFold!((a,b) => tuple(b[0], (!F(a[0],b[0]))?a[1]:a[1]+1))
		.chunkBy!((a,b) => a[1] == b[1])
		.map!(x => x.map!(y => y[0]));
	
	return tmp;
}

void main()
{
	[1,2,3,5,10,11,12,13,20,21,100].splitBy!((a,b) => a+1 != b)().writeln;
}


Andrea

More phobos-ized version:

https://run.dlang.io/is/iwgeAl

Andrea

On Tuesday, 17 October 2017 at 15:47:25 UTC, Andrea Fontana wrote:
> More phobos-ized version:
>
> https://run.dlang.io/is/iwgeAl

Thanks!

On Tuesday, 17 October 2017 at 14:15:02 UTC, Andrea Fontana wrote:
> auto splitBy(alias F, R)(R range)

Because of lazyness shouldn't it be named something with splitter, say splitterBy, instead?

On Wednesday, 18 October 2017 at 06:45:37 UTC, Nordlöw wrote:
> On Tuesday, 17 October 2017 at 14:15:02 UTC, Andrea Fontana wrote:
>> auto splitBy(alias F, R)(R range)
>
> Because of lazyness shouldn't it be named something with splitter, say splitterBy, instead?

Yes but I think it is something more similar to chunkBy, but chunkBy says that "predicate must be an equivalence relation, that is, it must be reflexive (pred(x,x) is always true), symmetric (pred(x,y) == pred(y,x)), and transitive (pred(x,y) && pred(y,z) implies pred(x,z)). If this is not the case, the range returned by chunkBy may assert at runtime or behave erratically."

If you try to use your data with chunkBy!"a != b+1", it does not work, as expected.

I think that my implementation could superseed the current one, since it seems to work in a more generic way.

Andrea

On Wednesday, 18 October 2017 at 07:01:19 UTC, Andrea Fontana wrote:
> If you try to use your data with chunkBy!"a != b+1", it does not work, as expected.

What's the motivation behind this limitation?

Without it

    chunkBy!"a + 1 == b"

is exactly what I want.

On Wednesday, 18 October 2017 at 07:26:20 UTC, Nordlöw wrote:
> On Wednesday, 18 October 2017 at 07:01:19 UTC, Andrea Fontana wrote:
>> If you try to use your data with chunkBy!"a != b+1", it does not work, as expected.
>
> What's the motivation behind this limitation?
>
> Without it
>
>     chunkBy!"a + 1 == b"
>
> is exactly what I want.

Probably it's an implementation problem. On source code you read:

// Issue 13595
version(none) // This requires support for non-equivalence relations
@system unittest
{
    import std.algorithm.comparison : equal;
    auto r = [1, 2, 3, 4, 5, 6, 7, 8, 9].chunkBy!((x, y) => ((x*y) % 3) == 0);
    assert(r.equal!equal([
        [1],
        [2, 3, 4],
        [5, 6, 7],
        [8, 9]
    ]));
}

And other.