I work on several projects that have initialization
requirements that
need to grab control after Py_Initialize(), but before
any user code
runs (via input, script, -c, etc...).

Note that these are Python clones that take advantage
of an installed
python (using its $prefix/lib/pythonx.x/*.py and
site-packages/*)

We could use 

PyImport_AppendInittab("sitecustomize",initsitecustomize);

But if there already IS customization in
sitecustomize.py, I've
blown it away (and have to look it up and force an
import).
And if someone uses the -S flag, I'm screwed.

I propose a hook styled after Py_AtExit(func) called
Py_AtInit(func)
which maintains a list of functions that are called in
Py_Initialize
right after main and site initializations.

If the hook isn't used, then the cost is a single extra
function
call at initialization.  Here's a spurious example:  A
customer wants
a version of python that has all the math functions and
his
extensions to act like builtins...

I would write (without refcnt or error checks ;-):

#include "Python.h"
static void after_init(void) {
    PyObject
*builtin,*builtin_dict,*math,*math_dict,*user,*user_dict;

    builtin = PyImport_ImportModule("__builtin__");
    builtin_dict = PyModule_GetDict(builtin);
    math = PyImport_ImportModule("math");
    math_dict = PyModule_GetDict(math);
    user = PyImport_ImportModule("user");
    user_dict = PyModule_GetDict(math);

    PyDict_Update(builtin_dictionary, math_dict);
    PyDict_Update(builtin_dictionary, user_dict);
}


int main(int argc, char** argv) {
    PyImport_AppendInittab("user",inituser);
    Py_AtInit(after_init);

    return Py_Main(argc, argv);
}

voila!  An extended Python with new builtins.
This is vastly better than hacking in through
site.py or sitecustomize

I actually want this to do some MPI initialization to
setup a
single user prompt with broadcast which has to run
after
Py_Initialize() but before the import of readline.

The patch looks good to me in principle. However, I wonder why you run the init functions after importing site; if you want it before any user-defined code, shouldn't you run it before site?

Also, can you please provide documentation patches?

Martin,

You are absolutely right, the call to initinitialize() should occur before
initsite().

I'll prepare a new patch with that change and also include doc updates.

Pat

Where is the new patch? :)

Thanks... I'll submit patches for 2.6, 2.7, and 3.2

On Tue, May 12, 2009 at 2:07 PM, Daniel Diniz <report@bugs.python.org> wrote:
>
> Changes by Daniel Diniz <ajaksu@gmail.com>:
>
>
> ----------
> stage:  -> test needed
> versions: +Python 2.7, Python 3.2 -Python 2.6
>
> _______________________________________
> Python tracker <report@bugs.python.org>
> <http://bugs.python.org/issue730473>
> _______________________________________
>

Patrick, its been 5 years since you promised a 2nd patch (which would now have to be against the current 3.3 tip). If you have lost interested, perhaps we should close this.

Just languishing for lo on these 5 years....