Title: Type System Concepts # Type System Concepts ## Introduction Most modern programming languages come with their own native object systems and additional fundamental algorithmic language constructs. Just as GLib serves as an implementation of such fundamental types and algorithms (linked lists, hash tables and so forth), the GLib Object System provides the required implementations of a flexible, extensible, and intentionally easy to map (into other languages) object-oriented framework for C. The substantial elements that are provided can be summarized as: - A generic type system to register arbitrary single-inherited flat and deep derived types as well as interfaces for structured types. It takes care of creation, initialization and memory management of the assorted object and class structures, maintains parent/child relationships and deals with dynamic implementations of such types. That is, their type specific implementations are relocatable/unloadable during runtime. - A collection of fundamental type implementations, such as integers, doubles, enums and structured types, to name a few. - A sample fundamental type implementation to base object hierarchies upon - the GObject fundamental type. - A signal system that allows very flexible user customization of virtual/overridable object methods and can serve as a powerful notification mechanism. - An extensible parameter/value system, supporting all the provided fundamental types that can be used to generically handle object properties or otherwise parameterized types. ## Background GObject, and its lower-level type system, GType, are used by GTK and most GNOME libraries to provide: - object-oriented C-based APIs and - automatic transparent API bindings to other compiled or interpreted languages. A lot of programmers are used to working with compiled-only or dynamically interpreted-only languages and do not understand the challenges associated with cross-language interoperability. This introduction tries to provide an insight into these challenges and briefly describes the solution chosen by GLib. The following chapters go into greater detail into how GType and GObject work and how you can use them as a C programmer. It is useful to keep in mind that allowing access to C objects from other interpreted languages was one of the major design goals: this can often explain the sometimes rather convoluted APIs and features present in this library. ### Data types and programming One could say that a programming language is merely a way to create data types and manipulate them. Most languages provide a number of language-native types and a few primitives to create more complex types based on these primitive types. In C, the language provides types such as char, long, pointer. During compilation of C code, the compiler maps these language types to the compiler's target architecture machine types. If you are using a C interpreter (assuming one exists), the interpreter (the program which interprets the source code and executes it) maps the language types to the machine types of the target machine at runtime, during the program execution (or just before execution if it uses a Just In Time compiler engine). Perl and Python are interpreted languages which do not really provide type definitions similar to those used by C. Perl and Python programmers manipulate variables and the type of the variables is decided only upon the first assignment or upon the first use which forces a type on the variable. The interpreter also often provides a lot of automatic conversions from one type to the other. For example, in Perl, a variable which holds an integer can be automatically converted to a string given the required context: ```perl my $tmp = 10; print "this is an integer converted to a string:" . $tmp . "\n"; ``` Of course, it is also often possible to explicitly specify conversions when the default conversions provided by the language are not intuitive. ### Exporting a C API C APIs are defined by a set of functions and global variables which are usually exported from a binary. C functions have an arbitrary number of arguments and one return value. Each function is thus uniquely identified by the function name and the set of C types which describe the function arguments and return value. The global variables exported by the API are similarly identified by their name and their type. A C API is thus merely defined by a set of names to which a set of types are associated. If you know the function calling convention and the mapping of the C types to the machine types used by the platform you are on, you can resolve the name of each function to find where the code associated to this function is located in memory, and then construct a valid argument list for the function. Finally, all you have to do is trigger a call to the target C function with the argument list. For the sake of discussion, here is a sample C function and the associated 32 bit x86 assembly code generated by GCC on a Linux computer: ```c static void function_foo (int foo) { } int main (int argc, char *argv[]) { function_foo (10); return 0; } ``` ```asm push $0xa call 0x80482f4 ``` The assembly code shown above is pretty straightforward: the first instruction pushes the hexadecimal value 0xa (decimal value 10) as a 32-bit integer on the stack and calls `function_foo`. As you can see, C function calls are implemented by GCC as native function calls (this is probably the fastest implementation possible). Now, let's say we want to call the C function `function_foo` from a Python program. To do this, the Python interpreter needs to: - Find where the function is located. This probably means finding the binary generated by the C compiler which exports this function. - Load the code of the function in executable memory. - Convert the Python parameters to C-compatible parameters before calling the function. - Call the function with the right calling convention. - Convert the return values of the C function to Python-compatible variables to return them to the Python code. The process described above is pretty complex and there are a lot of ways to make it entirely automatic and transparent to C and Python programmers: - The first solution is to write by hand a lot of glue code, once for each function exported or imported, which does the Python-to-C parameter conversion and the C-to-Python return value conversion. This glue code is then linked with the interpreter which allows Python programs to call Python functions which delegate work to C functions. - Another, nicer solution is to automatically generate the glue code, once for each function exported or imported, with a special compiler which reads the original function signature. The solution used by GLib is to use the GType library which holds at runtime a description of all the objects manipulated by the programmer. This so-called dynamic type library is then used by special generic glue code to automatically convert function parameters and function calling conventions between different runtime domains. The greatest advantage of the solution implemented by GType is that the glue code sitting at the runtime domain boundaries is written once: the figure below states this more clearly. ![](./glue.png) Currently, there exist multiple generic glue code which makes it possible to use C objects written with GType directly in a variety of languages, with a minimum amount of work: there is no need to generate huge amounts of glue code either automatically or by hand. Although that goal was arguably laudable, its pursuit has had a major influence on the whole GType/GObject library. C programmers are likely to be puzzled at the complexity of the features exposed in the following chapters if they forget that the GType/GObject library was not only designed to offer OO-like features to C programmers but also transparent cross-language interoperability. ## The GLib Dynamic Type System A type, as manipulated by the GLib type system, is much more generic than what is usually understood as an Object type. It is best explained by looking at the structure and the functions used to register new types in the type system. ```c typedef struct _GTypeInfo GTypeInfo; struct _GTypeInfo { /* interface types, classed types, instantiated types */ guint16 