How is the at sign (@) used in the c source code for numpy?

Question

I've been perusing some of the source code for numpy and I noticed that a lot of the c source code uses the construction @variablename@. For example, in the file "npy_math_complex.c.src" (located here):

/*==========================================================
* Constants
*=========================================================*/
static const @ctype@ c_1@c@ = {1.0@C@, 0.0};
static const @ctype@ c_half@c@ = {0.5@C@, 0.0};
static const @ctype@ c_i@c@ = {0.0, 1.0@C@};
static const @ctype@ c_ihalf@c@ = {0.0, 0.5@C@};

What do @ctype@ and @c@ mean? Are these some sort of macros? I'm guessing they aren't normal C-macros since I've looked at the relevant header files listed in the file and they don't appear to define any macros using "@".

Is @name@ some sort of macro used by distutils when compiling c code into a python module?

I've never seen an @ sign used in c code before and so I'm a bit confused...

Answer 1

It's because these files are templates. If I remember correctly NumPy uses several template engines (thanks @user2357112 for helping me finding the appropriate one):

• tempita (for Cython files)
• another one for C files
• another one for FORTRAN files

and the second one is actually responsible to convert these to "regular" C files - before these get compiled.

Basically these functions will be cloned a lot of times and for each of the functions a special placeholder is inserted between the %.

For example in this case it begins with:

/**begin repeat
 * #type = npy_float, npy_double, npy_longdouble#
 * #ctype = npy_cfloat,npy_cdouble,npy_clongdouble#
 * #c = f, , l#
 * #C = F, , L#
 * ....
 */

So in the first iteration @ctype@ will be replaced with npy_cfloat and @c@ with f and @C@ with F:

static const npy_cfloat c_1f = {1.0F, 0.0};
static const npy_cfloat c_halff = {0.5F, 0.0};
static const npy_cfloat c_if = {0.0, 1.0F};
static const npy_cfloat c_ihalff = {0.0, 0.5F};

In the next iteration is @ctype@ npy_cdouble, ...

static const npy_cdouble c_1 = {1.0, 0.0};
static const npy_cdouble c_half = {0.5, 0.0};
static const npy_cdouble c_i = {0.0, 1.0};
static const npy_cdouble c_ihalf = {0.0, 0.5};

and in the third iteration:

static const npy_clongdouble c_1l = {1.0L, 0.0};
static const npy_clongdouble c_halfl = {0.5L, 0.0};
static const npy_clongdouble c_il = {0.0, 1.0L};
static const npy_clongdouble c_ihalfl = {0.0, 0.5L};

these are then compiled as normal C files.