guile-gobject 0.2.0 "The Great Barrier Reef" has been released

[Martin Baulig <martin home-of-linux org>]

Hi guys,

just released the first public version of guile-gobject:

    ftp://ftp.gnome.org/pub/GNOME/earthquake/sources/guile-gobject/

or see the guile-gobject module in GNOME CVS.

And after writing 501 lines of RELEASE-NOTES-0.2.0.txt, here it is:

Release Notes for guile-gobject 0.2.0 "The Great Barrier Reef"
==============================================================

This is the first public release of guile-gobject a new set of
guile bindings for GNOME 2.

Today is my last day of GNOME hacking before I leave for about
four months to learn for the university - so I needed to do this
release even if it's not completely finished and there's almost
no documentation - but I spent every free minute to hack on the
code so there was no time to write much documentation.

To compile and use this stuff, you need a larger set of GNOME 2
modules which you can get at

        ftp://ftp.gnome.org/pub/GNOME/earthquake/

The guile-gobject modules contains the following scheme modules:

(gnome gobject):
================

The (gnome gobject) module is the guile wrapper for GNOME 2's type system,
libgruntime (glib/gobject in GNOME CVS). Well, actually "wrapper" is the
wrong word for it because it's much more than just a wrapper - it does not
only wrap the C library glib/gobject - it gives you full access to GNOME 2's
type system from guile - including support for creating new gruntime types
in guile and then using these types in C.

To understand this overview, you should be at least a bit familar with
gruntime - since it wraps it very closely.

Basic types:
------------

GRuntime defines a few basic C types for numbers, strings, enums etc - see
<gobject/gtype.h> for details.

To access these types from scheme, there are GOOPS classes:

   <gchar> <guchar> <gboolean> <gint> <guint> <glong> <gulong>
   <gfloat> <gdouble> <gchararray>

Instances of this type are internally stored as GValues:

   (make <gboolean> #:value #f)
   (make <guint> #:value 85)
   (make <gfloat> #:value 3.1415)
   (make <gchararray> #:value "Hello World!")

Enums and Flags:
----------------

In C, you can create a new enum/flags type by using (see <gobject/genums.h>)
the C functions g_enum_register_static() / g_flags_register_static ().

You can also create these types in scheme:

   (define foo-vtable '#((hello "Hello World" 1) (test "Test" 2)))
   (define gtype:foo (genum-register-static "Foo" foo-vtable))
   (define <foo> (gtype->class gtype:foo))

Now you can create instances:

   (make <foo> #:value 'hello)
   (make <foo> #:value "Hello World")
   (make <foo> #:value 1)

If there's an already existing enum or flags type, you can get information
about it:

   (genum->value-table gtype:foo)

Instances of a enum / flags type can be converted to a native scheme value:

   (define foo (make <foo> #:value 'hello))
   (genum->symbol foo)
   (genum->name foo)
   (gvalue->scm foo)

Closures:
---------

Closures are useful if you want to pass a scheme function as a callback
argument to a C function - in C there's a special GClosure type for this
(see <gobject/gclosure.h>) - in libbonobo, there's a bonobo_closure_invoke()
function which makes this even easier.

Let's assume we have a C function which return an `int' and takes two
arguments - a `long' and a callback:

    int
    my_func (long long_argument, GClosure *callback)
    {
        int int_ret;

        bonobo_closure_invoke (callback, G_TYPE_INT, &int_ret,
                               G_TYPE_LONG, long_argument,
                               0);

        return int_ret;
     }

Implementing the callback function in scheme is really easy:

   (define callback (lambda (x)
     (let ((retval (* x 8)))
       (display (list "HELLO" x "RETURNS" retval)) (newline)
       retval)))

   (define closure (make <gclosure> #:return-type gtype:gint
                         #:param-types (list gtype:gulong)
                         #:func callback))

Now you can invoke the callback either from scheme or from C:
(in the example above, `closure' is the same what a C function
with a `GClosure *' return value would return to scheme):

   (gclosure-invoke closure 256) ; returns 248

But you can also pass the closure as callback to the C function:

   (my-func 256 closure) ; returns 248

GType:
------

One of the ideas behind guile-gobject was to provide complete acess to
GNOME 2's type system - and to make it possible to use all GType's which
are known in C in scheme without any additional C glue code.

