7

I'm trying the fold expression with clang 3.6 '--std=c++1z', but something I don't quite get. The function that I'm testing is:

auto minus = [](auto... args) { return (args - ...); };
...
std::cout << minus(10, 3, 2) << std::endl;

according to n4191, I'm expecting it expands as a left fold to

(10 - 3) - 2

which gives result 5, however, the result is 9, which seems to be a right fold expansion, i.e.

10 - (3 - 2)

Am I missing anything or mis-understand n4191? Thanks

interjay
  • 107,303
  • 21
  • 270
  • 254
Ralph Zhang
  • 5,015
  • 5
  • 30
  • 40

1 Answers1

6

n4191 was revised by n4295. According to that, an expression of the form (e op ...) is a unary right fold, and that is expanded as: E1 op (... op (EN-1 op EN)), i.e. as a right fold expansion.

This does seem to be the reverse of what n4191 stated in terms of the fold direction. Clang 3.6 implements the n4295 proposal, as shown here.

... - args would be a unary left fold and expand in the direction you want.

interjay
  • 107,303
  • 21
  • 270
  • 254