Expand --no-touch-headers to include the private header.

[gob-dx.git] / doc / gob2.1.in

.\"
.\" gob manual page
.\" (C) 1999-2011 Jiri (George) Lebl <jirka@5z.com>
.\" 
.\" This manual page is covered by the terms of the GNU General
.\" Public License.  
.\"
.TH GOB2 1 "GOB2 @VERSION@" 
.SH NAME
GOB2 \- The GObject Builder
.SH SYNOPSIS
.PP
.B gob2
[ option ] ...
file

.SH DESCRIPTION
.PP
GObject Builder is a simple preprocessor for easily creating
GObject objects.  It does not parse any C code and ignores any C errors.  It
is in spirit similar to things like lex or yacc.  In some ways it
also resembles java.  But it is really just a simple preprocessor
for creating GObjects for use in C or C++ and it is not a programming
language.

.SH OPTIONS
.PP
.TP
.B \-? \-h \-\-help
Display a simple help screen.
.TP
.B \-\-version
Display version information
.TP
.B \-w \-\-exit\-on\-warn
Exit with an error code even when you encounter a warning.
.TP
.B \-\-no\-exit\-on\-warn
Exit with an error only on errors, not on warnings, this is the default.
.TP
.B \-\-for\-cpp
Generate C++ code.
.TP
.B \-\-no\-extern\-c
Never add the extern "C" to the header.
.TP
.B \-\-no\-gnu
Never generate any code with GNU C extensions.  However all the GNU C
extensions are always wrapped in #ifdef __GNUC__, so code using them compiles
correctly even on non-GNU compilers.  This option is for purists only.
(using GNU extensions some warnings are eliminated, some ugly hacks and there
is better argument type safety, so it\'s good to use them)
.TP
.B \-\-no\-touch
Don\'t touch output files unless they really
changed (implies \-\-no\-touch\-headers).  Be careful with automake, see section
PREVENTING SPURIOUS BUILDS.
.TP
.B \-\-no\-touch\-headers
Don\'t touch any generated header file unless that file really changed.
This avoids spurious rebuilds, but can confuse some make systems so it is not
enabled by default.
.TP
.B \-\-always\-private\-header
Always create a \fB<basename>-private.h\fR file, even if it would be empty.
.TP
.B \-\-ondemand\-private\-header
Create the private header only if it would have something in it, that is,
if there are some private data members or protected methods.
This is the default.
.TP
.B \-\-no\-private\-header
Never create a private header file.  If we use any private data members,
define the private data structure at the point in the .c source where
the class definition begins.
.TP
.B \-\-m4
Preprocess source with m4. Following args will be passed to m4.
.TP
.B \-\-m4\-dir
Print directory that will be searched for m4 files.
.TP
.B \-n \-\-no\-write
Do not write any output files, just check syntax of the input file.
.TP
.B \-\-no\-lines
Do not print out the \'#line\' statements into the output.  Useful for debugging
the auto-generated generated code.
.TP
.B \-\-no\-self\-alias
Do not create the Self and SelfClass type aliases and the SELF, IS_SELF
and SELF_CLASS macros.
.TP
.B \-\-no\-kill\-underscores
Do not remove the initial underscore from method names.
.TP
.B \-\-always\-private\-struct
Always include the private pointer in the public header file.  This is useful
for files which are part of a library and you want to reserve the right to add
some private data members without breaking binary compatibility.
.TP
.B \-o \-\-output\-dir
The directory into which output should be placed.
.TP
.B \-\-file\-sep[=c]
Replace default \'\-\' file name separator.  If no separator character
is given then none is used.  Only one character can be used.
.TP
.B \-\-gtk3
Use gtk3.

.SH TYPENAMES
.PP
Because we need to parse out different parts of the typename, sometimes you
need to specify the typename with some special syntax.  Types are specified in
capitalized form and words are separated by \':\'.  The first word of the type
(which can be empty) is the "namespace".  This fact is for example used for the
type checking macro and the type macro.  For "Gtk:New:Button", the macros will
be GTK_IS_NEW_BUTTON and GTK_TYPE_NEW_BUTTON.  This colon separated format of
typenames is used in the class declaration header and for method argument
types.

.SH OUTPUT FILES
.PP
The filenames are created from the typename.  The words are
separated by \'\-\' (this can be changed with
\fB\-\-file\-sep\fR option) and all in lower case.  For example for an object named
"Gtk:New:Button", the files are \fBgtk\-new\-button.c\fR and
\fBgtk\-new\-button.h\fR.
If you are using C++ mode, the output .c file will in fact be a .cc file.
If you have any private data members, a private header file will also
be created, called \fB<basename>\-private.h\fR (for the example above it
would be gtk\-new\-button\-private.h).
The public header file is created to be human readable and to be used as a
reference to the object.  The .c source file is not created as a human
readable source and is littered with #line statements, which make the
compiler attempt to point you to the right line in your .gob file in
case of parsing errors.  The output should not be edited by hand, and
you should only edit the .gob file.

.SH INCLUDING NORMAL C CODE IN THE OUTPUT FILES
.PP
To include some code directly in the output C file begin with \'%{\'
on an empty line and end the code with a \'%}\' on an empty line.  These
sections will appear in the output files in the order they are given.
There are several other \fIsections\fR to which you can put code.  You can
put it in the \'header\' section (which can be abbreviated \'h\') and it will
go into the public header file.  You can also put it in the \'privateheader\'
section (abbreviated \'ph\') which will make the code go into the private
header file.  Sometimes you want some code (other includes) to appear before
the extern "C" and the protecting define.  To do this you can put them
into the \'headertop\' (or \'ht\') section.  You may wish to include code or
comments in all the files, which you can do by putting them into the \'all\'
(or \'a\') section.  Similarly, code you wish to appear at the top of all
files go in the \'alltop\' (or \'at\') section.  When you want code
to appear as in alltop but only in the cfile you use the \'ctop\' (or \'ct\')
section.  Note that ctop requires 2.0.18.   Finally,
\'afterdecls\' includes code between the declarations and the method
implementations, but note that \'afterdecls\' requires version 2.0.16.
For example:
.nf

