+ * It specifies that the structure s_array contains six fields, that the \a name field is a classical null-terminated
+ * char* string (#xbt_string_t is just an helper type defined exactly to help the parsing macro to specify the semantic of the pointer),
+ * that \a father field is a simple reference, that the size of the array pointed by \a data is the \a length field, and that the
+ * \a matrix field is an arraywhich size is the result of \a rows times \a cols.
+ *
+ * \warning Since GRAS_DEFINE_TYPE is a macro, you shouldn't put any comma in your type definition
+ * (comma separates macro args). For example, change \verbatim int a, b;\endverbatim to \verbatim int a;
+int b;\endverbatim
+ *
+ * \section gras_dd_define \#define and fixed size array
+ *
+ * If you want to exchange arrays which size is given at compilation time by a
+ * \#defined constant, you need to keep GRAS informed. It would be done the
+ * following way:
+
+\verbatim #define BLOCK_SIZE 32
+GRAS_DEFINE_TYPE(s_toto,
+struct {
+ double data[BLOCK_SIZE];
+} s_toto;)
+
+void register_messages() {
+ gras_datadesc_type_t toto_type;
+
+ gras_datadesc_set_const("BLOCK_SIZE",BLOCK_SIZE);
+ toto_type = gras_datadesc_by_symbol(s_toto);
+}\endverbatim
+ *
+ * The form <tt>gras_datadesc_set_const("BLOCK_SIZE",BLOCK_SIZE);</tt> ensures
+ * that when you change the definition of the constant, GRAS keeps informed of
+ * the right value. Passing the numerical value of the constant as second
+ * argument would be a bad idea to that regard. Of course, the call to
+ * gras_datadesc_set_const() should come before any gras_datadesc_by_symbol()
+ * containing references to it.
+ *
+ * \section GRAS_dd_multidim Defining multidimentional arrays
+ *
+ * The mecanism for multidimensional arrays is known to be fragile and cumbersome. If you want to use it,
+ * you have to understand how it is implemented: the multiplication is performed using the sizes stack. In previous example,
+ * a \ref gras_datadesc_cb_push_int callback is added to the \a rows field and a \ref gras_datadesc_cb_push_int_mult one is
+ * added to \a cols. So, when the structure is sent, the \a rows field push its value onto the stack, then the \a cols field
+ * retrieve this value from the stack, compute (and push) the multiplication value. The \a matrix field can then retrieve this
+ * value by poping the array. There is several ways for this to go wrong:
+ * - if the matrix field is placed before the sizes, the right value won't get pushed into the stack soon enough.
+ * Reorder your structure fields if needed.
+ * - if you write GRAS_ANNOTE(size,cols*rows); in previous example (inverting rows and cols in annotation),
+ * \a rows will be given a \ref gras_datadesc_cb_push_int_mult. This cannot work since it will try to
+ * pop the value which will be pushed by \a cols <i>afterward</i>.
+ * - if you have more than one matrix in your structure, don't interleave the size. They are pushed/poped in the structure order.
+ * - if some of the sizes are used in more than one matrix, you cannot use this mecanism -- sorry.
+ *
+ * If you cannot express your datadescs with this mechanism, you'll have to use the more advanced
+ * (and somehow complex) one described in the \ref GRAS_dd_cb_full.
+ *
+ * \section GRAS_dd_multifile Projects spanning over multiple files
+ *
+ * GRAS_DEFINE_TYPE declares some symbols to work, it needs some special
+ * care when used in several files. In such case, you want the regular type
+ * definition in all files, but the gras specific symbol defined in only
+ * one file. For example, consider the following gras project sketch.