class_size; GBaseInitFunc base_init; GBaseFinalizeFunc base_finalize; /* classed types, instantiated types */ GClassInitFunc class_init; GClassFinalizeFunc class_finalize; gconstpointer class_data; /* instantiated types */ guint16 instance_size; guint16 n_preallocs; GInstanceInitFunc instance_init; /* value handling */ const GTypeValueTable *value_table; }; GType g_type_register_static (GType parent_type, const gchar *type_name, const GTypeInfo *info, GTypeFlags flags); GType g_type_register_fundamental (GType type_id, const gchar *type_name, const GTypeInfo *info, const GTypeFundamentalInfo *finfo, GTypeFlags flags); ``` `g_type_register_static()`, `g_type_register_dynamic()` and `g_type_register_fundamental()` are the C functions, defined in `gtype.h` and implemented in `gtype.c` which you should use to register a new GType in the program's type system. It is not likely you will ever need to use `g_type_register_fundamental()` but in case you want to, the last chapter explains how to create new fundamental types. Fundamental types are top-level types which do not derive from any other type while other non-fundamental types derive from other types. Upon initialization, the type system not only initializes its internal data structures but it also registers a number of core types: some of these are fundamental types. Others are types derived from these fundamental types. Fundamental and non-fundamental types are defined by: - class size: the `class_size` field in `GTypeInfo`. - class initialization functions (C++ constructor): the `base_init` and `class_init` fields in `GTypeInfo`. - class destruction functions (C++ destructor): the `base_finalize` and `class_finalize` fields in `GTypeInfo`. - instance size (C++ parameter to new): the `instance_size` field in `GTypeInfo`. - instantiation policy (C++ type of new operator): the `n_preallocs` field in `GTypeInfo`. - copy functions (C++ copy operators): the `value_table` field in `GTypeInfo`. - type characteristic flags: `GTypeFlags`. Fundamental types are also defined by a set of `GTypeFundamentalFlags` which are stored in a `GTypeFundamentalInfo`. Non-fundamental types are furthermore defined by the type of their parent which is passed as the `parent_type` parameter to `g_type_register_static()` and `g_type_register_dynamic()`. ## Copy functions The major common point between all GLib types (fundamental and non-fundamental, classed and non-classed, instantiatable and non-instantiatable) is that they can all be manipulated through a single API to copy/assign them. The `GValue` structure is used as an abstract container for all of these types. Its simplistic API (defined in `gobject/gvalue.h`) can be used to invoke the `value_table` functions registered during type registration: for example `g_value_copy()` copies the content of a `GValue` to another `GValue`. This is similar to a C++ assignment which invokes the C++ copy operator to modify the default bit-by-bit copy semantics of C++/C structures/classes. The following code shows how you can copy around a 64 bit integer, as well as a `GObject` instance pointer: ```c static void test_int (void) { GValue a_value = G_VALUE_INIT; GValue b_value = G_VALUE_INIT; guint64 a, b; a = 0xdeadbeef; g_value_init (&a_value, G_TYPE_UINT64); g_value_set_uint64 (&a_value, a); g_value_init (&b_value, G_TYPE_UINT64); g_value_copy (&a_value, &b_value); b = g_value_get_uint64 (&b_value); if (a == b) { g_print ("Yay !! 10 lines of code to copy around a uint64.\n"); } else { g_print ("Are you sure this is not a Z80 ?\n"); } } static void test_object (void) { GObject *obj; GValue obj_vala = G_VALUE_INIT; GValue obj_valb = G_VALUE_INIT; obj = g_object_new (VIEWER_TYPE_FILE, NULL); g_value_init (&obj_vala, VIEWER_TYPE_FILE); g_value_set_object (&obj_vala, obj); g_value_init (&obj_valb, G_TYPE_OBJECT); /* g_value_copy's semantics for G_TYPE_OBJECT types is to copy the reference. * This function thus calls g_object_ref. * It is interesting to note that the assignment works here because * VIEWER_TYPE_FILE is a G_TYPE_OBJECT. */ g_value_copy (&obj_vala, &obj_valb); g_object_unref (G_OBJECT (obj)); g_object_unref (G_OBJECT (obj)); } ``` The important point about the above code is that the exact semantics of the copy calls is undefined since they depend on the implementation of the copy function. Certain copy functions might decide to allocate a new chunk of memory and then to copy the data from the source to the destination. Others might want to simply increment the reference count of the instance and copy the reference to the new `GValue`. The value table used to specify these assignment functions is documented in `GTypeValueTable`. Interestingly, it is also very unlikely you will ever need to specify a `value_table` during type registration because these `value_tables` are inherited from the parent types for non-fundamental types. ## Conventions There are a number of conventions users are expected to follow when creating new types which are to be exported in a header file: - Type names (including object names) must be at least three characters long and start with "a–z", "A–Z" or "\_". - Use the `object_method` pattern for function names: to invoke the method named save on an instance of object type file, call `file_save`. - Use prefixing to avoid namespace conflicts with other projects. If your library (or application) is named `Viewer`, prefix all your function names with viewer_. For example: `viewer_object_method`. - Create a macro named `PREFIX_TYPE_OBJECT` which always returns the GType for the associated object type. For an object of type `File` in the `Viewer` namespace, use: `VIEWER_TYPE_FILE`. This macro is implemented using a function named `prefix_object_get_type`; for example, `viewer_file_get_type`. - Use `G_DECLARE_FINAL_TYPE` or `G_DECLARE_DERIVABLE_TYPE` to define various other conventional macros for your object: - `PREFIX_OBJECT (obj)`, which returns a pointer of type `PrefixObject`. This macro is used to enforce static type safety by doing explicit casts wherever needed. It also enforces dynamic type safety by doing runtime checks. It is possible to disable the dynamic type checks in production builds (see "Building GLib" section in the GLib API reference). For example, we would create `VIEWER_FILE (obj)` to keep the previous example. - `PREFIX_OBJECT_CLASS (klass)`, which is strictly equivalent to the previous casting macro: it does static casting with dynamic type checking of class structures. It is expected to return a pointer to a class structure of type `PrefixObjectClass`. An example is: `VIEWER_FILE_CLASS`. - `PREFIX_IS_OBJECT (obj)`, which returns a gboolean which indicates whether the input object instance pointer is non-`NULL` and of type `OBJECT`. For example, `VIEWER_IS_FILE`. - `PREFIX_IS_OBJECT_CLASS (klass)`, which returns a boolean if the input class pointer is a pointer to a class of type `OBJECT`. For example, `VIEWER_IS_FILE_CLASS`. - `PREFIX_OBJECT_GET_CLASS (obj)`, which returns the class pointer associated to an instance of a given type. This macro is used for static and dynamic type safety purposes (just like the previous casting macros). For example, `VIEWER_FILE_GET_CLASS`. The implementation of these macros is pretty straightforward: a number of simple-to-use macros are provided in `gtype.h`. For the example we used above, we would write the following trivial code to declare the macros: ```c #define VIEWER_TYPE_FILE viewer_file_get_type() G_DECLARE_FINAL_TYPE (ViewerFile, viewer_file, VIEWER, FILE, GObject) ``` Unless your code has special requirements, you can use the `G_DEFINE_TYPE` macro to define a class: ```c G_DEFINE_TYPE (ViewerFile, viewer_file, G_TYPE_OBJECT) ``` Otherwise, the `viewer_file_get_type` function must be implemented manually: ```c GType viewer_file_get_type (void) { static GType type = 0; if (type == 0) { const GTypeInfo info = { /* You fill this structure. */ }; type = g_type_register_static (G_TYPE_OBJECT, "ViewerFile", &info, 0); } return type; } ``` ## Non-instantiatable non-classed fundamental types A lot of types are not instantiatable by the type system and do not have a class. Most of these types are fundamental trivial types such as `gchar`, and are already registered by GLib. In the rare case of needing to register such