First try at making visibility=hidden the default.

theano/gof/cmodule.py

@@ -150,7 +150,7 @@ class DynamicModule(object):
 
         self.support_code = []
         self.functions = []
-        self.includes = ["<Python.h>", "<iostream>"]
+        self.includes = ["<Python.h>", "<iostream>", '"theano_mod_helper.h"']
         self.init_blocks = []
 
     def print_methoddef(self, stream):
@@ -171,14 +171,14 @@ static struct PyModuleDef moduledef = {{
       MyMethods,
 }};
 """.format(name=self.hash_placeholder), file=stream)
-            print(("PyMODINIT_FUNC PyInit_%s(void) {" %
+            print(("THEANO_INIT_FUNC PyInit_%s(void) {" %
                               self.hash_placeholder), file=stream)
             for block in self.init_blocks:
                 print('  ', block, file=stream)
             print("    PyObject *m = PyModule_Create(&moduledef);", file=stream)
             print("    return m;", file=stream)
         else:
-            print(("PyMODINIT_FUNC init%s(void){" %
+            print(("THEANO_INIT_FUNC init%s(void){" %
                               self.hash_placeholder), file=stream)
             for block in self.init_blocks:
                 print('  ', block, file=stream)
@@ -1449,7 +1449,8 @@ def std_include_dirs():
     py_plat_spec_inc = distutils.sysconfig.get_python_inc(plat_specific=True)
     python_inc_dirs = ([py_inc] if py_inc == py_plat_spec_inc
                        else [py_inc, py_plat_spec_inc])
-    return numpy_inc_dirs + python_inc_dirs
+    gof_inc_dir = os.path.dirname(__file__)
+    return numpy_inc_dirs + python_inc_dirs + [gof_inc_dir]
 
 
 def std_lib_dirs_and_libs():
@@ -1929,7 +1930,7 @@ class GCC_compiler(Compiler):
     @staticmethod
     def compile_str(module_name, src_code, location=None,
                     include_dirs=None, lib_dirs=None, libs=None,
-                    preargs=None, py_module=True):
+                    preargs=None, py_module=True, hide_symbols=True):
         """
         :param module_name: string (this has been embedded in the src_code
 
@@ -1952,6 +1953,10 @@ class GCC_compiler(Compiler):
         :param py_module: if False, compile to a shared library, but do not
             import it as a Python module.
 
+        :param hide_symbols: if True (the default) all symbols will be
+        hidden from the library symbol table (which means that other
+        objects can't use them.
+
         :returns: dynamically-imported python module of the compiled code.
             (unless py_module is False, in that case returns None.)
         """
@@ -2006,6 +2011,14 @@ class GCC_compiler(Compiler):
         else:
             cmd.extend(preargs)
         cmd.extend('-I%s' % idir for idir in include_dirs)
+        if hide_symbols and sys.platform != 'win32':
+            # This has been available since gcc 4.0 so we suppose it
+            # is always available. We pass it here since it
+            # significantly reduces the size of the symbol table for
+            # the objects we want to share. This in turns leads to
+            # improved loading times on most platforms (win32 is
+            # different, as usual).
+            cmd.append('-fvisibility=hidden')
         cmd.extend(['-o', lib_filename])
         cmd.append(cppfilename)
         cmd.extend(['-L%s' % ldir for ldir in lib_dirs])

theano/gof/cutils.py

@@ -188,6 +188,7 @@ def compile_cutils():
     code = ("""
         #include <Python.h>
         #include "numpy/arrayobject.h"
+        #include "theano_mod_helper.h"
 
         extern "C"{
         static PyObject *
@@ -236,7 +237,7 @@ def compile_cutils():
             CutilsExtMethods,
         };
 
-        PyMODINIT_FUNC
+        THEANO_INIT_FUNC
         PyInit_cutils_ext(void) {
             import_array();
             return PyModule_Create(&moduledef);
@@ -245,7 +246,7 @@ def compile_cutils():
         """
     else:
         code += """
-        PyMODINIT_FUNC
+        THEANO_INIT_FUNC
         initcutils_ext(void)
         {
           import_array();

theano/gof/lazylinker_c.c

 #include <Python.h>
+#include "theano_mod_helper.h"
 #include "structmember.h"
 #include <sys/time.h>
 
-// Old Python compatibility from here:
-// http://www.python.org/dev/peps/pep-0353/
-#if PY_VERSION_HEX < 0x02050000 && !defined(PY_SSIZE_T_MIN)
-typedef int Py_ssize_t;
-#define PY_SSIZE_T_MAX INT_MAX
-#define PY_SSIZE_T_MIN INT_MIN
-// This one was taken from:
-// http://svn.python.org/projects/python/trunk/Modules/_ctypes/ctypes.h
-#define PyNumber_AsSsize_t(ob, exc) PyInt_AsLong(ob)
-#endif
-
 #if PY_VERSION_HEX >= 0x03000000
 #include "numpy/npy_3kcompat.h"
 #define PyCObject_AsVoidPtr  NpyCapsule_AsVoidPtr
@@ -1033,10 +1023,6 @@ static PyMethodDef lazylinker_ext_methods[] = {
   {NULL, NULL, 0, NULL}        /* Sentinel */
 };
 
