guile-gtk-0.9 released



I'm happy to announce version 0.9 of guile-gtk.  It is available from

    http://www.ping.de/sites/zagadka/guile-gtk/

It is a package to make Gtk, the GIMP toolkit, accessible from Guile,
GNU's Scheme implementation.  A patch for integrating it into
gnome-guile will follow shortly.

A lot of things have changed since the last release, so here is the
NEWS entry:

* Gtk+ version 1.0 is required from now on.

* The module has been renamed to (gtk gtk) from (toolkits gtk).  The
  Gdk bindings have been moved to the module (gtk gdk).

* Dynamic linking of compiled code modules works differently from the
  usual setup of Guile.  No strange symlinks are made.  See the README
  and "gtk/dynlink.scm" for details.

* The Scheme code (gtk/*.scm and event-repl.scm) is now
  installed under the prefix of this package, not under the prefix
  where Guile was installed.  This will not make a difference most of
  the time, because Guile and this package will likely be installed
  with the same prefix, but being too clever about where to install
  Scheme files caused trouble for package maintainers, I think.

  When Guile is part of a distribution and sits in /usr, I expect it
  to still have /usr/local/share/guile in its path.  Is this
  unreasonable?

* New Scheme function `gtk-standalone?' that returns whether the Guile
  interpreter is running a Gtk-aware read-eval-print-loop or not.
  When `(gtk-standalone?)' returns true you should call `gtk-main' or
  `gtk-exit' from your script at the approriate times.  When it
  returns false, you can assume that someone else will take care of
  running the Gtk event loops and quitting the interpreter.

  It would be ideal to arrange thins so that gtk-standalone could
  always return false.  I'm not at all sure how to achieve this, tho.

* New Scheme function `gtk-standalone-main' that can be used to
  conditionally run a stand-alone session.

	gtk-standalone-main TOPLEVEL

  When (gtk-standlone?) is true: connect gtk-exit to the "destroy"
  signal of TOPLEVEL and call gtk-main.

* New Scheme function `gtk-callback-trampoline' that can be used to
  intercept every callback to a Scheme procedure.  Currently, this is
  used to catch errors that occur in callbacks and popup a window
  showing the error message with a backtrace.

* The types `int' and `uint' from the definition files are now handled
  properly.

* The type `string' can now be used as the return type of functions.
  Be sure to understand that the string is assumed to to be in
  malloced memory and that the caller takes ownership of that memory.
  Whe you want to return a string whose memory should not be freed by
  the caller, use the type `static_string' instead.

* New Scheme functions `gtk-object-new' and `gtk-object-set'.
  They can be used to create new GtkObjects or set their attributes
  with a variable-length keyword/value list.  `gtk-widget-new' and
  `gtk-widget-set' are provided, too.

    (gtk-object-new TYPE KEY VAL KEY VAL ...)
    (gtk-widget-new TYPE KEY VAL KEY VAL ...)
    (gtk-object-set OBJ KEY VAL KEY VAL ...)
    (gtk-widget-set OBJ KEY VAL KEY VAL ...)
	
  TYPE is the name of the type that the new object should be created
  of.  It should be a symbol.  KEY is either a symbole or a keyword.
  It can be a fully qualified attribute name like "GtkWidget::parent",
  or it can be a shortened version like "parent".  See
  "examples/simple.scm" for an example.  Here is a very short one:

    (gtk-object-new 'GtkButton 'label "Hi")

  The TYPE names must already be known to either Guile-Gtk or Gtk+.
  This means that they must either be listed in some *.defs file that
  has been translated into glue code and the initialization function
  of that code has been called; or that the type has already been used
  somewhere else (and can thus be found with `gtk_type_from_name').

  The "GtkObject::signal::<name>" form is not supported.

* New Scheme function `gtk-object-get' and `gtk-widget-get'.  Use them
  like

     (gtk-widget-get label 'label)

* Guile-gtk now uses the improved reference counting of Gtk+.  This
  should not be too visible from the outside.

  The difference is that no GtkObject will silently disappear when
  Scheme still has a reference to it.  Destruction of a GtkObject is
  now decoupled from its `finalization'.  Calling `gtk-widget-destroy'
  continues to remove a widget from the display and clean it up in
  other ways, but the actual GtkWidget structure will only be freed
  when the last reference to it has been revoked.  Thus, reference
  counting is now only a resource management technique (as it should
  be) and because Scheme has automatic garbage collection you do not
  need to worry about it.  Consequently, the `gtk_object_ref' and
  `gtk_object_unref' calls are not exported to Scheme.

  Another consequence is that the Scheme value that stands in for a
  particular GtkObject will not vanish as long as the GtkObject is
  there.  This means that you can have weak references to them that
  will only then be invaliddated when truely no-one cares about the
  GtkObject any longer (both from within Scheme and from the outside).

  Unfortunately, it is possible to form cycles of GtkObjects and other
  Scheme values that are not detected by the garbage collector.  These
  cycles will not be collected altho they are garbage.  The explicit
  destruction of a GtkObject that is part of this cycle will very
  likely break it and make all members collectable; so as long as you
  destroy all your object, everything should be fine.