a type in the type system, fill a `GTypeInfo` structure with zeros since these types are also most of the time fundamental: ```c GTypeInfo info = { .class_size = 0, .base_init = NULL, .base_finalize = NULL, .class_init = NULL, .class_finalize = NULL, .class_data = NULL, .instance_size = 0, .n_preallocs = 0, .instance_init = NULL, .value_table = NULL, }; static const GTypeValueTable value_table = { .value_init = value_init_long0, .value_free = NULL, .value_copy = value_copy_long0, .value_peek_pointer = NULL, .collect_format = "i", .collect_value = value_collect_int, .lcopy_format = "p", .lcopy_value = value_lcopy_char, }; info.value_table = &value_table; type = g_type_register_fundamental (G_TYPE_CHAR, "gchar", &info, &finfo, 0); ``` Having non-instantiatable types might seem a bit useless: what good is a type if you cannot instantiate an instance of that type? Most of these types are used in conjunction with `GValue`s: a `GValue` is initialized with an integer or a string and it is passed around by using the registered type's `value_table`. `GValue`s (and by extension these trivial fundamental types) are most useful when used in conjunction with object properties and signals. ## Instantiatable classed types: objects Types which are registered with a class and are declared instantiatable are what most closely resembles an object. Although `GObject`s are the most well known type of instantiatable classed types, other kinds of similar objects used as the base of an inheritance hierarchy have been externally developed and they are all built on the fundamental features described below. For example, the code below shows how you could register such a fundamental object type in the type system (using none of the GObject convenience API): ```c typedef struct { GObject parent_instance; /* instance members */ char *filename; } ViewerFile; typedef struct { GObjectClass parent_class; /* class members */ /* the first is public, pure and virtual */ void (*open) (ViewerFile *self, GError **error); /* the second is public and virtual */ void (*close) (ViewerFile *self, GError **error); } ViewerFileClass; #define VIEWER_TYPE_FILE (viewer_file_get_type ()) GType viewer_file_get_type (void) { static GType type = 0; if (type == 0) { const GTypeInfo info = { .class_size = sizeof (ViewerFileClass), .base_init = NULL, .base_finalize = NULL, .class_init = (GClassInitFunc) viewer_file_class_init, .class_finalize = NULL, .class_data = NULL, .instance_size = sizeof (ViewerFile), .n_preallocs = 0, .instance_init = (GInstanceInitFunc) viewer_file_init, }; type = g_type_register_static (G_TYPE_OBJECT, "ViewerFile", &info, 0); } return type; } ``` Upon the first call to `viewer_file_get_type`, the type named `ViewerFile` will be registered in the type system as inheriting from the type `G_TYPE_OBJECT`. Every object must define two structures: its class structure and its instance structure. All class structures must contain as first member a `GTypeClass` structure. All instance structures must contain as first member a `GTypeInstance` structure. The declaration of these C types, coming from `gtype.h` is shown below: ``` struct _GTypeClass { GType g_type; }; struct _GTypeInstance { GTypeClass *g_class; }; ``` These constraints allow the type system to make sure that every object instance (identified by a pointer to the object's instance structure) contains in its first bytes a pointer to the object's class structure. This relationship is best explained by an example: let's take object B which inherits from object A: ```c /* A definitions */ typedef struct { GTypeInstance parent; int field_a; int field_b; } A; typedef struct { GTypeClass parent_class; void (*method_a) (void); void (*method_b) (void); } AClass; /* B definitions. */ typedef struct { A parent; int field_c; int field_d; } B; typedef struct { AClass parent_class; void (*method_c) (void); void (*method_d) (void); } BClass; ``` The C standard mandates that the first field of a C structure is stored starting in the first byte of the buffer used to hold the structure's fields in memory. This means that the first field of an instance of an object B is A's first field which in turn is `GTypeInstance`'s first field which in turn is `g_class`, a pointer to B's class structure. Thanks to these simple conditions, it is possible to detect the type of every object instance by doing: ```c B *b; b->parent.parent.g_class->g_type ``` or, more compactly: ```c B *b; ((GTypeInstance *) b)->g_class->g_type ``` ### Initialization and destruction Instantiation of these types can be done with `g_type_create_instance()`, which will look up the type information structure associated with the type requested. Then, the instance size and instantiation policy (if the `n_preallocs` field is set to a non-zero value, the type system allocates the object's instance structures in chunks rather than mallocing for every instance) declared by the user are used to get a buffer to hold the object's instance structure. If this is the first instance of the object ever created, the type system must create a class structure. It allocates a buffer to hold the object's class structure and initializes it. The first part of the class structure (ie: the embedded parent class structure) is initialized by copying the contents from the class structure of the parent class. The rest of class structure is initialized to zero. If there is no parent, the entire class structure is initialized to zero. The type system then invokes the `base_init` functions `(GBaseInitFunc)` from topmost fundamental object to bottom-most most derived object. The object's `class_init` `(GClassInitFunc)` function is invoked afterwards to complete initialization of the class structure. Finally, the object's interfaces are initialized (we will discuss interface initialization in more detail later). Once the type system has a pointer to an initialized class structure, it sets the object's instance class pointer to the object's class structure and invokes the object's `instance_init` `(GInstanceInitFunc)` functions, from top-most fundamental type to bottom-most most-derived type. Object instance destruction through `g_type_free_instance()` is very simple: the instance structure is returned to the instance pool if there is one and if this was the last living instance of the object, the class is destroyed. Class destruction (the concept of destruction is sometimes partly referred to as finalization in GType) is the symmetric process of the initialization: interfaces are destroyed first. Then, the most derived `class_finalize` `(GClassFinalizeFunc)` function is invoked. Finally, the `base_finalize` `(GBaseFinalizeFunc)` functions are invoked from bottom-most most-derived type to top-most fundamental type and the class structure is freed. The base initialization/finalization process is very similar to the C++ constructor/destructor paradigm. The practical details are different though and it is important not to get confused by superficial similarities. GTypes have no instance destruction mechanism. It is the user's responsibility to implement correct destruction semantics on top of the existing GType code. (This is what `GObject` does) Furthermore, C++ code equivalent to the `base_init` and `class_init` callbacks of GType is usually not needed because C++ cannot really create object types at runtime. The instantiation/finalization process can be summarized as follows:
Invocation time Function invoked Function's parameters
First call to g_type_create_instance() for target type type's base_init function On the inheritance tree of classes from fundamental type to target type. base_init is invoked once for each class structure.
target type's class_init function On target type's class structure
interface initialization, see the section called “Interface Initialization”
Each call to g_type_create_instance() for target type target type's instance_init function On object's instance
Last call to g_type_free_instance() for target type interface destruction, see the section called “Interface Destruction”
target type's class_finalize function On target type's class structure
type's base_finalize function On the inheritance tree of classes from fundamental type to target type. base_finalize is invoked once for each class structure.
