Row/column major ordering about mir HOT 9 CLOSED

See also #9

from mir.

Hmm I like your idea of overloading opCall, but it works only for element access.
What do you think about something like the following?
(the last two are proposed)

It should basically work similar to opIndex then ;-)

auto s = iota(4).sliced(2,2);
s[].writeln; // [[0, 1], [2, 3]]
s().writeln; // [0, 2], [1, 3]]
s(1).writeln; // [[0,2]]

from mir.

Hmm I like your idea of overloading opCall, but it works only for element access.

Yes, first we need a good DIP for opCall to be like opIndex. See http://wiki.dlang.org/DIPs

It should basically work similar to opIndex then ;-)

Behavior with .writeln looks bad to me. opCall should only select sub-dimensions without any transpostion.

from mir.

Yes, first we need a good DIP for opCall to be like opIndex. See http://wiki.dlang.org/DIPs

I guess one could use iota instead of .., apart from that variadic arguments should be supported ;-)

Behavior with .writeln looks bad to me. opCall should only select sub-dimensions without any transpostion.

What would you suggest s(1) should return then?

from mir.

I guess one could use iota instead of .., apart from that variadic arguments should be supported ;-)

Anyway $ is not defined for opCall

What would you suggest s(1) should return then?

s().writeln; // [[0, 1], [2, 3]] instead of [[0, 2], [1, 3]]
s(1).writeln; // [0, 2] instead of [[0,2]]

from mir.

Anyway $ is not defined for opCall

You could also overload with tuples and replace the $ with -1

s = iota(9).sliced(3, 3);
// 0 1 2
// 3 4 5
// 6 7 8

// 0 3 6
// 1 4 7
// 2 5 8
s(tuple(1,2), tuple(1, 2)); // [[4,7], [5,8]]
s( tuple(1,-1), tuple(1, -1)); // same

Oh sry - made the brackets for s(1) by mistake.

For s() I am not sure yet, but by looking at the examples with everted above, I notice two things

b.everted.reshape(2, 3).everted
b().reshape(2, 3)(); // similar, but definitely harder to read

1. the opCall syntax is quite hard to distinguish/read and to understand whats going on. Why not add a simple method, e.g. b.F(2, 3) (probably better naming required though)

2. The main point of this issue was that in the examples the math index order is not stored, leading that b.everted.reshape(2, 3).everted evaluates differently than maybe expected and one needs to make a copy of the array..

from mir.

You could also overload with tuples and replace the $ with -1

-1 would significantly reduce performance.

from mir.

-1 would significantly reduce performance.

and if i think about it one couldn't do stuff like (2, $-3)

from mir.

duplicate of #10

from mir.