  For a cycle to go undetected, it must involve one or more Scheme
  value (because GtkObject do not form cycles themselves) and at least
  one `untraced' pointer from one GtkObject to another one.  Currently
  only the parent->child link is traced, but the plan is to extend
  this tracing to all inter-object pointers and thus to eliminate the
  potential for cycles completely.

  Another drawback of untraced references is that they cause certain
  GtkObjects to be collected later than necessary.  For example, a
  GtkAdjustment that is in use by a GtkRange *and* known to Scheme has
  a reference count of 2, one from the GtkRange, the other from the
  Scheme proxy object.  The GC might discover that the GtkRange is
  garbage and collect it, thus releasing one referece to the
  GtkAdjustment.  The GtkAdjustment, however, has not been detected as
  garbage, because there was an unidentified (untraced) reference to
  it.  Only the next GC will collect the GtkAdjustment.

* The function `gtk-destroyed?' has been removed, as a consequence of
  the ref_counting changes.  In some cases, it might be possible to
  use `gtk-object-destroyed' instead.  This function determines
  whether `gtk-object-destroy' has been called on a GtkObject.

* Access to information contained in GdkEvents is almost there.  There
  is a fair number of accessor functions defined in gdk.defs.guile,
  but I'm not too happy about this approach.  GdkEvent is a quite
  complicated structure, we should not export this complexity to
  Scheme.

* Boxed types are implemented differently and can now be reliably compared
  with `eq?`.

* The syntax of `define-boxed' has been changed to allow for the
  definition of options (like `fields', see below).  It is now just

    (define-boxed NAME OPTION...)

  The copy, free and size parameters are now specified with the
  options `copy', `free', and `size'.  The `size' option is, erm,
  optional, but `copy' and `free' are mandatory.  For example,

    (define-boxed GdkEvent
      gdk_event_copy
      gdk_event_free
      "sizeof(GdkEvent)")

  becomes

    (define-boxed GdkEvent
      (copy gdk_event_copy)
      (free gdk_event_free)
      (size "sizeof(GdkEvent)"))

* Boxed types can now have `fields', like object types.

* The `field' option for object and boxed types can now also specify
  settable fields by including the `setter' option for an individual
  field.  For example, within the defintion for GtkAdjustement,

    (fields
      (float value (setter #t)))

  will generate a function

    gtk-adjustement-set-value! ADJ VALUE

* The `gen-typeinfo' script has been renamed to `build-guile-gtk' and
  is installed.  It should be ready to be used for generating glue
  code for external *.defs files.

  The usage of build-guile-gtk is

     guile-guile-gtk [GLOBAL-OPTIONS] CMD [CMD-OPTIONS] DEFS...

  Supported global options are

     -I DIR          Add DIR to the search path for *.defs files.

  Supported CMDs are

  * glue

  The command "glue" will write glue code for the definitions from the
  file DEFS to stdout.  DEFS is the name of a file with the familiar
  feature definitions.  When DEFS does not exist, it is searched for
  in the `import path' which initially consist of the directory
  "$(prefix)/share/gtk".  You can add directories to the import path
  with the -I global option.

  From with a DEFS file you can import other files with the statement
  
     (import DEFS)

  This will read DEFS and make all type definitions in it available.
  No glue code will be generated for the definitions in DEFS.  DEFS is
  searched in the same way as the file given on the command line.

  After reading the foo.defs file (either from the command line or due
  to a import statement) build-guile-gtk tries to read a file named
  "foo.defs.guile".

  The statement

     (add-options SYM OPTS...)

  can be used to add extra `options' to arbitrary features from a
  *.defs file.  SYM identifies the feature, it is either a type or
  function name.  OPTS is a list of options.  An option is itself a
  list of the form

     (NAME VALS...)

  where NAME is symbol that identifies the option.  The statement
  `add-options' is Guile-gtk specific and should only be used in
  *.defs.guile files.  In addition, there is also a statement

     (options OPTS)

  that can be used to set global options.  The exact set of options is
  rather ad-hoc currently.

  * link

  Generate, compile and link a program that is just like guile-gtk but
  with all bindings from the DEFS files preloaded.  This is useful for
  systems that can't do dynamic linking of the bindings.  All
  arguments after "link" that are not *.defs files are passed to the
  compiler unchanged.

  When you want to have more control, you might find these commands
  useful:

  * main

  The "main" command will output, to stdout, C code for a main
  function that initializes all definitions from the DEFS files given
  as arguments.

  * cflags

  Output necessary options to compile the generated glue code or the
  output of the "main" command.

  * libs

  Output necessary linker options to link the main function.
  
* The `guile.details' file is gone.  Its function is now performed by
  the options mentioned in the last paragraph.

* The file `gtk.defs' has been split into `gdk.defs' and `gtk.defs',
  mostly to test out the new features of build-guile-gtk.

* There is a new example program `test-gdk.scm' that shows how one
  might go about drawing from Scheme.



[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]