## Non-instantiatable classed types: interfaces GType's interfaces are very similar to Java's interfaces. They allow to describe a common API that several classes will adhere to. Imagine the play, pause and stop buttons on hi-fi equipment—those can be seen as a playback interface. Once you know what they do, you can control your CD player, MP3 player or anything that uses these symbols. To declare an interface you have to register a non-instantiatable classed type which derives from `GTypeInterface`. The following piece of code declares such an interface: ```c #define VIEWER_TYPE_EDITABLE viewer_editable_get_type () G_DECLARE_INTERFACE (ViewerEditable, viewer_editable, VIEWER, EDITABLE, GObject) struct _ViewerEditableInterface { GTypeInterface parent; void (*save) (ViewerEditable *self, GError **error); }; void viewer_editable_save (ViewerEditable *self, GError **error); ``` The interface function, `viewer_editable_save` is implemented in a pretty simple way: ```c void viewer_editable_save (ViewerEditable *self, GError **error) { ViewerEditableinterface *iface; g_return_if_fail (VIEWER_IS_EDITABLE (self)); g_return_if_fail (error == NULL || *error == NULL); iface = VIEWER_EDITABLE_GET_IFACE (self); g_return_if_fail (iface->save != NULL); iface->save (self); } ``` `viewer_editable_get_type` registers a type named `ViewerEditable` which inherits from `G_TYPE_INTERFACE`. All interfaces must be children of `G_TYPE_INTERFACE` in the inheritance tree. An interface is defined by only one structure which must contain as first member a `GTypeInterface` structure. The interface structure is expected to contain the function pointers of the interface methods. It is good style to define helper functions for each of the interface methods which simply call the interface's method directly: `viewer_editable_save` is one of these. If you have no special requirements you can use the `G_IMPLEMENT_INTERFACE` macro to implement an interface: ```c static void viewer_file_save (ViewerEditable *self) { g_print ("File implementation of editable interface save method.\n"); } static void viewer_file_editable_interface_init (ViewerEditableInterface *iface) { iface->save = viewer_file_save; } G_DEFINE_TYPE_WITH_CODE (ViewerFile, viewer_file, VIEWER_TYPE_FILE, G_IMPLEMENT_INTERFACE (VIEWER_TYPE_EDITABLE, viewer_file_editable_interface_init)) ``` If your code does have special requirements, you must write a custom `get_type` function to register your GType which inherits from some GObject and which implements the interface `ViewerEditable`. For example, this code registers a new `ViewerFile` class which implements `ViewerEditable`: ```c static void viewer_file_save (ViewerEditable *editable) { g_print ("File implementation of editable interface save method.\n"); } static void viewer_file_editable_interface_init (gpointer g_iface, gpointer iface_data) { ViewerEditableInterface *iface = g_iface; iface->save = viewer_file_save; } GType viewer_file_get_type (void) { static GType type = 0; if (type == 0) { const GTypeInfo info = { .class_size = sizeof (ViewerFileClass), .base_init = NULL, .base_finalize = NULL, .class_init = (GClassInitFunc) viewer_file_class_init, .class_finalize = NULL, .class_data = NULL, .instance_size = sizeof (ViewerFile), .n_preallocs = 0, .instance_init = (GInstanceInitFunc) viewer_file_init }; const GInterfaceInfo editable_info = { .interface_init = (GInterfaceInitFunc) viewer_file_editable_interface_init, .interface_finalize = NULL, .interface_data = NULL, }; type = g_type_register_static (VIEWER_TYPE_FILE, "ViewerFile", &info, 0); g_type_add_interface_static (type, VIEWER_TYPE_EDITABLE, &editable_info); } return type; } ``` `g_type_add_interface_static()` records in the type system that the given `ViewerFile` type implements also `ViewerEditable` (`viewer_editable_get_type()` returns the type of `ViewerEditable`). The `GInterfaceInfo` structure holds information about the implementation of the interface: ```c struct _GInterfaceInfo { GInterfaceInitFunc interface_init; GInterfaceFinalizeFunc interface_finalize; gpointer interface_data; }; ``` ### Interface initialization When an instantiatable classed type which implements an interface (either directly or by inheriting an implementation from a superclass) is created for the first time, its class structure is initialized following the process described in the section called "Instantiatable classed types: objects". After that, the interface implementations associated with the type are initialized. First a memory buffer is allocated to hold the interface structure. The parent's interface structure is then copied over to the new interface structure (the parent interface is already initialized at that point). If there is no parent interface, the interface structure is initialized with zeros. The `g_type` and the `g_instance_type` fields are then initialized: `g_type` is set to the type of the most-derived interface and `g_instance_type` is set to the type of the most derived type which implements this interface. The interface's `base_init` function is called, and then the interface's `default_init` is invoked. Finally if the type has registered an implementation of the interface, the implementation's `interface_init` function is invoked. If there are multiple implementations of an interface the `base_init` and `interface_init` functions will be invoked once for each implementation initialized. It is thus recommended to use a `default_init` function to initialize an interface. This function is called only once for the interface no matter how many implementations there are. The `default_init` function is declared by `G_DEFINE_INTERFACE` which can be used to define the interface: ```c G_DEFINE_INTERFACE (ViewerEditable, viewer_editable, G_TYPE_OBJECT) static void viewer_editable_default_init (ViewerEditableInterface *iface) { /* add properties and signals here, will only be called once */ } ``` Or you can do that yourself in a GType function for your interface: ```c GType viewer_editable_get_type (void) { static gsize type_id = 0; if (g_once_init_enter (&type_id)) { const GTypeInfo info = { sizeof (ViewerEditableInterface), NULL, /* base_init */ NULL, /* base_finalize */ viewer_editable_default_init, /* class_init */ NULL, /* class_finalize */ NULL, /* class_data */ 0, /* instance_size */ 0, /* n_preallocs */ NULL /* instance_init */ }; GType type = g_type_register_static (G_TYPE_INTERFACE, "ViewerEditable", &info, 0); g_once_init_leave (&type_id, type); } return type_id; } static void viewer_editable_default_init (ViewerEditableInterface *iface) { /* add properties and signals here, will only called once */ } ``` In summary, interface initialization uses the following functions:
Invocation time Function Invoked Function's parameters Remark
First call to g_type_create_instance() for any type implementing interface interface's base_init function On interface's vtable Rarely necessary to use this. Called once per instantiated classed type implementing the interface.
First call to g_type_create_instance() for each type implementing interface interface's default_init function On interface's vtable Register interface's signals, properties, etc. here. Will be called once.
First call to g_type_create_instance() for any type implementing interface implementation's interface_init function On interface's vtable Initialize interface implementation. Called for each class that that implements the interface. Initialize the interface method pointers in the interface structure to the implementing class's implementation.