-#ifndef PyMODINIT_FUNC  /* declarations for DLL import/export */
-#define PyMODINIT_FUNC void
-#endif
-
 #if defined(NPY_PY3K)
 static struct PyModuleDef moduledef = {
         PyModuleDef_HEAD_INIT,
@@ -1052,11 +1038,11 @@ static struct PyModuleDef moduledef = {
 #endif
 #if defined(NPY_PY3K)
 #define RETVAL m
-PyMODINIT_FUNC
+THEANO_INIT_FUNC
 PyInit_lazylinker_ext(void) {
 #else
 #define RETVAL
-PyMODINIT_FUNC
+THEANO_INIT_FUNC
 initlazylinker_ext(void) 
 {
 #endif
@@ -1076,4 +1062,3 @@ initlazylinker_ext(void)
 
     return RETVAL;
 }
-

theano/gof/theano_mod_helper.h

+#ifndef THEANO_MOD_HELPER
+#define THEANO_MOD_HELPER
+
+#ifndef _WIN32
+#define MOD_PUBLIC __attribute__((visibility ("default")))
+#else
+#define MOD_PUBLIC
+#endif
+
+#define THEANO_INIT_FUNC MOD_PUBLIC PyMODINIT_FUNC
+
+#endif

theano/sandbox/cuda/cuda_ndarray.cu

@@ -2,6 +2,7 @@
 
 #include <Python.h>
 #include <structmember.h>
+#include "theano_mod_helper.h"
 
 #include <numpy/arrayobject.h>
 #include <iostream>
@@ -3475,10 +3476,6 @@ static PyMethodDef module_methods[] = {
     {NULL, NULL, NULL, NULL}  /* Sentinel */
 };
 
-#ifndef PyMODINIT_FUNC  /* declarations for DLL import/export */
-#define PyMODINIT_FUNC void
-#endif
-
 #define CNDA_MOD_NAME "cuda_ndarray"
 #define CNDA_DOCSTRING "CUDA implementation of a numpy ndarray-like object."
 
@@ -3493,10 +3490,10 @@ static struct PyModuleDef cuda_ndarray_moduledef =
     module_methods
 };
 
-PyMODINIT_FUNC
+THEANO_INIT_FUNC
 PyInit_cuda_ndarray(void)
 #else
-PyMODINIT_FUNC
+THEANO_INIT_FUNC
 initcuda_ndarray(void)
 #endif
 {

theano/sandbox/cuda/cuda_ndarray.cuh

@@ -10,6 +10,8 @@
 
 #include "numpy/npy_3kcompat.h"
 
+#include "theano_mod_helper.h"
+
 // Py3k strings are unicode, these mimic old functionality.
 //
 // NOTE: npy_3kcompat.h replaces PyString_X with PyBytes_X, which breaks
@@ -53,7 +55,7 @@
 #endif
 #define ALWAYS_INLINE
 #else //else _WIN32
-#define DllExport
+#define DllExport  MOD_PUBLIC
 #define ALWAYS_INLINE __attribute__((always_inline))
 #endif
 

theano/sandbox/cuda/nvcc_compiler.py

@@ -208,7 +208,7 @@ class NVCC_compiler(Compiler):
     def compile_str(
             module_name, src_code,
             location=None, include_dirs=[], lib_dirs=[], libs=[], preargs=[],
-            rpaths=rpath_defaults, py_module=True):
+            rpaths=rpath_defaults, py_module=True, hide_symbols=True):
         """:param module_name: string (this has been embedded in the src_code
         :param src_code: a complete c or c++ source listing for the module
         :param location: a pre-existing filesystem directory where the
@@ -225,6 +225,9 @@ class NVCC_compiler(Compiler):
         :param py_module: if False, compile to a shared library, but
             do not import as a Python module.
 
+        :param hide_symbols: if True (the default), hide all symbols
+        from the library symbol table unless explicitely exported.
+
         :returns: dynamically-imported python module of the compiled code.
             (unless py_module is False, in that case returns None.)
 
@@ -320,6 +323,9 @@ class NVCC_compiler(Compiler):
             # in both math_functions.h and pymath.h,
             # by not including the one in pymath.h
             cmd.extend(['-D HAVE_ROUND'])
+        else:
+            if hide_symbols:
+                preargs2.append('-fvisibility=hidden')
 
         if local_bitwidth() == 64:
             cmd.append('-m64')

theano/scan_module/scan_perform_ext.py

@@ -109,7 +109,8 @@ except ImportError:
                     preargs.append("-D NPY_F_CONTIGUOUS=NPY_ARRAY_F_CONTIGUOUS")
 
             cmodule.GCC_compiler.compile_str(dirname, code, location=loc,
-                                             preargs=preargs)
+                                             preargs=preargs,
+                                             hide_symbols=False)
             # Save version into the __init__.py file.
             init_py = os.path.join(loc, '__init__.py')
             open(init_py, 'w').write('_version = %s\n' % version)