  %alltop{
        /* this will be at the very top of all output files */
  %}

  %ctop{
        /* this will be at the very top of the C file */
        /* Requires 2.0.18 */
  %}

  %headertop{
        /* this will be on top of the public header */
  %}

  %privateheader{
        /* this will go into the private header file */
  %}

  %h{
        /* will be included in the header */
        void somefunc(int i);
  %}

  %a{
        /* will be included in all files */
  %}

  %afterdecls{
        /* between the declarations and the method implementations */
        /* Requires gob version 2.0.16 */
  %}

  %{
        /* will be included in the C file */
        void somefunc(int i)
        {
              /* some code */
        }
  %}

.fi

.SH INCLUDE FILES
.PP
Gob will automatically include the class header file at the top of the .c 
source file.  If you wish to include it somewhere else, put the include
into some %{ %} section above the class definition, and gob will not include
it automatically.  This way you can avoid circular includes and control
where in the file do you want to include the header.
.PP
If you made any data members private, gob will also create a source file
that will be called \fB<basename>\-private.h\fR.  Same rule as above applies
for this just as it does for the regular header file.  If you do explicitly
include the regular header file, you should always include this private
header file below it.  That is, if you use any private data members.  If you
don\'t, the private header file automatically includes the public header file,
and thus the public header file will be indirectly included at the very top
of the file.

.SH THE CLASS HEADER
.PP
There can be only one class per input file.  Defining a class
is sort of like in Java, you define the class and write inline code
directly into the class definition.  To define a class you need to specify
the new object name and the name of the object from which it is derived
from, such as this "class <new type> from <parent type> { <class code> }".
For example:
.nf

  class Gtk:New:Button from Gtk:Button {
          <class code>
  }

.fi
.PP
To make an abstract class (to pass G_TYPE_FLAG_ABSTRACT) add \'(abstract)\'
before the curly braces above.  This works since version 2.0.13.

.PP
To make a simple dynamic class which can be registered with a GTypeModule,
add \`(dynamic)\' to the class header.
This will cause a type registration method to be defined, which you would
normally call from the load method of a GTypeModule.
In the above example, the registration function will look like
.nf

  void gtk_new_button_register_type(GTypeModule *type_module)

.fi
.SH DATA MEMBERS
.PP
There are five types of data members.  Three of them are normal data members,
one is class wide (global) in scope and one is a virtual one, usually linked to
a normal data member or a class wide data member.  The three normal data
members are public, protected and private.  Public and protected are basically
just entries in the object structure, while private has it\'s own dynamically
allocated private structure.  Protected members are always put after the public
one in the structure and are marked protected in the header file.  There is
only one identifier allowed per typename unlike in normal C.  Example:
.nf

  public int i;
  private GtkWidget *h;
  protected long k;

.fi
.PP
Public and protected data members are accessed normally as members of
the object struct.  Example where \'i\' is as above a public data member:
.nf

  object\->i = 1;

.fi
.PP
The private data members are defined in a structure which is only available
inside the .c file, or by including a private header file.  You must access
them using the structure _priv.  Example
where \'h\' is the private data member (as in the above example):
.nf

  object\->_priv\->h = NULL;

.fi
The _priv structure is defined in the \fB<basename>\-private.h\fR.
This file is automatically included if you don\'t include it yourself.  You
should always explicitly include it in your .gob file if you explicitly also
include the main header file.  The reason it is a separate header file is
that you can also include it in other places that need to access this objects
private data, such as if you have the majority of functionality of an object
in a separate .c file.  Or if a derived object needs to access the protected
methods.
.PP
In case you use the \fB\-\-no\-private\-header\fR option, no
private header file is created and you can only access the _priv pointer
below the class definition in the .gob file.
.PP
Also note that this structure is dynamically allocated, and is freed in the
finalize handler.  If you override the finalized handler, your code will be
run first and only then will the _priv structure be freed.
.PP
.B "Classwide data members:"
.PP
Sometimes you want a datamember to be shared by all objects.  You then need
the "classwide" scope keyword.  So for example the following adds a global
member foo:
.nf

  classwide int foo;

.fi
To access the member you can use the SELF_GET_CLASS macro (or
YOUR_OBJECT_NAME_GET_CLASS) to get at the class.
Thus the following would work:
.nf

  SELF_GET_CLASS(object)\->foo = 20;

.fi
.PP
.B "Automatic Initialization:"
.PP
You can automatically initialize the public private and protected data members
without having to add an init method.  The advantage here is that
initialization is kept close to the definition of the data member and thus
it\'s easier to check.  To do this, just add a \'=\' followed by a number or
a token.  It is also possible to include arbitrary C code for more elaborate
initializations by putting it all in curly braces.  Note that the curly braces
will not be printed into the output, but since gob does not C parsing it needs
them to figure out where the C code ends.  The code will be inserted into the
init method, above the user defined body.  So for example the following
will initialize an integer to \-1 and a string with a newly allocated string
of "hello".
.nf

  public int foo = \-1;
  private char *bar = {g_strdup("hello")};

.fi
.PP
.B "Automatic Destruction:"
.PP
Most data stored as pointers needs to have a function called when the object
is finalized to either free the data.  Gob will let you
define a function to be called on the data the object is finalized.  This is
achieved by putting \'destroywith\' followed by a function name after the
variable definition.  It is only called if the data you defined this on
is not NULL, so you cans specify functions which do not handle NULL.  It
is very much like the GDestroyNotify function used in GTK+ and glib in many
places.  Unlike many other places, gob will not enforce any kind of type
safety here so be a little bit more careful.  Any function you give it will
be called as a "void function(void *)".  It will in fact be cast into such
a form before called.  This is to avoid spurious warnings for gtk calls to
subclass methods.  The function needs not be of that form exactly, it just has
to take one argument which is the pointer to the data.  You should also not
define this on any non-pointer data as the results may be undefined.
Example:
.nf

  public char *foo = {g_strdup("bar")}
          destroywith g_free;