### Interface Destruction When the last instance of an instantiatable type which registered an interface implementation is destroyed, the interface's implementations associated to the type are destroyed. To destroy an interface implementation, GType first calls the implementation's `interface_finalize` function and then the interface's most-derived `base_finalize` function. Again, it is important to understand, as in the section called "Interface Initialization", that both `interface_finalize` and `base_finalize` are invoked exactly once for the destruction of each implementation of an interface. Thus, if you were to use one of these functions, you would need to use a static integer variable which would hold the number of instances of implementations of an interface such that the interface's class is destroyed only once (when the integer variable reaches zero). The above process can be summarized as follows:
Invocation time Function Invoked Function's parameters
Last call to g_type_free_instance() for type implementing interface interface's interface_finalize function On interface's vtable
interface's base_finalize function On interface's vtable
## The GObject base class The previous chapter discussed the details of GLib's Dynamic Type System. The GObject library also contains an implementation for a base fundamental type named `GObject`. `GObject` is a fundamental classed instantiatable type. It implements: - memory management with reference counting - construction/Destruction of instances - generic per-object properties with set/get function pairs - easy use of signals All the GNOME libraries which use the GLib type system (like GTK and GStreamer) inherit from `GObject` which is why it is important to understand the details of how it works. ### Object instantiation The `g_object_new()` family of functions can be used to instantiate any GType which inherits from the GObject base type. All these functions make sure the class and instance structures have been correctly initialized by GLib's type system and then invoke at one point or another the constructor class method which is used to: - allocate and clear memory through `g_type_create_instance()` - initialize the object's instance with the construction properties. GObject explicitly guarantees that all class and instance members (except the fields pointing to the parents) to be set to zero. Once all construction operations have been completed and constructor properties set, the constructed class method is called. Objects which inherit from `GObject` are allowed to override this constructed class method. The example below shows how `ViewerFile` overrides the parent's construction process: ```c #define VIEWER_TYPE_FILE viewer_file_get_type () G_DECLARE_FINAL_TYPE (ViewerFile, viewer_file, VIEWER, FILE, GObject) struct _ViewerFile { GObject parent_instance; /* instance members */ char *filename; guint zoom_level; }; /* will create viewer_file_get_type and set viewer_file_parent_class */ G_DEFINE_TYPE (ViewerFile, viewer_file, G_TYPE_OBJECT) static void viewer_file_constructed (GObject *obj) { /* update the object state depending on constructor properties */ /* Always chain up to the parent constructed function to complete object * initialisation. */ G_OBJECT_CLASS (viewer_file_parent_class)->constructed (obj); } static void viewer_file_finalize (GObject *obj) { ViewerFile *self = VIEWER_FILE (obj); g_free (self->filename); /* Always chain up to the parent finalize function to complete object * destruction. */ G_OBJECT_CLASS (viewer_file_parent_class)->finalize (obj); } static void viewer_file_class_init (ViewerFileClass *klass) { GObjectClass *object_class = G_OBJECT_CLASS (klass); object_class->constructed = viewer_file_constructed; object_class->finalize = viewer_file_finalize; } static void viewer_file_init (ViewerFile *self) { /* initialize the object */ } ``` If the user instantiates an object `ViewerFile` with: ```c ViewerFile *file = g_object_new (VIEWER_TYPE_FILE, NULL); ``` If this is the first instantiation of such an object, the `viewer_file_class_init` function will be invoked after any `viewer_file_base_class_init` function. This will make sure the class structure of this new object is correctly initialized. Here, `viewer_file_class_init` is expected to override the object's class methods and setup the class' own methods. In the example above, the constructed method is the only overridden method: it is set to `viewer_file_constructed`. Once `g_object_new()` has obtained a reference to an initialized class structure, it invokes its constructor method to create an instance of the new object, if the constructor has been overridden in `viewer_file_class_init`. Overridden constructors must chain up to their parent’s constructor. In order to find the parent class and chain up to the parent class constructor, we can use the `viewer_file_parent_class` pointer that has been set up for us by the `G_DEFINE_TYPE` macro. Finally, at one point or another, `g_object_constructor` is invoked by the last constructor in the chain. This function allocates the object's instance buffer through `g_type_create_instance()` which means that the `instance_init` function is invoked at this point if one was registered. After `instance_init` returns, the object is fully initialized and should be ready to have its methods called by the user. When `g_type_create_instance()` returns, `g_object_constructor` sets the construction properties (i.e. the properties which were given to `g_object_new()`) and returns to the user's constructor. The process described above might seem a bit complicated, but it can be summarized easily by the table below which lists the functions invoked by `g_object_new()` and their order of invocation:
Invocation time Function invoked Function's parameters Remark
First call to g_object_new() for target type target type's base_init function On the inheritance tree of classes from fundamental type to target type. base_init is invoked once for each class structure. Never used in practice. Unlikely you will need it.
target type's class_init function On target type's class structure Here, you should make sure to initialize or override class methods (that is, assign to each class' method its function pointer) and create the signals and the properties associated to your object.
interface's base_init function On interface's vtable
interface's interface_init function On interface's vtable
Each call to g_object_new() for target type target type's class constructor method: GObjectClass->constructor On object's instance If you need to handle construct properties in a custom way, or implement a singleton class, override the constructor method and make sure to chain up to the object's parent class before doing your own initialization. In doubt, do not override the constructor method.
type's instance_init function On the inheritance tree of classes from fundamental type to target type. The instance_init provided for each type is invoked once for each instance structure. Provide an instance_init function to initialize your object before its construction properties are set. This is the preferred way to initialize a GObject instance. This function is equivalent to C++ constructors.
target type's class constructed method: GObjectClass->constructed On object's instance If you need to perform object initialization steps after all construct properties have been set. This is the final step in the object initialization process, and is only called if the constructor method returned a new object instance (rather than, for example, an existing singleton).
Readers should feel concerned about one little twist in the order in which functions are invoked: while, technically, the class' constructor method is called before the GType's `instance_init` function (since `g_type_create_instance()` which calls `instance_init` is called by `g_object_constructor` which is the top-level class constructor method and to which users are expected to chain to), the user's code which runs in a user-provided constructor will always run after GType's `instance_init` function since the user-provided constructor must (you've been warned) chain up before doing anything useful. ### Object memory management The memory-management API for GObjects is a bit complicated but the idea behind it is pretty simple: the goal is to provide a flexible model based on reference counting which can be integrated in applications which use or require different memory management models (such as garbage collection). The methods which are used to manipulate this reference count are described below. #### Reference count The functions `g_object_ref()` and `g_object_unref()` increase and decrease the reference count, respectively. These functions are thread-safe. `g_clear_object()` is a convenience wrapper around `g_object_unref()` which also clears the pointer passed to it. The reference count is initialized to one by `g_object_new()` which means that the caller is currently the sole owner of the newly-created reference. (If the object is derived from `GInitiallyUnowned`, this reference is "floating", and must be "sunk", i.e. transformed into a real reference.) When the reference count reaches zero, that is, when `g_object_unref()` is called by the last owner of a reference to the object, the `dispose()` and the `finalize()` class methods are invoked. Finally, after `finalize()` is invoked, `g_type_free_instance()` is called to free the object instance. Depending on the memory allocation policy decided when the type was registered (through one of the `g_type_register_*` functions), the object's instance memory will be freed or returned to the object pool for this type. Once the object has been freed, if it was the last instance of the type, the type's class will be destroyed as described in the section called "Instantiatable classed types: objects" and the section called "Non-instantiatable classed types: interfaces". The table below summarizes the destruction process of a `GObject`:
Invocation time Function invoked Function's parameters Remark
Last call to g_object_unref() for an instance of target type target type's dispose class function GObject instance When dispose ends, the object should not hold any reference to any other member object. The object is also expected to be able to answer client method invocations (with possibly an error code but no memory violation) until finalize is executed. dispose can be executed more than once. dispose should chain up to its parent implementation just before returning to the caller.
target type's finalize class function GObject instance Finalize is expected to complete the destruction process initiated by dispose. It should complete the object's destruction. finalize will be executed only once. finalize should chain up to its parent implementation just before returning to the caller. See the section on "Reference counts and cycles" for more information.
