Proper namespacing for dtype and stype in C API

[v0dro]

Currently the enum defining stype and dtype for an nmatrix does not namespace it's values.

Prefixing every value with NM_ would suffice. @mohawkjohn thoughts?

[translunar]

You're mistaken. It namespaces them in C++, and in C it adds the nm_ prefix:

• https://github.com/SciRuby/nmatrix/blob/master/ext/nmatrix/nmatrix.h#L193
• https://github.com/SciRuby/nmatrix/blob/master/ext/nmatrix/nmatrix.h#L225

[v0dro]

No what I'm saying is that the contents of the enum should be prefixed too. So it should be NM_FLOAT64 not FLOAT64.

A quick program after compiling with gcc's -E option yields this output: Program:

#include <stdio.h>
#define NM_DEF_ENUM(name, ...)     \
  typedef enum nm_ ## name {       \
    __VA_ARGS__                    \
  } nm_ ## name;

NM_DEF_ENUM(dtype_t,    BYTE                =  0,  // unsigned char
                        INT8                =  1,  // char
                        INT16               =  2,  // short
                        INT32               =  3,  // int
                        INT64               =  4,  // long
                        FLOAT32         =  5,  // float
                        FLOAT64         =  6,  // double
                        COMPLEX64       =  7,  // Complex64 class
                        COMPLEX128  =  8,  // Complex128 class
                        RUBYOBJ         = 9);  // Ruby VALUE type

int main()
{ 
  nm_dtype_t e;

  e = BYTE;
  printf("BYTE : %d\n", e);
  return 0;
}

Output:

typedef enum nm_dtype_t { BYTE = 0, INT8 = 1, INT16 = 2, INT32 = 3, INT64 = 4, FLOAT32 = 5, FLOAT64 = 6, COMPLEX64 = 7, COMPLEX128 = 8, RUBYOBJ = 9 } nm_dtype_t;
                                            ;
int main()
{
  nm_dtype_t e;

  e = BYTE;
  printf("BYTE : %d\n", e);
  return 0;
}

Having NM_ everywhere would avoid clashes with other libraries who might use data type names similar to ours.

[translunar]

Why? It's in a namespace?

[v0dro]

Not sure what you mean. What's in a namespace?

The terms that are inside the enumerator (like BYTE and INT32) are not namespaced.

[translunar]

Yes they are. It's a macro which puts them in a namespace for c++ or gives them prefixes for C. On Sat, Feb 20, 2016 at 12:06 AM Sameer Deshmukh notifications@github.com wrote:

Not sure what you mean. What's in a namespace?

The terms that are inside the enumerator (like BYTE and INT32) are not namespaced.

— Reply to this email directly or view it on GitHub https://github.com/SciRuby/nmatrix/issues/468#issuecomment-186519213.

[v0dro]

In my program I was able to access them as BYTE or FLOAT64 (see main() function above).

[translunar]

Oh, hmm. You may actually be right, and I'm not sure how that happened if so. Sorry. ^^;;

Just to check, can you figure out how to generate the post-processed sources using gcc and confirm this? You're looking for a .i file, I think.

[v0dro]

You mean for the whole gem? Won't that be complex to sift through? Copy pasting the macros from nmatrix.h into another file and compiling with -E would be easier if I'm getting at what you're getting at.

[translunar]

Uh, no. Just add the gcc flag for outputting preprocessed files to extconf, and go look in the build directory.