.fi
Note that the function name you give must be a real function and not macro.
Also note that this is always called in the "finalize" method of GObject.
It is always called after any user defined body of the finalize handler.
.PP
Sometimes you may want to run arbitrary code on destruction.  While this can
be perfectly well done in the finalize handler.  Depending on the style you
may want to include all destruction/initialization code together with the
definition of the data member.  Thus you may want to put arbitrary code which
will then be inserted into the "finalize" method of GObject.  This can be
done with the "destroy" keyword followed by arbitrary code in curly braces.  
Inside this code a macro called VAR will be define which refers to your
variable.  So for example destroying a GString can be either done with
a helper routine or the following code:
.nf

  public GString *string = {g_string_new(NULL)}
          destroy {
                  if(VAR) g_string_free(VAR, TRUE);
          };

.fi
The thing to remember with these is that there are many ways to do this
and you\'d better be consistent in your code in how you use the above things.
Also defining a helper routine that will do the destruction will be a nicer
thing to do if that\'s a possibility.  The "destroy" keyword with code does
take up more space in the file and it may become more cluttered.
.PP
The data is zeroed out after being destroyed.  This is to make debugging easier
in case your code might try to access an already finalized object.  In case
you have overridden the finalize method, your code will be run first and
only then will the destructors be called.  You should not however make any
assumptions about the order at which the destructors are called.  If you have
interdependencies between destructors for different data members, you will
have to do this in your own finalize override function.
.PP
.B "Automatic Unreffing:"
.PP
This is very much like the automatic destruction, but is instead run in the
dispose method (it is among other places called from the "destroy" method of
GtkObject).  All data and other objects that you need to unref should be
done here, and not at finalize time.  The semantics are otherwise the same
as for the "destroywith" and "destroy" keywords, except that you use
"unrefwith" and "unref".
.nf

  public G:Object *foo = NULL
          unrefwith g_object_unref;
  public G:Object *bar = NULL
          unref {
                g_object_unref (VAR);
          };

.fi

.SH GOBJECT PROPERTIES
.PP
The fourth type of a data member a property type.  It is a named data member
which is one of the features of the GObject system.  It just defines a way to
get and set some data, but you have to take care of storing that data
somewhere.  So it is normal to also have a normal private (or public)
data member where you store the real data.
You normally need to define a
get and a set handler.  They are fragments of C code that will be used to get
the value or set the value of the argument.  Inside them you can use the define
VAL to which you assign the data or get the data.  You should treat this VAL
as a GValue which stores the data of the correct type.
You can also use the
identifier "self" as pointer to the object instance.  The type is defined as
one of the GObject type enums, but without the G_TYPE_ prefix.  There are
also some attributes of a property which you can set.  For example the
following is a definition of an integer property \'height\' which will
be synchronized with a private integer data member also of the name \'height\'.
.nf

  private int height;
  property INT height
         (
          name = "height",
          nick = _("Short nickname"),
          blurb = _("Long description"),
          minimum = 10,
          maximum = 200,
          default_value = 100)
        set { self\->_priv\->height = g_value_get_int (VAL); }
        get { g_value_set_int (VAL, self\->_priv\->height); };

.fi
.PP
The attributes are really optional though you should at least set some
of them.
All property types have a \'nick\' and a \'blurb\' attribute and you should
set those accordingly.  This will make runtime querying the object
nicer as things such as gui editors and class browsers can be more
verbose about the class itself.
.PP
The \'name\' property is canonical name of property. It is useful when you try to
implement properties with no C names like \'vertical-scroll\'. The \'name\'
property can be omitted.
.PP
You can use the \'_("string")\' notation
instead of just "string", and that will mark the string for translation.
.PP
Almost all types also have a \'default_value\' attribute which sets the initial
value of this property (on object initialization, the set handler will be run
automatically with this value).  This value will be overridden if the user 
sets a value of this property on the call to g_object_new.
.PP
All the numeric types (including CHAR) have \'minimum\' and \'maximum\'
attributes which can restrict the range.  If you do not specify these
the range will be the full range that the data type can handle.
.PP
Types such as UNICHAR and BOOLEAN only have the \'nick\', \'blurb\' and
\'default_value\' attributes.
.PP
The ENUM type has an \'enum_type\' attribute which is the exact
type of the enum.  This is so that the property knows which exact
type you can set, rather then just knowing it is an enum.  You should
always create an enum type specific for the enum itself (see section
on the enum types)
.PP
Similarly FLAGS type has a \'flags_type\' which again you should set to
the specific type of this flags data member.
.PP
There is a STRING type which has only the extra \'default_value\' attribute.
.PP
The OBJECT type is one of the types that doesn\'t have a \'default_value\' and it
only has an \'object_type\' attribute (in addition to nick and blurb of course)
that is the exact object type that this property accepts.  The object_type
should be as a type, that is for example \'Gtk:Button\'.
.PP
There is a BOXED type which is a pointer which has a boxed type defined
(such that GObject knows how to copy and destroy this pointer).  Here
you will need to specify the \'boxed_type\' attribute with the specific
type of the boxed pointer.
.PP
There is also a POINTER type, which has only the \'nick\' and \'blurb\'
attributes.  This is for storing arbitrary pointers.  You should be
careful with this one, as GObject knows nothing about the data
stored at this pointer.  It is somewhat like a \'void *\' type.
.PP
There is also the PARAM type for storing parameters with a \'param_type\'
attribute.
.PP
You should notice that this list is pretty much like the list of g_param_spec_*
functions from gobject/gparamspecs.h, and the attributes are like the
arguments of those functions.  Note however that value array is NOT supported
yet.
.PP
You can also specify extra flags, such as CONSTRUCT or CONSTRUCT_ONLY using the
\'flags\' attribute.  You can specify multiple flags by oring them together with
\'|\'.  These flags correspond to the GParamFlags enumeration except do not
include the G_PARAM_ prefix.  So for example to define an enumeration property,
which is a CONSTRUCT_ONLY property, we could do the following:
.nf

  private SomeEnumerationType foo;
  property ENUM foo
         (nick = _("Short nickname"),
          blurb = _("Long description"),
          enum_type = Some:Enumeration:Type
          default_value = SOME_ENUMERATION_VALUE,
          flags = CONSTRUCT_ONLY,
          link);