Last call to g_object_unref() for the last instance of target type interface's interface_finalize function On interface's vtable Never used in practice. Unlikely you will need it.
interface's base_finalize function On interface's vtable Never used in practice. Unlikely you will need it.
target type's class_finalize function On target type's class structure Never used in practice. Unlikely you will need it.
type's base_finalize function On the inheritance tree of classes from fundamental type to target type. base_init is invoked once for each class structure. Never used in practice. Unlikely you will need it.
#### Weak References Weak references are used to monitor object finalization: `g_object_weak_ref()` adds a monitoring callback which does not hold a reference to the object but which is invoked when the object runs its dispose method. Weak references on the object are automatically dropped when the instance is disposed, so there is no need to invoke `g_object_weak_unref()` from the `GWeakNotify` callback. Remember that the object instance is not passed to the `GWeakNotify` callback because the object has already been disposed. Instead, the callback receives a pointer to where the object previously was. Weak references are also used to implement `g_object_add_weak_pointer()` and `g_object_remove_weak_pointer()`. These functions add a weak reference to the object they are applied to which makes sure to nullify the pointer given by the user when object is finalized. Similarly, `GWeakRef` can be used to implement weak references if thread safety is required. #### Reference counts and cycles GObject's memory management model was designed to be easily integrated in existing code using garbage collection. This is why the destruction process is split in two phases: the first phase, executed in the `dispose()` handler is supposed to release all references to other member objects. The second phase, executed by the `finalize()` handler is supposed to complete the object's destruction process. Object methods should be able to run without program error in-between the two phases. This two-step destruction process is very useful to break reference counting cycles. While the detection of the cycles is up to the external code, once the cycles have been detected, the external code can invoke `g_object_run_dispose()` which will indeed break any existing cycles since it will run the dispose handler associated to the object and thus release all references to other objects. This explains one of the rules about the `dispose()` handler stated earlier: the `dispose()` handler can be invoked multiple times. Let's say we have a reference count cycle: object A references B which itself references object A. Let's say we have detected the cycle and we want to destroy the two objects. One way to do this would be to invoke `g_object_run_dispose()` on one of the objects. If object A releases all its references to all objects, this means it releases its reference to object B. If object B was not owned by anyone else, this is its last reference count which means this last unref runs B's dispose handler which, in turn, releases B's reference on object A. If this is A's last reference count, this last unref runs A's dispose handler which is running for the second time before A's finalize handler is invoked! The above example, which might seem a bit contrived, can really happen if GObjects are being handled by language bindings—hence the rules for object destruction should be closely followed. ### Object properties One of GObject's nice features is its generic get/set mechanism for object properties. When an object is instantiated, the object's `class_init` handler should be used to register the object's properties with `g_object_class_install_properties()`. The best way to understand how object properties work is by looking at a real example of how it is used: ```c // Implementation typedef enum { PROP_FILENAME = 1, PROP_ZOOM_LEVEL, N_PROPERTIES } ViewerFileProperty; static GParamSpec *obj_properties[N_PROPERTIES] = { NULL, }; static void viewer_file_set_property (GObject *object, guint property_id, const GValue *value, GParamSpec *pspec) { ViewerFile *self = VIEWER_FILE (object); switch ((ViewerFileProperty) property_id) { case PROP_FILENAME: g_free (self->filename); self->filename = g_value_dup_string (value); g_print ("filename: %s\n", self->filename); break; case PROP_ZOOM_LEVEL: self->zoom_level = g_value_get_uint (value); g_print ("zoom level: %u\n", self->zoom_level); break; default: /* We don't have any other property... */ G_OBJECT_WARN_INVALID_PROPERTY_ID (object, property_id, pspec); break; } } static void viewer_file_get_property (GObject *object, guint property_id, GValue *value, GParamSpec *pspec) { ViewerFile *self = VIEWER_FILE (object); switch ((ViewerFileProperty) property_id) { case PROP_FILENAME: g_value_set_string (value, self->filename); break; case PROP_ZOOM_LEVEL: g_value_set_uint (value, self->zoom_level); break; default: /* We don't have any other property... */ G_OBJECT_WARN_INVALID_PROPERTY_ID (object, property_id, pspec); break; } } static void viewer_file_class_init (ViewerFileClass *klass) { GObjectClass *object_class = G_OBJECT_CLASS (klass); object_class->set_property = viewer_file_set_property; object_class->get_property = viewer_file_get_property; obj_properties[PROP_FILENAME] = g_param_spec_string ("filename", "Filename", "Name of the file to load and display from.", NULL /* default value */, G_PARAM_CONSTRUCT_ONLY | G_PARAM_READWRITE); obj_properties[PROP_ZOOM_LEVEL] = g_param_spec_uint ("zoom-level", "Zoom level", "Zoom level to view the file at.", 0 /* minimum value */, 10 /* maximum value */, 2 /* default value */, G_PARAM_READWRITE); g_object_class_install_properties (object_class, N_PROPERTIES, obj_properties); } ``` ```c // Use ViewerFile *file; GValue val = G_VALUE_INIT; file = g_object_new (VIEWER_TYPE_FILE, NULL); g_value_init (&val, G_TYPE_UINT); g_value_set_char (&val, 11); g_object_set_property (G_OBJECT (file), "zoom-level", &val); g_value_unset (&val); ``` The client code above looks simple but a lot of things happen under the hood: `g_object_set_property()` first ensures a property with this name was registered in file's `class_init` handler. If so it walks the class hierarchy, from bottom-most most-derived type, to top-most fundamental type to find the class which registered that property. It then tries to convert the user-provided GValue into a GValue whose type is that of the associated property. If the user provides a signed char GValue, as is shown here, and if the object's property was registered as an `unsigned int`, `g_value_transform()` will try to transform the input signed char into an unsigned int. Of course, the success of the transformation depends on the availability of the required transform function. In practice, there will almost always be a transformation which matches and conversion will be carried out if needed. After transformation, the GValue is validated by `g_param_value_validate()` which makes sure the user's data stored in the GValue matches the characteristics specified by the property's GParamSpec. Here, the GParamSpec we provided in `class_init` has a validation function which makes sure that the GValue contains a value which respects the minimum and maximum bounds of the GParamSpec. In the example above, the client's GValue does not respect these constraints (it is set to 11, while the maximum is 10). As such, the `g_object_set_property()` function will return with an error. If the user's GValue had been set to a valid value, `g_object_set_property()` would have proceeded with calling the object's `set_property` class method. Here, since our implementation of ViewerFile did override this method, execution would jump to `viewer_file_set_property` after having retrieved from the GParamSpec the `param_id` which had been stored by `g_object_class_install_property()`. Once the property has been set by the object's `set_property` class method, execution returns to `g_object_set_property()` which makes sure that the "notify" signal is emitted on the object's instance with the changed property as parameter unless notifications were frozen by `g_object_freeze_notify()`. `g_object_thaw_notify()` can be used to re-enable notification of property modifications through the “notify” signal. It is important to remember that even if properties are changed while property