That is - if you use GNOME 2's type system to define a new GType "Foo" in
C, you can automatically use this new type in scheme - and all you need to
know is either its name or its GType.

In GNOME 2, new types are normally declared in the header file like this:

   #define FOO_TYPE_HELLO (foo_hello_get_type ())
   GType foo_hello_get_type (void) G_GNUC_CONST;

To wrap this new type in scheme, you can either provide C glue code for
the foo_hello_get_type () function and do it like this:

   (define foo-type:hello (foo-hello-get-type))
   (define <foo-hello> (gtype->class foo-type:hello))

Or you can just access the new type by its name - in C, the type is created
by calling:

   GType g_type_register_static (GType               parent_type,
                                 const gchar        *type_name,
                                 const GTypeInfo    *info,
                                 GTypeFlags          flags);

The `type_name' uniquely identifies the type in the gruntime system - so
you can get the type by its name:

   (define foo-type:hello (gtype-from-name "FooHello"))
   (define <foo-hello> (gtype->class foo-type:hello))

To create the GOOPS class for the new type, you should always use
`gtype->class' - it takes care of creating the type's parent classes
if necessary, putting the class into the correct metaclass, creating
all the class slots etc.

GObject:
--------

A <gobject> (`GObject *' in C) is the basic object type in the gruntime
library - guile-gobject allows you both to access existing GObject types
and to create new GObject types in scheme.

To access an already existing GObject type, all you need is it's GType:

   (define gtktype:label (gtype-from-name "GtkLabel"))
   (define <gtk-label> (gtype->class gtktype:label))

Now you can create instances:

   (make <gtk-label> #:label "Hello World!")

You can also emit signals on the new GObject instance:

   (gobject-signal-emit test 'roswell)

Or connect signal handlers:

   (gobject-signal-connect test 'roswell
     (lambda (object) (display (list "ROSWELL" object)) (newline)))

Get properties:

   (gobject-get-property test 'foo)

Set properties:

   (gobject-set-property test 'foo 345)
   
Creating new GObject types:
---------------------------

To create a new GObject type:

   (define gtype:test (gobject-type-register-static gtype:gobject "Test"))
   (define <test> (gtype->class gtype:test))

This creates an entirely new GObject type - deriving from the parent
type gtype:gobject - but you can also derive any other GObject type
including types which you created in scheme:

   (define gtype:super (gobject-type-register-static gtype:test "Super"))
   (define <super> (gtype->class gtype:super))

Now you can add new signals to your new type:

   ;; Define a new signal "roswell".
   (gobject-class-define-signal <test> 'roswell gtype:void)

   ;; This is the default handler for this signal.
   (define-method (test:roswell (object <test>))
     (display (list "ROSWELL" object)) (newline))

Or add a new property to it:

   (define param (make <gparam-long> #:name 'test))
   (gobject-class-install-property <test> param)

(gnome corba):
==============

CORBA support for guile :-)

This module depends heavily on ORBit2 - the ORB which we use in GNOME 2 -
it doesn't work with any other ORB, but read the bootstraping section
below.

Bootstrapping your IDL:
-----------------------

ORBit2 is a CORBA 2.3 compliant ORB - and, of course, it's interoperable
with other ORBs.

However, it's not very useful to have an opaque <CORBA:Object> in scheme
if you don't know anything about this object. It's also not very useful if
you can't write your own CORBA servants in scheme.

Basically, there are two ways to solve this problem: parsing the IDL at
runtime or doing it the GNOME 2 way - loading a shared library which
contains all the necessary information from the IDL.

Some of the core GNOME 2 libraries like Bonobo provide so called "imodule"
libraries which are installed in `$(libdir)/<modulename>_imodule.la'.

To create such a library for your own IDL, you need to run ORBit2's
IDL compiler, orbit-idl, with the --imodule argument - see the demos/corba/
directory in this distribution for an example.

NOTE: The rest of this section refers to the sample IDL which can be found
      in  demos/corba/Foo.idl in this distribution.

