On: ngn/k, an AGPL K interpreter

[geocar]

Cool!

I've often thought about the best way to do this, and your suggestion is compelling. I'm going to try it out.

[bakul → geocar]

:-) The macrology is a bit grody and my DO implementation doesn't extend past 3 args.

[geocar → bakul]

I'm interested to know what you think is "grody". This is what I came up with:

    #define DO(...) for(long long _a[]={__VA_ARGS__,1,1,1,1},i=0;i<*_a;++i) \
      for(long long j=0;j<_a[1];++j)for(long long k=0;k<_a[2];++k)

It doesn't seem bad to me, and it'd be pretty easy to extend or add asserts to catch when you introduce a fourth arg.

gcc also appears smart enough to remove the unnecessary loads, so I'm going to keep playing with it.

[bakul → geocar]

Clever use of `__VA_ARGS__`! Does gcc remove `_a` as well? Well, may be it was just my implementation:

  #define mkmacro(func, ...) mkmacro_(func, VA_NUM_ARGS(__VA_ARGS__))
  #define mkmacro_(func, nargs) mkmacro__(func, nargs)
  #define mkmacro__(func, nargs) func ## nargs
  #define DO(...) mkmacro(DO, __VA_ARGS__)(__VA_ARGS__)
  #define DO1(n) for(I i=0;i<(n); i++)
  #define DO2(m,n) DO1(m) for(I j=0;j<(n);j++)
  #define DO3(l,m,n) DO2(l,m) for(I k=0;k<(n);k++)

[bakul → bakul]

I just noticed two missing defines. Sorry about that. Fixed below.

I did some testing and seems like this generates much less code at least with clang (with -O3 -fomit-frame-pointer) than geocar's much briefer solution!

  #define VA_NUM_ARGS(...) \
    VA_NUM_ARGS_IMPL(__VA_ARGS__,5,4,3,2,1)
  #define VA_NUM_ARGS_IMPL(_1,_2,_3,_4,_5,N,...) N
  #define mkmacro(func, ...) \
    mkmacro_(func, VA_NUM_ARGS(__VA_ARGS__))
  #define mkmacro_(func, nargs) mkmacro__(func, nargs)
  #define mkmacro__(func, nargs) func ## nargs

  #define DO(...) mkmacro(DO, __VA_ARGS__)(__VA_ARGS__)
  #define DO1(n) for(I i=0;i<(n); i++)
  #define DO2(m,n) DO1(m) for(I j=0;j<(n);j++)
  #define DO3(l,m,n) DO2(l,m) for(I k=0;k<(n);k++)