Re: libseed-list js extensions for native modules

[Jonatan Liljedahl <lijon kymatica com>]

I reverted my seed-sandbox.c to current git, but I had only this small 
change which I thought was a typo fix:

--- a/modules/sandbox/seed-sandbox.c
+++ b/modules/sandbox/seed-sandbox.c
@@ -126,7 +126,7 @@ seed_static_function context_funcs[] = {
   {"eval", seed_context_eval, 0},
   {"add_globals", seed_sandbox_context_add_globals, 0},
   {"destroy", seed_sandbox_context_destroy, 0},
-  {"get_global_object", seed_sandbox_context_get_global_object },
+  {"get_global_object", seed_sandbox_context_get_global_object, 0},
   {NULL, NULL, 0}
 };

Anyhow, now it loads the sandbox module...

But I still can't add to the prototype:

> imports.sandbox.Context.prototype.foo = function() { print("test") };
function () { print("test"); }
> c = new imports.sandbox.Context
[object Context]
> c.foo()
TypeError Result of expression 'c.foo' [undefined] is not a function.
> c.eval("1+2")
3

Could this be a difference in JSC version? I'm with webkit 1.1.15.2-1 in 
ubuntu.

/Jonatan

Alan Knowles wrote:
> It failed to load the sandbox module by the looks of things.
>
> It's probably worth running with --seed-debug=all to find out what's going on. - if you built seed with debug on.
>
> Regards
> Alan
>
>
>  --- On 29/Jun/2010, Jonatan Liljedahl wrote: 
> > Doesn't work here, but now something else is strange (first try to get 
> > sandbox failed):
> >
> >  > imports.sandbox.Context.prototype.foo = function() { print("test") };
> > function () { print("test"); }
> >  > c = new imports.sandbox.Context
> > TypeError Result of expression 'imports.sandbox' [undefined] is not an 
> > object.
> >  > c = new imports.sandbox.Context
> > [object Context]
> >  > c.foo()
> > TypeError Result of expression 'c.foo' [undefined] is not a function.
> >  >
> >
> >
> > Alan Knowles wrote:
> > > Did I miss something - this works perfectly..
> > >
> > >
> > > GLib = imports.gi.GLib;
> > >
> > > imports.sandbox.Context.prototype.eval_file = function(fn) { 
> > >         var  script = {}; 
> > >         print("load " + fn);
> > >         GLib.file_get_contents(fn,script); 
> > >         return this.eval(script.contents); 
> > > };
> > >
> > > c = new imports.sandbox.Context;
> > > c.eval_file("test.js");
> > >
> > > #output>>>
> > > load test.js
> > >
> > > ** (seed:3644): CRITICAL **: Line 10 in /tmp/seedtest.js: GFileError Failed to open file 'test.js': No such file or directory
> > >
> > > Regards
> > > Alan
> > >
> > >
> > >
> > >  --- On 29/Jun/2010, Jonatan Liljedahl wrote: 
> > > > > This should be made to work - It's probably going to take a bit of
> > > >  > working out how to support it...
> > > >
> > > > I'm trying to understand how this works, am I right that native modules 
> > > > create their classes with seed_create_class() (JSClassCreate()), while 
> > > > the Gir importer sets up an object with a 'prototype' property "by hand"?
> > > >
> > > >  From the JSC API docs:
> > > >
> > > > "Standard JavaScript practice calls for storing function objects in 
> > > > prototypes, so they can be shared. The default JSClass created by 
> > > > JSClassCreate follows this idiom, instantiating objects with a shared, 
> > > > automatically generating prototype containing the class's function 
> > > > objects. The kJSClassAttributeNoAutomaticPrototype attribute specifies 
> > > > that a JSClass should not automatically generate such a prototype."
> > > >
> > > > So, the question is why the prototype is not writable and if it can be 
> > > > made to be that. I tried this hackish function on the constructor but it 
> > > > didn't work:
> > > >
> > > > gboolean
> > > > seed_make_extendable (JSContextRef ctx, JSObjectRef object)
> > > > {
> > > >    JSStringRef jname = JSStringCreateWithUTF8CString ("prototype");
> > > >    JSValueRef exception = NULL;
> > > >    JSValueRef proto = JSObjectGetPrototype(ctx, object);
> > > >    if (proto)
> > > >      {