Once you installed this library, you can read it in scheme by calling

   (corba-primitive-open-module "Foo")

This creates all the GOOPS classes and methods you need.

If there's a a CORBA interface Foo::Hello, the bootstrap process will create
a GOOPS class <Foo:Hello> which serves as stub class and another GOOPS class
<POA:Foo:Hello> which serves as skeleton class.

All stub classes are derived from <CORBA:Object> and their CORBA class
hierarchy is preserved in scheme.

All skeleton classes are derived from <PortableServer-ServantBase> and
their CORBA class hierarchy is preserved as well.

Calling CORBA Methods:
---------------------

This section refers to the sample IDL in demos/corba/Foo.idl in this
distribution.

To call a CORBA method, all you need to do is to invoke the corresponding
method in the stub class - let's assume `hello' is an instance of the
<Foo:Hello> class:

   (Foo:Hello:doHello hello) ; calls the CORBA method `Foo::Hello::doHello'
                               on the CORBA Object `hello'.

So to call CORBA methods, you don't even need to know that it's CORBA :-)

Oh, btw. the cool thing about CORBA and scheme is that you don't need to
worry about these annoying CORBA exceptions - since you automatically get
a scheme exception `corba-system-exception' / `corba-user-exception'.

Implementing CORBA servants:
----------------------------

Well, I guess the interesting part is to implement CORBA servants in
scheme - so let's assume you want to write a servant for the `Foo::Hello'
interface.

The first thing you need to do is to derive its POA class (as a special
"feature" you can also use the POA class directly to specify the default
behavior for all servants of this interface, see below):

   (define-class <hello> (<POA:Foo:Hello>))

Then, you define methods 

   (define-method (Foo:Hello:doHello (hello <hello>))
     (display (list "Hello World!" hello)) (newline))

If you call `(next-method)', the POA classes method will be run - and
the default is to throw a CORBA::NO_IMPLEMENT system exception.

However, you can override this:

   (define-method (Foo:Bar:Baz:haveFun (object <POA:Foo:Bar:Baz>) a b)
     (display (list "Default Foo:Bar:Baz:haveFun handler!" a b))
     (newline))

If you created all the methods, you can create servants and call
`corba-servant->reference' to get a CORBA::Object reference:

   (define servant (make <hello>))
   (define hello (corba-servant->reference servant))

Now you have a CORBA Object `hello' (for guile, this is an instance of
the GOOPS class <Foo:Hello>) and you can invoke methods on it:

   (Foo:Hello:doHello hello)

Even if this looks like there's just a scheme method being called - this
is a "real" CORBA call - for scheme `hello' is a "normal" CORBA Object.

NOTE: Any CORBA Objects which you create in guile are "owned" by guile's
      garbage collector - so make sure to CORBA_Object_duplicate() in a
      C function before you store it somewhere !

Implementing CORBA servants - multiple inheritance:
---------------------------------------------------

Like in C, you can also create servants for CORBA interfaces which are
derived from other interfaces:

   (define-class <maximum> (<hello> <POA:Foo:MaximumHello>))
   (define-method (Foo:Hello:doHello (hello <maximum>))
     (display (list "Hello Maximum World!" hello))
     (newline)
     (next-method))

   (define maximum-servant (make <maximum>))
   (define maximum (corba-servant->reference maximum-servant))

This creates a new servant for the CORBA interface Foo::MaximumHello which
is derived from Foo::Hello and Foo::Bar::Baz - this inheritance is reflected
in scheme.

   ;; Calls method `Foo:Hello:doHello' in class <maximum> and then
   ;; in <hello> because of the (next-method).
   (Foo:Hello:doHello maximum)

   ;; Calls method `Foo:Bar:Baz:haveFun' in class <POA:Foo:Bar:Baz> -
   ;; the default handler.
   (Foo:Bar:Baz:haveFun maximum 1 2)

Since we're using real CORBA calls, all of this also works for calls which
are coming "from the outside" - ie. from C or a remote process.

Implementing CORBA servants - an important limitation:
------------------------------------------------------

CORBA servants can be implemented either in C or in scheme - but you cannot
mix them - to make it clear, an example:

In the example above, you learned how to create a CORBA servant for the
Foo::MaximumHello CORBA interface in scheme.