.fi
.PP
The above example also gives an example of automatic linking to a standard data
memember.  By including the attribute \'link\' a get and set handlers will be
automatically added without having to type them by hand.  This is useful for a
vast majority data types that are just linked to some standard data member and
do not need to do anything extra on get or set.
.PP
Another extra feature of properties is the possibility of automatically
exporing methods to get and set the property.  That is without having to
use g_object_set and g_object_get.  This is achieved by adding an
\'export\' attribute to the list of property attributes.
.PP
If you do not define a set or get handler, the property will automatically
be only readable or writable as appropriate.
.PP
Gob2 also creates macros which can be used for type safe access to
properties through g_object_set and g_object_get.
The macros are called <type>_PROP_<argument name>(x) and
<type>_GET_PROP_<argument name>(x).  They define both the string and the
value part of the argument.  So for setting an argument of height, one would
use (for object type My:Object):
.nf

  g_object_set (G_OBJECT (object),
                MY_OBJECT_PROP_HEIGHT (7),
                NULL);

.fi
And for getting, you would use:
.nf

  int height;
  g_object_get (G_OBJECT (object),
                MY_OBJECT_GET_PROP_HEIGHT (&height),
                NULL);

.fi
Note however that the type safety only works completely on GNU C compilers.
The code will compile on other compilers but with minimal type safety.
For complete type safety it is useful to use the get/set methods that
are defined by using the \'export\' attribute.
.PP
To get bettery type safety on some of the property types, you can specify
the \'type\' attribute which will add casts where appropriate in code dealing
with this property.  This is especially useful for POINTER and OBJECT types.
But even for others.
.PP
You can also override properties from parent objects (that is override their
implementation, not their attributes).  Do this by adding the
special \'override\' attribute.  For example if the parent object
had a \'height\' property then you could override it by
.nf

  private int height;
  property INT height
         (override)
        set { self\->_priv\->height = g_value_get_int (VAL); }
        get { g_value_set_int (VAL, self\->_priv\->height); };

.fi
Overriding is supported since gob 2.0.10.

.SH METHODS
.PP
There is a whole array of possible methods.  The three normal,
"familiar" method types are private, protected and public.  Public are
defined as normal functions with a prototype in the header file.  
Protected methods are defined as normal methods (which you can call from other
files), but their prototype is placed in the private header file.  Private
methods
are defined as static functions with prototypes at the top of the .c
file.  Then there are signal, virtual and override methods.  More on those
later.  You can also
define init and class_init methods with a special definition if you want
to add code to the constructors or you can just leave them out.
You can also not define a body for a method, by just using \';\' instead of a
body.  This will define an empty function.  You can\'t do this for non-void
regular public, private or protected methods, however it is acceptable for
non-void virtual, signal and override methods.
.PP
.B "Function argument lists:"
.PP
For all but the init and class_init methods, you use the
following syntax for arguments.  The first argument can be just "self",
which gob will translate into a pointer to the object instance.  The rest
of the arguments are very similar to normal C arguments.  If the
typename is an object pointer you should use the syntax defined above
with the words separated by \':\'
.nf
<type> <argument id>
or
<type> <argument id> (check <list of checks>)
.fi
.PP
The checks are glib type preconditions, and can be the following:
"null", which tests pointers for being NULL, "type" which checks GTK+
object pointers for being the right type, "<test> <number>" which tests
numeric arguments for being a certain value.  The test can be a <,>,<=,>=
!= or ==.  Example:
.nf
  
  public int
  foo (self,
       int h (check > 0 < 11),
       Gtk:Widget *w (check null type))

.fi
.PP
This will be the prototype of a function which has a self pointer
as the first argument, an integer argument which will be checked and has
to be more then 0 and less then 11, and a pointer to a GtkWidget object
instance and it is checked for being null and the type will also be
checked.
.PP
.B "Function attributes:"
.PP
For method that aren't virtual, signal or override methods, and aren't
init or class_init, GLib function attribute macros G_GNUC_PRINTF,
G_GNUC_SCANF, and G_GNUC_FORMAT can optionally be included after the
argument list.  Simply include an \'attr\' keyword and the C code to include
in the file.  You have to include braces and anything inside the braces
will be printed into the header file after the function declaration and
before the trailing semicolon.  The braces themselves are not printed.
For example:
.nf

  public void
  print (self, const char *format (check null), ...)
    attr {G_GNUC_PRINTF(2, 3)}

.fi
.PP
This will produce a prototype which will generate a warning at compile
time if the contents of the format argument (argument number 2) aren't
consistent with the types and number of the subsequent variadic
arguments (the first of which is argument number 3).  Only one \'attr\'
keyword per method is allowed.
If you have more than one attribute to include, you should
put them all within the braces.
Note that function attributes were aded in version 2.0.16.
.PP
.B "Error return:"
.PP
Methods which have a return value, there also has to be something
returned if there is an error, such as if a precondition is not met.
The default is 0, casted to the type of the method.  If you need to
return something else then you can specify an \'onerror\' keyword after
the prototype and any optional function attribute macros, and after
that a number, a token (an identifier) or a bit of C code enclosed in
braces {}.  The braces will not be printed into the output, they just
delimit the string.  For example:
.nf

  public void * get_something (self, int i (check >= 0)) onerror NULL {
          ...
  }

.fi
The onerror value is also used in overrides that have a return value, in
case there isn\'t a parent method, PARENT_HANDLER will return it.  More about
this later.
.PP
.B "Default return:"
.PP
Some signal and virtual methods have a return type.  But what happens if
there is no default handler and no one connects to a signal.  GOB2 will
normally have the wrappers return whatever you specify with onerror or \'0\'
if you haven\'t specified anything.  You can also specify a default
return value with the keyword \'defreturn\'.  It\'s use is identical to the
use of onerror, and you can in fact use both at the same time.  Example
.nf

  virtual int get_some_int (self) onerror \-1 defreturn 10 ;