change notification is frozen, the "notify" signal will be emitted once for each of these changed properties as soon as the property change notification is thawed: no property change is lost for the "notify" signal, although multiple notifications for a single property are compressed. Signals can only be delayed by the notification freezing mechanism. It sounds like a tedious task to set up GValues every time when one wants to modify a property. In practice one will rarely do this. The functions `g_object_set_property()` and `g_object_get_property()` are meant to be used by language bindings. For application there is an easier way and that is described next. ### Accessing multiple properties at once It is interesting to note that the `g_object_set()` and `g_object_set_valist()` (variadic version) functions can be used to set multiple properties at once. The client code shown above can then be re-written as: ```c ViewerFile *file; file = /* */; g_object_set (G_OBJECT (file), "zoom-level", 6, "filename", "~/some-file.txt", NULL); ``` This saves us from managing the GValues that we were needing to handle when using `g_object_set_property()`. The code above will trigger one notify signal emission for each property modified. Equivalent `_get` versions are also available: `g_object_get()` and `g_object_get_valist()` (variadic version) can be used to get numerous properties at once. These high level functions have one drawback — they don't provide a return value. One should pay attention to the argument types and ranges when using them. A known source of errors is to pass a different type from what the property expects; for instance, passing an integer when the property expects a floating point value and thus shifting all subsequent parameters by some number of bytes. Also forgetting the terminating NULL will lead to undefined behaviour. This explains how `g_object_new()`, `g_object_newv()` and `g_object_new_valist()` work: they parse the user-provided variable number of parameters and invoke `g_object_set()` on the parameters only after the object has been successfully constructed. The "notify" signal will be emitted for each property set. ## The GObject messaging system ### Closures Closures are central to the concept of asynchronous signal delivery which is widely used throughout GTK and GNOME applications. A closure is an abstraction, a generic representation of a callback. It is a small structure which contains three objects: - a function pointer (the callback itself) whose prototype looks like: `return_type function_callback (... , gpointer user_data);` - the `user_data` pointer which is passed to the callback upon invocation of the closure - a function pointer which represents the destructor of the closure: whenever the closure's refcount reaches zero, this function will be called before the closure structure is freed The `GClosure` structure represents the common functionality of all closure implementations: there exists a different closure implementation for each separate runtime which wants to use the GObject type system. The GObject library provides a simple GCClosure type which is a specific implementation of closures to be used with C/C++ callbacks. A `GClosure` provides simple services: - invocation (`g_closure_invoke()`): this is what closures were created for; they hide the details of callback invocation from the callback invoker. - notification: the closure notifies listeners of certain events such as closure invocation, closure invalidation and closure finalization. Listeners can be registered with `g_closure_add_finalize_notifier()` (finalization notification), `g_closure_add_invalidate_notifier()` (invalidation notification) and `g_closure_add_marshal_guards()` (invocation notification). There exist symmetric deregistration functions for finalization and invalidation events (`g_closure_remove_finalize_notifier()` and `g_closure_remove_invalidate_notifier()`) but not for the invocation process #### C Closures If you are using C or C++ to connect a callback to a given event, you will either use simple `GCClosures` which have a pretty minimal API or the even simpler `g_signal_connect()` functions (which will be presented a bit later). `g_cclosure_new()` will create a new closure which can invoke the user-provided `callback_func` with the user-provided `user_data` as its last parameter. When the closure is finalized (second stage of the destruction process), it will invoke the `destroy_data` function if the user has supplied one. `g_cclosure_new_swap()` will create a new closure which can invoke the user-provided `callback_func` with the user-provided `user_data` as its first parameter (instead of being the last parameter as with `g_cclosure_new()`). When the closure is finalized (second stage of the destruction process), it will invoke the `destroy_data` function if the user has supplied one. #### Non-C closures (for the fearless) As was explained above, closures hide the details of callback invocation. In C, callback invocation is just like function invocation: it is a matter of creating the correct stack frame for the called function and executing a call assembly instruction. C closure marshallers transform the array of GValues which represent the parameters to the target function into a C-style function parameter list, invoke the user-supplied C function with this new parameter list, get the return value of the function, transform it into a GValue and return this GValue to the marshaller caller. A generic C closure marshaller is available as `g_cclosure_marshal_generic()` which implements marshalling for all function types using libffi. Custom marshallers for different types are not needed apart from performance critical code where the libffi-based marshaller may be too slow. An example of a custom marshaller is given below, illustrating how GValues can be converted to a C function call. The marshaller is for a C function which takes an integer as its first parameter and returns `void`. ```c g_cclosure_marshal_VOID__INT (GClosure *closure, GValue *return_value, guint n_param_values, const GValue *param_values, gpointer invocation_hint, gpointer marshal_data) { typedef void (*GMarshalFunc_VOID__INT) (gpointer data1, gint arg_1, gpointer data2); register GMarshalFunc_VOID__INT callback; register GCClosure *cc = (GCClosure*) closure; register gpointer data1, data2; g_return_if_fail (n_param_values == 2); data1 = g_value_peek_pointer (param_values + 0); data2 = closure->data; callback = (GMarshalFunc_VOID__INT) (marshal_data ? marshal_data : cc->callback); callback (data1, g_marshal_value_peek_int (param_values + 1), data2); } ``` There exist other kinds of marshallers, for example there is a generic Python marshaller which is used by all Python closures (a Python closure is used to invoke a callback written in Python). This Python marshaller transforms the input GValue list representing the function parameters into a Python tuple which is the equivalent structure in Python. ### Signals GObject's signals have nothing to do with standard UNIX signals: they connect arbitrary application-specific events with any number of listeners. For example, in GTK, every user event (keystroke or mouse move) is received from the windowing system and generates a GTK event in the form of a signal emission on the widget object instance. Each signal is registered in the type system together with the type on which it can be emitted: users of the type are said to connect to the signal on a given type instance when they register a closure to be invoked upon the signal emission. Users can also emit the signal by themselves or stop the emission of the signal from within one of the closures connected to the signal. When a signal is emitted on a given type instance, all the closures connected to this signal on this type instance will be invoked. All the closures connected to such a signal represent callbacks whose signature looks like: ```c return_type function_callback (gpointer instance, ..., gpointer user_data); ``` #### Signal registration To register a new signal on an existing type, we can use any of `g_signal_newv()`, `g_signal_new_valist()` or `g_signal_new()` functions: ```c guint g_signal_newv (const gchar *signal_name, GType itype, GSignalFlags signal_flags, GClosure *class_closure, GSignalAccumulator accumulator, gpointer accu_data, GSignalCMarshaller c_marshaller, GType return_type, guint n_params, GType *param_types); ``` The number of parameters to these functions is a bit intimidating but they are relatively simple: - `signal_name`: is a string which can be used to uniquely identify a given signal - `itype`: is the instance type on which this signal can be emitted - `signal_flags`: partly defines the order in which closures which were connected to the signal are invoked - `class_closure`: this is the default closure for the signal: if it is not `NULL` upon the signal emission, it will be invoked upon this emission of the signal. The moment where this closure is invoked compared to other closures connected to that signal depends partly on the `signal_flags` - `accumulator`: this is a function pointer which is invoked after each closure has been invoked. If it returns `FALSE`, signal emission is stopped. If it returns `TRUE`, signal emission proceeds normally. It is also used to compute the return value of the signal based on the return value of all the invoked closures. For example, an accumulator could ignore `NULL` returns from closures; or it could build a list of the values returned by the closures - `accu_data`: this pointer will be passed down to each invocation of the accumulator during emission - `c_marshaller`: this is the default C marshaller for any closure which is connected to this signal - `return_type`: this is the type of the return value of the signal - `n_params`: this is the number of parameters this signal takes - `param_types`: this is an array of GTypes which indicate the type of each parameter of the signal. The length of this array is indicated by `n_params`. As you can see from the above definition, a signal is basically a description of the closures which can be connected to this signal and a description of the order in which the closures connected to this signal will be invoked. #### Signal connection If you want to connect to a signal with a closure, you have three possibilities: - you can register a class closure at signal registration: this is a system-wide operation. i.e.: the class closure will be invoked during each emission of a given signal on any of the instances of the type which supports that signal - tou can use `g_signal_override_class_closure()` which overrides the class closure of a given type. It is possible to call this function only on a derived type of the type on which the signal was registered. This function is of use only to language bindings - you can register a closure with the `g_signal_connect()` family of functions. This is an instance-specific operation: the closure will be invoked only during emission of a given signal on a given instance It is also possible to connect a different kind of callback on a given signal: emission hooks are invoked whenever a given signal is emitted whatever the instance on which it is emitted. Emission hooks are connected with `g_signal_add_emission_hook()` and removed with `g_signal_remove_emission_hook()`. #### Signal emission Signal emission is done through the use of the `g_signal_emit()` family of functions. ```c void g_signal_emitv (const GValue instance_and_params[], guint signal_id, GQuark detail, GValue *return_value); ``` - the `instance_and_params` array of GValues contains the list of input parameters to the signal. The first element of the array is the instance pointer on which to invoke the signal. The following elements of the array contain the list of parameters to the signal - `signal_id` identifies the signal to invoke - `detail` identifies the specific detail of the signal to invoke. A detail is a kind of magic token/argument which is passed around during signal emission and which is used by closures connected to the signal to filter out unwanted signal emissions. In most cases, you can safely set this value to zero. See the section called "The detail argument" for more information about this parameter - `return_value` holds the return value of the last closure invoked during emission if no accumulator was specified. If an accumulator was specified during signal creation, this accumulator is used to calculate the return value as a function of the return values of all the closures invoked during emission. If no closure is invoked during emission, the `return_value` is nonetheless initialized to zero/`NULL` Signal emission can be decomposed in 6 steps: 1. `RUN_FIRST`: if the `G_SIGNAL_RUN_FIRST` flag was used during signal registration and if there exists a class closure for this signal, the class closure is invoked. 2. `EMISSION_HOOK`: if any emission hook was added to the signal, they are invoked from first to last added. Accumulate return values. 3. `HANDLER_RUN_FIRST`: if any closure were connected with the `g_signal_connect()` family of functions, and if they are not blocked (with the `g_signal_handler_block()` family of functions) they are run here, from first to last connected. 4. `RUN_LAST`: if the `G_SIGNAL_RUN_LAST` flag was set during registration and if a class closure was set, it is invoked here. 5. `HANDLER_RUN_LAST`: if any closure were connected with the `g_signal_connect_after()` family of functions, if they were not invoked during `HANDLER_RUN_FIRST` and if they are not blocked, they are run here, from first to last connected. 6. `RUN_CLEANUP`: if the `G_SIGNAL_RUN_CLEANUP` flag was set during registration and if a class closure was set, it is invoked here. Signal emission is completed here. If, at any point during the emission (except in the `RUN_CLEANUP` or `EMISSION_HOOK` states), one of the closures stops the signal emission with `g_signal_stop_emission()`, the emission jumps to the `RUN_CLEANUP` state. If, at any point during emission, one of the closures or emission hook emits the same signal on the same instance, emission is restarted from the `RUN_FIRST` state. The accumulator function is invoked in all states, after invocation of each closure (except in `RUN_EMISSION_HOOK` and `RUN_CLEANUP`). It accumulates the closure return value into the signal return value and returns `TRUE` or `FALSE`. If, at any point, it does not return `TRUE`, emission jumps to `RUN_CLEANUP` state. If no accumulator function was provided, the value returned by the last handler run will be returned by `g_signal_emit()`. #### The detail argument All the functions related to signal emission or signal connection have a parameter named the detail. Sometimes, this parameter is hidden by the API but it is always there, in one form or another. Of the three main connection functions, only one has an explicit detail parameter as a GQuark: `g_signal_connect_closure_by_id()`. The two other functions, `g_signal_connect_closure()` and `g_signal_connect_data()` hide the detail parameter in the signal name identification. Their `detailed_signal` parameter is a string which identifies the name of the signal to connect to. The format of this string should match `signal_name::detail_name`. For example, connecting to the signal named `notify::cursor_position` will actually connect to the signal named `notify` with the `cursor_position` detail. Internally, the detail string is transformed to a GQuark if it is present. Of the four main signal emission functions, one hides it in its signal name parameter: `g_signal_emit_by_name()`. The other three have an explicit detail parameter as a GQuark again: `g_signal_emit()`, `g_signal_emitv()` and `g_signal_emit_valist()`. If a detail is provided by the user to the emission function, it is used during emission to match against the closures which also provide a detail. If a closure's detail does not match the detail provided by the user, it will not be invoked (even though it is connected to a signal which is being emitted). This completely optional filtering mechanism is mainly used as an optimization for signals which are often emitted for many different reasons: the clients can filter out which events they are interested in before the closure's marshalling code runs. For example, this is used extensively by the notify signal of GObject: whenever a property is modified on a GObject, instead of just emitting the notify signal, GObject associates as a detail to this signal emission the name of the property modified. This allows clients who wish to be notified of changes to only one property to filter most events before receiving them. As a simple rule, users can and should set the detail parameter to zero: this will disable completely this optional filtering for that signal.