Now let's assume you already have an implementation for the Foo::Hello
interface in C.

Even if Foo::MaximumHello is derived from Foo::Hello - you cannot use the
Foo::Hello C implementation in scheme.

This limitation may sound obvious, but it's not so obvious at all if you're
a bit familiar with CORBA. In C, you would normally expect to have a `vepv'
and a `epv' vector in a CORBA servant - and to be able to poke around in the
vepv to override methods.

As an ORBit2 specific implementation details, servants which you create from
scheme don't have a `vepv' at all and the `epv' is not what you'd expect -
the `epv' entries are scheme vectors and not pointers to C functions.

[Implementation details:

 This works because ORBit2 has a feature to call one single function to marshal
 any calls on a CORBA object - the `impl_finder_func' and `relay_call' fields
 in the class info (see ORBit2 source code for details).

 This feature was explicitly added to ORBit2 to make it easy to use from
 scripting languages - it'd be difficult to provide "normal" `epv' entries
 for the scheme methods (you'd have to generate a C function at runtime which
 also contains a data pointer somewhere - before we added this feature to
 ORBit2 I was using weird assembler tricks here and it was also using much
 more memory - about 40 more bytes / CORBA method).
]

CORBA structs / sequences:
--------------------------

There's support to access CORBA structs / sequences from scheme including
a special record type for structs - see the (gnome corba) module for
details.

Summary, Remaining Issues, ....:
================================

To summarize, this code is in a state where the "base engine" is almost
completed - that is, the code which deals with the classes types etc.

What needs to be done is to write the glue code to actually call all the
C functions:

* we probably need to write a module (gnome glib) which knows how to deal
  with glib data types (GList etc.) and which contains the glue for all
  the glib functions.

* (gnome gtk) is completely unfinished / not existant at all.

  However, we already have all the functionality and we also have all the
  types (just call `gtype-from-name' for all of GTK+'s types) - so it should
  be possible to write a g-wrap file for all the GTK+ functions.

  As a first step, we should try to reuse guile-gtk's .defs files.

* (gnome corba types) needs some love.

* (gnome corba primitives) - need to add support for the remaining types
  to the marshallers.

* (gnome corba) needs some love - someone needs to write good scheme
  bindings for structs and sequences.

* Bug files

* More ....

Last words:
===========

If was really fun to hack on this stuff and on GNOME 2 in general and I'd
really like to continue hacking on it.

However, it's already past midnight and I need to draw a clear line here.
I guess it also doesn't really matter since in the Free Software word, a
module is never really finished and there's always someone else who
continues hacking on it.

So, normally I'd terminate a release notes file with something like
"comments, ..., are welcome" - but unfortunately the situation is a bit
different.

Since this announcement will also go to guile-devel, a little explanation:

Today, I mean yesterday - it's already past midnight - was my last day of
GNOME 2 hacking for the next four months. Hacking on GNOME was so much fun
for me - but now I need to concentrate on my real life and learn for the
university.

Most likely, I won't be completely away during this time, but I won't have
any time to hack, all I can do is to read my mail or to spent a few hours
in IRC from time to time.

However, after hacking so much on GNOME 2, I need to become familiar with
being a student again, with spending most of the day with learning, getting
up early and going to the library and all this stuff.

On Thursday evening, I'll shut my machine down for at least 2-3 weeks or
even longer - depending on how long it takes until I have enough distance
to GNOME so that I can safely read my mail or come to IRC.

I also set myself a hard limit of Friday morning 04:12 am local time -
that's when the first train leaves the station - and I'll sit in this train
to spend the weekend in Berlin. It's obvious that I can't boot my machine
anymore when I'm back in Trier - since this weekend is the "barrier" between
being a GNOME hacker and being a student ....

So, any mails which arrive me until Thursday afternoon European time will
get answered - anything which arrives later ....

Hmm, I guess that's the longest release notes file I ever wrote :-)

-------------
September 10th, 2001
Martin Baulig <martin gnome org>


NOTE: Please don't use baulig suse de anymore, it'll become disfunctional
      in two weeks since I'm no longer employed at SuSE !

-- 
Martin Baulig
martin gnome org (private)
baulig suse de (work)