.fi
That is an empty virtual method (in C++ terms a pure virtual).  If you never
specify any handler for it in the derived children it will just return 10.
.PP
.B "Constructor methods:"
.PP
There are two methods that handle the construction of an object, init and
class_init.  You define them by just using the init or class_init keyword 
with an untyped argument in the argument list.  The argument will be
usable in your function as a pointer to your object or class depending if
it\'s init or class_init.
For example:
.nf

  init (self) {
          /* initialize the object here */
          self\->a = 9;
          self\->b = 9;
  }

  class_init (class) {
          /* initialize the class, this is rarely needed */
          class\->blah = NULL;
  }

.fi
The class_init function is very rarely needed as all standard class
initialization is taken care of for you by gob itself.  The init function
should on the other hand be used whenever you need to construct or initialize
anything in the object to put it into a sane state.
.PP
.B "Constructor, dispose, finalize methods:"
.PP
Since 2.0.16, you can also easily add code to the object's constructor,
dispose, and finalize methods.  See GObject documentation on how these are
run.  The code you add will be run before calling the parents function
for dispose and finalize, and after the parent function for constructor.
The syntax is just like init and class_init.
For example:
.nf

  constructor (self) {
        /* constructor method */
  }

  dispose (self) {
        /* dispose method */
  }

  finalize (self) {
        /* finalize method */
  }

.fi
You can also just override those methods as usual, but the above is much easier and nearly as flexible.
.PP
.B "Virtual methods:"
.PP
Virtual methods are basically pointers in the class structure,
so that one can override the method in derived methods.  That is to implement
the method in a derived class, you must then use an override method (more
on those later).
They can be empty
(if you put \';\' instead of the C code).  A wrapper will also be defined
which makes calling the methods he same as public methods.  This type of
method is just a little bit "slower" then normal functions, but not as
slow as signals.  You define them by using "virtual" keyword before the
prototype.  If you put the keyword "private" right after the "virtual"
keyword, the wrapper will not be a public method, but a private one.
You can do the same with "protected" to make a protected wrapper.
.PP
.B "Signals:"
.PP
Signals are methods to which the user can bind other handlers
and override the default handler.  The default handler is basically the
method body.  This is the most versatile and flexible type of a method
and also the slowest.  You need to specify a whole bunch of things when
you define a signal.  One thing is when the default handler will be run,
first or last.  You specify that by "first" or "last" right after the
"signal" keyword.  Then you need to define the GObject enum types (again
without the G_TYPE_ prefix).  For that you define the return types
and the types of arguments after the "self" pointer (not including the
"self" pointer).  You put it in the following syntax "<return type> (<list
of arguments>)".  If the return type is void, the type should be "NONE",
the same should be for the argument list.  The rest of the prototype is
the same as for other method types.  The body can also be empty, and
also there is a public method wrapper which you can use for calling the
signal just like a public method.  Example:
.nf

  signal first INT (POINTER, INT)
  int do_something (self, Gtk:Widget *w (check null type), int length)
  {
          ...
  }
  
.fi
or
.nf

  signal last NONE (NONE) void foo (self);

.fi
.PP
You can include name of signal, if this name is not a C variable name. Example:
.nf

  signal first INT "do-something" (POINTER, INT)
  int do_something (self, Gtk:Widget *w (check null type), int length)
  {
          ...
  }
  
.fi
.PP
If you don\'t want the wrapper that emits the signal to be public, you can
include the keyword "private" after the "signal" keyword.  This will make
the wrapper a normal private method.  You can also make a protected wrapper
by using "protected" instead of "private".
.PP
If you don\'t define a "first" or a "last", the default will be taken as
"last".
.PP
You can also add additional flags.  You do this just like with the argument
flags, although this is probably very rare.  These are the G_SIGNAL_* flags,
and you can add them without the G_SIGNAL_ prefix into a parenthesis, just
after the "signal" keyword.  By default all public signals are G_SIGNAL_ACTION.
.PP
Also gob2 creates a wrapper macros for typesafe signal connection.  That is
you will be warned by the compiler if you pass a callback that is not the
correct prototype.  This will again only warn you on gcc, but it will
compile without warning on another compiler.  So as with all the typesafety
hacks in gob, it is better to test your objects under gcc to get any warnings
even if you are using a different compiler in the end.
.PP
The methods that are created for you are:
.nf

  <class_name>_connect__<signal_name> (<object>, <callback>, <data>)
  <class_name>_connect_after__<signal_name> (<object>, <callback>, <data>)
  <class_name>_connect_data__<signal_name> (<object>, <callback>, <data>,
                                            <destroy_notify>, <flags>)

.fi
.PP
These three functions correspond to the g_signal_connect,
g_signal_connect_after and g_signal_connect_data functions that you would
normally use, except they are for a specific signal.  Also do note
the two underscores between the method name and the signal name.  For
example to connect the signal "foo" on the object "Test:Object" you
would do:
.nf

  test_object_connect__foo (object, callback, data);

.fi
.PP
To use BOXED in the signal arguments you need to tell gob which type of boxed
argument you want to use.  For this you can just add BOXED_GTK_TYPE_STRING
instead of BOXED.  For example BOXED_GTK_TYPE_TREE_ITER for GtkTreeIter.
This works since version 2.0.13.
.PP
.B "Override methods:"
.PP
If you need to override some method (a signal or a virtual method
of some class in the parent tree of the new object), you can define and
override method.  After the "override" keyword, you should put the
typename of the class you are overriding a method from.  Other then that
it is the same as for other methods.  The "self" pointer in this case
should be the type of the method you are overriding so that you don\'t
get warnings during compilation.  Also to call the method of the parent
class, you can use the PARENT_HANDLER macro with your arguments.  Example:
.nf

  override (Gtk:Container) void
  add (Gtk:Container *self (check null type), Gtk:Widget *wid (check null type))
  {
          /* some code here */
          PARENT_HANDLER(self, wid);
  }

