1. Build NumPy. You can either build NumPy specficially:

   make numpy
   

   or build NumPy for a specific platform:

   make numpy-iOS
   

   or build all supported binary packages:

   make app_packages
   

   This will produce:

   • a folder named dist/app_packages, containing the python code required by the package
   • a folder for each supported mobile platform (iOS, tvOS and watchOS) containing the static fat binary libraries needed to support the Python code.

2. Copy the contents of dist/app_packages into your project's site_packages or app_packages directory. This will make the Python library available to your project.

3. Copy the static binary libraries in the platform directory (e.g., the contents of dist/iOS) into your project and add them as static libraries in your project. The location where you copy the files doesn't matter - they just need to be part of the project. If you're using a BeeWare template, we'd suggest putting them in the Support folder.

4. Add the following function definition to the file that configures your Python environment (if you're using a BeeWare template, this will be the main.m file; in other projects, it's whichever file contains the code that invokes Py_Initialize() and PyEval_InitThreads():

   void numpy_importer() {
       PyRun_SimpleString(
           "import sys, importlib\n" \
           "class NumpyImporter(object):\n" \
           "    def find_module(self, fullname, mpath=None):\n" \
           "        if fullname in (" \
           "                    'numpy.core.multiarray', " \
           "                    'numpy.core.umath', " \
           "                    'numpy.fft.fftpack_lite', " \
           "                    'numpy.linalg._umath_linalg', " \
           "                    'numpy.linalg.lapack_lite', " \
           "                    'numpy.random.mtrand', " \
           "                ):\n" \
           "            return self\n" \
           "        return\n" \
           "    def load_module(self, fullname):\n" \
           "        f = '__' + fullname.replace('.', '_')\n" \
           "        mod = sys.modules.get(f)\n" \
           "        if mod is None:\n" \
           "            mod = importlib.__import__(f)\n" \
           "            sys.modules[fullname] = mod\n" \
           "            return mod\n" \
           "        return mod\n" \
           "sys.meta_path.append(NumpyImporter())"
       );
   }
   

5. Add the following external function declarations to the file that configures your Python enviroment:

   extern PyMODINIT_FUNC PyInit_multiarray(void);
   extern PyMODINIT_FUNC PyInit_umath(void);
   extern PyMODINIT_FUNC PyInit_fftpack_lite(void);
   extern PyMODINIT_FUNC PyInit__umath_linalg(void);
   extern PyMODINIT_FUNC PyInit_lapack_lite(void);
   extern PyMODINIT_FUNC PyInit_mtrand(void);
   

6. Add the following function calls before invoking Py_Initialize():

   PyImport_AppendInittab("__numpy_core_multiarray", &PyInit_multiarray);
   PyImport_AppendInittab("__numpy_core_umath", &PyInit_umath);
   PyImport_AppendInittab("__numpy_fft_fftpack_lite", &PyInit_fftpack_lite);
   PyImport_AppendInittab("__numpy_linalg__umath_linalg", &PyInit__umath_linalg);
   PyImport_AppendInittab("__numpy_linalg_lapack_lite", &PyInit_lapack_lite);
   PyImport_AppendInittab("__numpy_random_mtrand", &PyInit_mtrand);
   

7. Install the numpy importer, by invoking numpy_importer() after invoking PyEval_InitThreads(), but before you run any Python scripts.

If you've followed these instructions, and you're using a BeeWare project template, your iOS/<project name>/main.m file (in the Supporting Files folder in the Xcode project) will look something like this:

//
//  main.m
//  A main module for starting Python projects under iOS.
//

#import <Foundation/Foundation.h>
#import <UIKit/UIKit.h>
#include <Python.h>
#include <dlfcn.h>


void numpy_importer() {
    PyRun_SimpleString(
        "import sys, importlib\n" \
...
        "sys.meta_path.append(NumpyImporter())"
    );
}


extern PyMODINIT_FUNC PyInit_multiarray(void);
extern PyMODINIT_FUNC PyInit_umath(void);
extern PyMODINIT_FUNC PyInit_fftpack_lite(void);
extern PyMODINIT_FUNC PyInit__umath_linalg(void);
extern PyMODINIT_FUNC PyInit_lapack_lite(void);
extern PyMODINIT_FUNC PyInit_mtrand(void);


int main(int argc, char *argv[]) {
    int ret = 0;
    unsigned int i;
    NSString *tmp_path;
    NSString *python_home;
    NSString *python_path;
    wchar_t *wpython_home;
    const char* main_script;
    wchar_t** python_argv;

    @autoreleasepool {
...

        // iOS provides a specific directory for temp files.
        tmp_path = [NSString stringWithFormat:@"TMP=%@/tmp", resourcePath, nil];
        putenv((char *)[tmp_path UTF8String]);

        PyImport_AppendInittab("__numpy_core_multiarray", &PyInit_multiarray);
        PyImport_AppendInittab("__numpy_core_umath", &PyInit_umath);
        PyImport_AppendInittab("__numpy_fft_fftpack_lite", &PyInit_fftpack_lite);
        PyImport_AppendInittab("__numpy_linalg__umath_linalg", &PyInit__umath_linalg);
        PyImport_AppendInittab("__numpy_linalg_lapack_lite", &PyInit_lapack_lite);
        PyImport_AppendInittab("__numpy_random_mtrand", &PyInit_mtrand);

        NSLog(@"Initializing Python runtime");
        Py_Initialize();

...

        // If other modules are using threads, we need to initialize them.
        PyEval_InitThreads();

        numpy_importer();

        // Start the main.py script
        NSLog(@"Running %s", main_script);

...
    }

    exit(ret);
    return ret;
}