> > > >        JSObjectSetProperty (ctx, (JSObjectRef) object, jname, proto,
> > > >        kJSPropertyAttributeDontEnum|kJSPropertyAttributeDontDelete,
> > > > 			   &exception);
> > > >      }
> > > >
> > > >    JSStringRelease (jname);
> > > >    return TRUE;
> > > > }
> > > >
> > > > Once JSClassCreate has made the prototype, it seems to be fixed and 
> > > > read-only.
> > > >
> > > > Perhaps one should create classes more "manually" like the Gir importer 
> > > > does instead of using JSClassCreate()?
> > > >
> > > > > Looking through the code, it's either we have to explicitly overlay
> > > > > the prototype properties in the constructor, or we set the
> > > > > parent_class of Context to a standard object..
> > > > parent_class can only be a JSClassRef, and it says it should be NULL for 
> > > > the normal Object class, which I guess it should be. Seems like 
> > > > JavaScriptCore has classic inherited OOP internally, and that 
> > > > parent_class has nothing to do with prototype. (but that all objects has 
> > > > Object as parent_class)
> > > >
> > > > I'm not sure I understand what you mean by overlay the prototype 
> > > > properties..
> > > >
> > > > /Jonatan
> > > >
> > > > > Regarsd Alan
> > > > >
> > > > > --- On 21/Jun/2010, Jonatan Liljedahl wrote:
> > > > > > Also, it should work to add methods to SomeClass.prototype in
> > > > > > native modules. Currently it does not:
> > > > > >
> > > > > > imports.sandbox.Context.prototype.eval_file = function(fn) { var
> > > > > > script = {}; GLib.file_get_contents(fn,script); return
> > > > > > this.eval(script.contents); };
> > > > > >
> > > > > > c = new imports.sandbox.Context; c.eval_file(foobar);
> > > > > >
> > > > > > TypeError Result of expression 'ctx.eval_file' [undefined] is not a
> > > > > >  function.
> > > > > >
> > > > > > It does work for GI modules, as seen in extensions/Gtk.js
> > > > > >
> > > > > > /Jonatan
> > > > > >
> > > > > > Jonatan Liljedahl wrote:
> > > > > > > The gi_importer tries to import extensions/name.js by evaluating
> > > > > > >  "imports.extensions.name". It would be nice if this also worked
> > > > > > > for native modules, I imagine it to be quite common that you
> > > > > > > write a module with core parts in C and some high-level stuff in
> > > > > > > js.
> > > > > > >
> > > > > > > The question is, how should this be handled regarding searchpaths
> > > > > > > etc?
> > > > > > >
> > > > > > > If your module is installed in /usr/local/lib/seed, then you
> > > > > > > probably installed the extension js in
> > > > > > > /usr/local/share/seed/extensions, and imports.extensions would be
> > > > > > > the right thing.
> > > > > > >
> > > > > > > But what happens if your module is installed somewhere else? It
> > > > > > > should probably start with the same dir as the native module,
> > > > > > > then go on with imports.searchPath, and if the file is found call
> > > > > > >  seed_importer_handle_file() on it..
> > > > > > >
> > > > > > > BTW, what happens if you have 'extensions' folder in more than
> > > > > > > one search path? will extensions.foo search for 'foo' in all of
> > > > > > > those or only the first match for 'extensions'? Just doing a
> > > > > > > 'imports.extensions' will of course create only one dir_importer
> > > > > > > for that folder, hiding possible 'extensions' folders later in
> > > > > > > the searchpath. I don't know if this is a good thing...
> > > > > > >
> > > > > > > /Jonatan _______________________________________________ 
> > > > > > > libseed-list mailing list libseed-list gnome org 
> > > > > > > http://mail.gnome.org/mailman/listinfo/libseed-list
> > > > > > _______________________________________________ libseed-list
> > > > > > mailing list libseed-list gnome org 
> > > > > > http://mail.gnome.org/mailman/listinfo/libseed-list
> > > > > _______________________________________________ libseed-list mailing
> > > > > list libseed-list gnome org 
> > > > > http://mail.gnome.org/mailman/listinfo/libseed-list