.fi
If the function has a return value, then PARENT_HANDLER is an expression that
you can use.  It will return whatever the parent handler returned, or the
"onerror" expression if there was no parent handler.
.PP
.B "Method names:"
.PP
Inside the code, aliases are set for the methods, so that you don\'t
have to type the class name before each call, just type \fBself_\fR instead
of the name of the class.  So to call a method called \fBblah\fR, you
would use the name \fBself_blah\fR.
Example:
.nf

  private int
  foo (self)
  {
          return self\->len;
  }
  
  private int
  bar (self, int i)
  {
          return self_foo (self) + i;
  }

.fi

.SH MAKING NEW OBJECTS
.PP
You should define a new method which should be a normal public method.  Inside
this method, you can use the GET_NEW macro that is defined for you and that
will fetch a new object, so a fairly standard new method would look like:
.nf

  public GObject *
  new (void) {
          GObject *ret = GET_NEW;
          return G_OBJECT (ret);
  }

.fi
.PP
You should not a subtle peculiarity of the GObject system here.  If there is
any code inside the G_OBJECT macro argument, it will get executed multiple
times.  This means that things such as G_OBJECT(GET_NEW) would actually create
4 objects, leaking 3 of them.  A good rule (as with anywhere in C) is to be
careful with all macros.

.SH SELF REFERENCES
.PP
.B "Self alias casts:"
.PP
There are some standard casts defined for you.  Instead of using the full
macros inside the .c file, you can use SELF, IS_SELF and SELF_CLASS.  Using
these makes it easier to for example change class names around.
.PP
.B "Self alias types:"
.PP
There are also the Self and SelfClass types inside
your .c file.  These serve the same function as the above, they make it easier
to type and easier to change typenames around which can help a lot during
prototyping stage.  However you should note that the Self type should not be
used in function prototypes as one of the arguments or as a return value type.
This is because this is a simple C typedef which is only available inside your
\&.c file and not in the header files.  You can disable both the self casting
macros and the self type aliases by passing \-\-no\-self\-alias to gob.

.SH DEALING WITH DIFFERENT GOB VERSIONS
.PP
.B "Defines:"
.PP
In your generated C file, you can use the defines GOB_VERSION_MAJOR
GOB_VERSION_MINOR and GOB_VERSION_PATCHLEVEL if you wish to for example
use a feature that is only available in some newer gob version.  Note however
that you can only use these defines in the C code portions of your .gob file,
and #ifdef\'s cannot span multiple functions.  Check the BUGS section
for more on using the C preprocessor and gob.
.PP
.B "Minimum version requires:"
.PP
You can also make your .gob file require at least certain version of gob.  You
do this by putting \'requires x.y.z\' (where x.y.z is the version number)
outside of any C block, comment or class, usually you should make this the
first line in the file or close to the top.  If gob finds this and the version
of gob used to compile the code is lower then that listed in the require, gob
will generate an error and exit.  For example to require that gob2 version
2.0.0 or higher be used to compile a file, put this at the top of that file:
.nf

  requires 2.0.0

.fi

.SH CREATING NEW ENUM, FLAGS and ERROR TYPES
.PP
You can create new GObject ENUM, FLAGS and GError types for use in your
classes easily.  Glib includes some utilities for handling these, however
it may be cleaner to use the below specified way in your classes.  It also
then doesn\'t require any Makefile setup.  Make sure this is defined in the
same section as the class, that is not in any of the \'%?{\' \'%}\' sections.
.PP
You use the keywords \'enum\' \'flags\' and \'error\' as you would use the
\'class\' keyword.  Then you give a prefix for the values in the enumeration.
Then you define a list of values just like in C.  For \'enum\' types you can
also specify the values assigned to each string.  Then you specify the type in
the standard gob style of specifying types.  Here are a few examples
of all of these:
.nf

  enum LAME_CLIENT {
        IS_CONNECTED,
        NONE = 9,
        LAST
  } Test:Enum;

  flags BUGA_BUGA {
        ONE,
        TWO,
        MANY,
  } Some:Flags;

  error TEST_OBJECT_ERROR {
        BAD_THIS,
        BAD_THAT
  } Test:Object:Error;

.fi
.PP
This will for example define an enum that is equivalent to the following
C code:
.nf

  typedef enum {
        LAME_CLIENT_IS_CONNECTED,
        LAME_CLIENT_NONE = 9,
        LAME_CLIENT_LAST
  } TestEnum;

.fi

.SH C++ MODE
.PP
There is a C++ mode so that gob creates C++ compiler friendly files.  You need
to use the \-\-for\-cpp argument to gob.  This will make the generated file have
a .cc instead of a .c extension, and several things will be adjusted to
make it all work for a C++ compiler.  One thing that will be missing is an
alias to the new method, as that clashes with C++, so instead you\'ll have to
use the full name of the method inside your code.  Also note that gob does
not use any C++ features, this option will just make the generated code
compile with a C++ compiler.

.SH OVERRIDING THE GET_TYPE METHOD
.PP
The get_type is not really a method, but a function which initializes your
object.  Recently objects appeared which require you to make a custom
get_type function.  So it is possible
to override this function.  To do so, just define a new public method called
get_type, with no arguments.  Example:
.nf

  public GType
  get_type (void)
  {
        /* code goes here */
        return some_type;
  }

