Subj : JS_InitClass segfaults in my JS/Python bridge
To   : netscape.public.mozilla.jseng
From : jjl@pobox.com (John J. Lee)
Date : Fri Aug 01 2003 04:07 pm

I'm making a JavaScript / Python bridge, and I'm having trouble
dynamically creating proxy JS classes for Python classes that get
bound to JS.  I'm embedding JS in Python rather than the other way
round.  I've looked at both perlconnect (both halves) and Claes
Jacobssen's Javascript.xs, and stared at the docs, and I'm none the
wiser.  I think I must be doing something really stupid, but I can't
see it.  I'm a complete JS newbie.

(I'm using spidermonkey 1.5-rc5, libjs.so, on linux.)

JS and Python are pretty similar in many ways (JS property <--> Python
attribute; also, methods are just properties/attributes in both
languages), so the idea is not to define any methods or properties at
JS_InitClass time, but instead use my own setProperty / getProperty
functions to look up the corresponding Python attributes.

Now, the JS_InitClass docs say:

| If you provide a constructor for the class, then you should also pass
| an object to parent_proto.

but I've seen several people do (note third and fifth args):

// set up my_class, then
JSObject *obj = JS_InitClass(
    cx, global_obj, NULL, &my_class, my_constructor, 0,
    NULL, NULL, NULL, NULL);

apparently without harm.  That's what I'm trying to do (I don't see
any need ATM for the proxies to have a special prototype), but I get a
segfault or illegal instruction inside JS_InitClass.  In fact, I get
that no matter what I do -- I've tried using a prototype object of my
own created with JS_NewObject, using the global object, and using
NULL.  They all segfault or cause illegal instructions.

Any clues?  I've reduced the problem to the boilerplate setting up of
runtime and context, then setting up a JSClass and calling
JS_InitClass, so I'm fairly confident the problem is in the function
below.

[In case anybody feels like looking at it, the complete code (
stripped down to minimal 120 lines + extern type declarations that
causes the segfault) is here:

http://wwwsearch.sourceforge.net/spidermonkey.tar.gz

Unfortunately, the test.py in there doesn't happen to trigger the
segfault... In the unlikely event that anyone who wants to try
building it and getting the segfault, let me know and I'll put up the
code that triggers it.]

This is Pyrex code -- essentially Python with C types, compiles to C.
But don't panic!  The cdef class is really just a C struct (as far as
C is concerned), '->' operators are written as '.', casts (type)object
are written as <type>object, there are no semicolons and comments
start with a '#'.

Pyrex does treat python objects specially, but there's nothing too
tricky there: it compiles len(name), klass.__name__, the use of name
in strcpy etc. into appropriate Python C API calls (including
refcounting, but there's no refcounting worthy of note in the function
below).


cdef class Class:
    # struct members
    cdef Context context  # another struct, containing the JSContext * etc.
    cdef JSClass *jsc
    cdef JSObject *base_obj
    cdef object klass  # Python class (PyObject *) -- not relevant here

    # Define a special Python method (don't worry your pretty head about it ;).
    # It's just a C function that gets called from Python.
    def __new__(self, Context context, flags, klass):
        self.jsc = <JSClass *> xmalloc(sizeof(JSClass))

        name = klass.__name__
        self.jsc.name = <char *> xmalloc((len(name)+1)*sizeof(char))
        strcpy(self.jsc.name, name)

        self.jsc.flags = flags
        self.jsc.addProperty = JS_PropertyStub
        self.jsc.delProperty = JS_PropertyStub
        self.jsc.getProperty = JS_PropertyStub  # will implement later
        self.jsc.setProperty = JS_PropertyStub  # ditto
        self.jsc.enumerate = JS_EnumerateStub
        self.jsc.resolve = JS_ResolveStub
        self.jsc.convert = JS_ConvertStub
        self.jsc.finalize = JS_FinalizeStub

        # context.cx is the JSContext *, context.globj is the global object
	self.base_obj = JS_InitClass(
            context.cx, context.globj,
            NULL,  # !!! no matter what this is, JS_InitClass crashes
            self.jsc, constructor_stub, 0,
            NULL, NULL, NULL, NULL)


Since Python constructors have variable numbers of arguments, I'm not
sure what that 0 should be, but that can't be causing the problems I'm
having ATM.

Heeeelllp...


John

.