.fi

.SH INTERFACES
.PP
Currently gob will only allow you to implement interfaces (that is, define new
classes which implement an interface) and doesn\'t yet have support for making
new interfaces, but this will be coming at some point in the future.
.PP
To define a class that implements an interface add a class flag \'interface\'
with the type name of the interface as an argument.  Then to implement
a specific method of the interface, just add \'interface <typename>\'
before the method definition.  The method can, and probably should be,
private.
.PP
Interface methods behave very similarly to virtual/override methods.
As with override methods, you can use PARENT_HANDLER in the definition of an
interface method to call the implementation of the same interface method in
a parent class.
.PP
The NAME_parent_iface variable may be used to access the complete interface
structure from a parent class, where NAME is the implemented interface (with
colons replaced by underscores).
This enables access to the full parent implementation of an interface.
For example, if a class implements the Gtk:Tree:Model interface then the
implementation of the same interface in the parent class is available via
the Gtk_Tree_Model_parent_iface pointer.
If an interface is not implemented in the parent class, then the corresponding
parent_iface value will be a null pointer.
.PP
The following example implements a new object, that implements the
Gtk:Tree:Model interface and implements the get_flags method of that
interface.  Do note that except for standard (GTK+ and glib) specific
interfaces which seem
to have a non-standard name for the interface structure, the structure
should end with and Iface, if you are implementing an interface.  That
is for example for the Gtk:Tree:Model, the structure containing the
table of methods should be named GtkTreeModelIface.
.nf
  class Some:Object from G:Object
          (interface Gtk:Tree:Model)
  {
          /* function implemented for the Gtk:Tree:Model interface */
          interface Gtk:Tree:Model
          private GtkTreeModelFlags
          get_flags (Gtk:Tree:Model *self (check null type))
          {
                /* Here would be the implementation */
                return (GtkTreeModelFlags)0;
          }
  }

.fi
.PP
If you want to implement multiple interfaces just list more class flag lines
as follows:
.nf

  class Some:Object from G:Object
          (interface Gtk:Tree:Model)
          (interface Gtk:Editable)
  {
          /* ... */
  }

.fi

.SH DIRECT BonoboObject SUPPORT
.PP
If you want to build a BonoboObject class gob2 has direct support for these.
Just create a new object that derives from
Bonobo:Object.
Then use a "BonoboObject" class flag with the interface name as an
argument.  The interface name should be as you would type it in C, that is with
underscores as namespace separators.  Then you add the methods (using exact
same names as in the idl file) and prepend those methods with a BonoboObject
keyword.  For example imagine you have an interface GNOME/Foo/SomeInterface,
with a method fooBar that takes a single string:
.nf

  class Foo:Some:Interface from Bonobo:Object
    (BonoboObject GNOME_Foo_SomeInterface) {

          BonoboObject
          private void
          fooBar (PortableServer_Servant servant,
                  const CORBA_char *string,
                  CORBA_Environment *ev)
          {
                  Self *self = SELF (bonobo_object_from_servant (servant));

                  /* your code here */
          }

          /* rest of class */
  }

.fi
Note that the implementation method can be private, in fact that\'s probably
a good idea to do.  It won\'t work to make this a signal, it can however
be a virtual.  Note that the method prototype must match the one from the
interface header file, or you will get a bad assignment warning.  You should
check the header file generated by orbit\-idl and see the epv structure
for the correct prototypes if you can\'t figure them out from the idl itself.
Also note that the first argument is not "self", but the servant and you must
use bonobo_object_from_servant function to get the actual object pointer.

.SH DIRECT LIBGLADE SUPPORT
.PP
Gob can simplify writing a libglade class.  Just create a new object that
derives from a GtkContainer widget.  Then use a "GladeXML" class flag
with the glade file name, root widget and optional domain  as arguments
between double quotes.  For example:
.nf

class My:Glade from Gtk:Window (GladeXML "gob\-libglade.glade" "root")
{
  ....
}

.fi
Note however that then "gob\-libglade.glade" would have to be in the current
directory.  You could specify a path, but that may not work for all
installations.  You can replace the glade filename with a token to be used
in the generated .c file and you can then have a macro with the filename,
as follows:
.nf

class My:Glade from Gtk:Window (GladeXML GLADE_FILE "root")
{
  ....
}

.fi
And somewhere in your header files you would have
.nf

#define GLADE_FILE "/path/to/file.glade"

.fi

You can declare widgets as data members by adding a 'GladeXML' to
the definition.
.nf

private Gtk:Button * button1 GladeXML;

.fi
This will automatically set the "button1" from the GladeXML file.

All signals created with glade are automatically connected if you defined
those class methods in your class.  For example suppose in glade that
we set the "connect" signal on button1 to go to on_button1_clicked, then
in our gob file we can just write:
.nf

public void
on_button1_clicked(self, GtkButton * button)
{
}

.fi

See the examples directory for a full example.  Note that this feature
requires version at least 2.0.12.


.SH IDENTIFIER CONFLICTS
.PP
Gob will need to define some local variables and functions in the generated
files, so you need to take some precaution not to conflict with these.  The
general rule of thumb is that all of these start with three underscores.  There
is one, "parent_class" which doesn\'t because it\'s intended for use in your
code.  For virtuals or signals, you cannot use the identifier __parent__
which is used for the parent of the object.  You should actually never access
__parent__ either as it not guaranteed that it will stay named this way.
Data members cannot be named __parent__ nor _priv.  For methods, you cannot
use the identifiers "init" or "class_init" unless you mean the constructor
methods.  You shouldn\'t generally use 3 underscores even in override method
argument lists and virtual and signal method names as it might confuse the
PARENT_HANDLER macro.  In fact avoiding all names with three underscores is
the best policy when working with gob.
.PP
There are a couple of defines which you shouldn\'t be redefining in the code
or other headers.  These are SELF, IS_SELF, SELF_CLASS, SELF_TYPE, ARG, VAR,
PARENT_HANDLER, GET_NEW, GOB_VERSION_MAJOR, GOB_VERSION_MINOR and
GOB_VERSION_PATCHLEVEL.
.PP
As for types, there are Self and SelfClass types which are only defined in your
source files.  Their generation (just like the generation of the SELF macros)
can be turned off, see command line options.

.SH USING GTK\-DOC STYLE INLINE DOCUMENTATION
.PP
If you want to use gtk\-doc style inline documentation for your objects, you
can do one of two things.  First, you could include the inline documentation
comments in your %{ %} section which will then be put verbatim into the
output source file.  This is the way you should use for functions you define
outside of the class.
.PP
For class methods, you should use a gtk+ style comment, however it can be
indented any number of tabs or spaces and you can use the short method name
without the type prefix.  Gob will automatically try to extract these and
translate to full names and put them in the output source file.  An example
would be:
.nf

  class Gtk:Button:Example from Gtk:Button {
          /**
           * new:
           *
           * Makes a new #GtkButtonExample widget
           *
           * Returns: a new widget
           **/
          public
          GtkWidget *
          new(void)
          {
                  return (GtkWidget *)GET_NEW;
          }
  } 

.fi
If the function you are documenting is a signal or a virtual then it will
be documenting the wrapper that starts that virtual function or emits
that signal.

.SH DEALING WITH CIRCULAR HEADERS
.PP
Sometimes you may need to use an object of type MyObjectA in the MyObjectB
class and vice versa.  Obviously you can\'t include headers for both.  So you
need to just declare the typedef in the header of A for B, and the other way
around as well.  The headers generated include a protecting
define before it declares the typedef.  This define is the
__TYPEDEF_<upper case object name>__.  So inside my\-object\-a.h there will be
this:
.nf

  #ifndef __TYPEDEF_MY_OBJECT_A__
  #define __TYPEDEF_MY_OBJECT_A__
  typedef struct _MyObjectA MyObjectA;
  #endif

.fi
Now instead of including my\-object\-a.h in the header section of
my\-object\-b.gob, just copy the above code there and you\'re set for using
MyObjectA as a type in the method parameters and public types.
.PP
Another way to get out of this problem is if you can use those types only
in the private members, in which case they won\'t be in the generated public
header.

.SH BUILDING WITH MAKE
.PP
If you are using normal makefiles, what you need to do is to add a generic
rule for .gob files.  So you would include the following in the Makefile
and then just use the .c and .h files as usual (make sure the space
before the \'gob2\' is a tab, not spaces):
.nf

  %.c %.h %\-private.h: %.gob
          gob2 $<

.fi

.SH BUILDING WITH AUTOCONF and AUTOMAKE
.PP
This is a little bit more involved.  Basically the first thing to do is to
check for GOB2 in your configure.in file.  You can use the supplied m4 macro
which will also check the version of gob.  Basically you include this:
.nf

  GOB2_CHECK([2.0.0])

.fi
This will replace @GOB2@ in your makefiles with the full path of gob2.  Thus
when adding the generic rule to your Makefile.am file, it should look like:
.nf

  %.c %.h %\-private.h: %.gob
          @GOB2@ $<

.fi
.PP
For Makefile.am you have to set up a couple more things.  First you have to
include the generated .c and .h files into BUILT_SOURCES variable.  You
have to include both the .gob and the .c and .h files in the SOURCES for your
program.

.SH PREVENTING SPURIOUS BUILDS
.PP
When nothing has changed you might not really want to rebuild everything and
gob provides options \-\-no\-touch (since 2.0.13) and \-\-no\-touch\-headers to avoid
this.  When working with build systems such as automake you have to be more
careful as just using those options can cause automake to get confused and you
will need to use something like the following:
.nf

  foo_SOURCES = foo.gob foo.gob.stamp foo.c foo.h foo\-private.h
  BUILT_SOURCES = foo.gob.stamp
  MAINTAINERCLEANFILES = foo.gob.stamp

  %.gob.stamp: %.gob
          @GOB2@ \-\-no\-touch $<
          @touch $@

.fi 

.SH DEBUGGING
.PP
GOB does several things to make debugging the code easier.  First it adds
preprocessor commands into the output c file that point to the correct places
in your .gob input file.  However sometimes there might be some bigger
confusion and this is just not helpful.  In this case you will probably want
to have gcc point you directly at the generated files.  For this use
the \-\-no\-lines command line option.  You should also note that these commands
are not generated for the public header file at all.  If there is an error which
points you to the public header file, make sure you fix this error in the .gob
file, otherwise your changes will not have any effect after gob recompiles the
sources again.
.PP
Sometimes you might want to know which method you are in for some debugging
output.  GOB will define __GOB_FUNCTION__ macro, which is just a string constant
with a pretty name of the method.

.SH M4 SUPPORT
.PP
It is possible to have your .gob file also preprocessed by m4.  This is useful
if you have a lot of files and you\'d like to have some preprocessor put in
some common features.  All you have to do is add \-\-m4 to the command line
of gob2 and gob2 will first run your file through m4.  You can print the
directory that is searched for m4 files by running "gob2 \-\-m4\-dir"
.PP
All the arguments after \-\-m4 will be passed to m4 itself, so it has to be the
last gob2 argument on the command line.  This way you can specify arbitrary
options to pass to m4.

.SH BUGS
.PP
The lexer does not actually parse the C code, so I\'m sure that some corner
cases or maybe even some not so corner cases of C syntax might confuse gob
completely.  If you find any, send me the source that makes it go gaga and
I\'ll try to make the lexer try to handle it properly, but no promises.
.PP
Another thing is that gob ignores preprocessor macros.  Since gob counts
braces, the following code won\'t work:
.nf

  #ifdef SOME_DEFINE
  if(foo) {
  #else
  if(bar) {
  #endif
          blah();
  }

.fi
To make this work, you\'d have to do this:
.nf

  #ifdef SOME_DEFINE
  if(foo)
  #else
  if(bar)
  #endif
  {
          blah();
  }

.fi
There is no real good way we can handle this without parsing C code, so we
probably never will.  In the future, I might add #if 0 as a comment but
that\'s about as far as I can really take it and even that is problematic.
Basically, if you use gob, just don\'t use the C preprocessor too extensively.
And if you use it make sure that you do not cross the boundaries of the C
code segments.
.PP
Comments will not get through to the generated files unless inside C code.
This is not the case for gtk\-doc style comments which are supported.
.PP
The short name aliases are actually implemented as pointers to functions.  Thus
if you want to get the pointer of a function using the short name alias you
can\'t use the \'&\'.  Thus:
.nf

  void (*foo)(Self *);

  /* this will NOT work */
  foo = &self_short_name;

  /* this will work */
  foo = self_short_name;

  /* Both of these will work */
  foo = &my_class_long_name;
  foo = my_class_long_name;

.fi

.SH AUTHOR
.PP
George Lebl <jirka@5z.com>
.PP
GOB2 Homepage: http://www.jirka.org/gob.html