ruby/doc/extension.rdoc

# extension.rdoc -  -*- RDoc -*- created at: Mon Aug  7 16:45:54 JST 1995

= Creating extension libraries for Ruby

This document explains how to make extension libraries for Ruby.

== Basic Knowledge

In C, variables have types and data do not have types.  In contrast,
Ruby variables do not have a static type, and data themselves have
types, so data will need to be converted between the languages.

Objects in Ruby are represented by the C type `VALUE'.  Each VALUE
data has its data type.

To retrieve C data from a VALUE, you need to:

1. Identify the VALUE's data type
2. Convert the VALUE into C data

Converting to the wrong data type may cause serious problems.

=== Ruby data types

The Ruby interpreter has the following data types:

T_NIL       :: nil
T_OBJECT    :: ordinary object
T_CLASS     :: class
T_MODULE    :: module
T_FLOAT     :: floating point number
T_STRING    :: string
T_REGEXP    :: regular expression
T_ARRAY     :: array
T_HASH      :: associative array
T_STRUCT    :: (Ruby) structure
T_BIGNUM    :: multi precision integer
T_FIXNUM    :: Fixnum(31bit or 63bit integer)
T_COMPLEX   :: complex number
T_RATIONAL  :: rational number
T_FILE      :: IO
T_TRUE      :: true
T_FALSE     :: false
T_DATA      :: data
T_SYMBOL    :: symbol

In addition, there are several other types used internally:

T_ICLASS    :: included module
T_MATCH     :: MatchData object
T_UNDEF     :: undefined
T_NODE      :: syntax tree node
T_ZOMBIE    :: object awaiting finalization

Most of the types are represented by C structures.

=== Check type of the VALUE data

The macro TYPE() defined in ruby.h shows the data type of the VALUE.
TYPE() returns the constant number T_XXXX described above.  To handle
data types, your code will look something like this:

  switch (TYPE(obj)) {
    case T_FIXNUM:
      /* process Fixnum */
      break;
    case T_STRING:
      /* process String */
      break;
    case T_ARRAY:
      /* process Array */
      break;
    default:
      /* raise exception */
      rb_raise(rb_eTypeError, "not valid value");
      break;
  }

There is the data type check function

  void Check_Type(VALUE value, int type)

which raises an exception if the VALUE does not have the type
specified.

There are also faster check macros for fixnums and nil.

  FIXNUM_P(obj)
  NIL_P(obj)

=== Convert VALUE into C data

The data for type T_NIL, T_FALSE, T_TRUE are nil, false, true
respectively.  They are singletons for the data type.
The equivalent C constants are: Qnil, Qfalse, Qtrue.
RTEST() will return true if a VALUE is neither Qfalse nor Qnil.
If you need to differentiate Qfalse from Qnil,
specifically test against Qfalse.

The T_FIXNUM data is a 31bit or 63bit length fixed integer.
This size depends on the size of long: if long is 32bit then
T_FIXNUM is 31bit, if long is 64bit then T_FIXNUM is 63bit.
T_FIXNUM can be converted to a C integer by using the
FIX2INT() macro or FIX2LONG().  Though you have to check that the
data is really FIXNUM before using them, they are faster.  FIX2LONG()
never raises exceptions, but FIX2INT() raises RangeError if the
result is bigger or smaller than the size of int.
There are also NUM2INT() and NUM2LONG() which converts any Ruby
numbers into C integers.  These macros include a type check,
so an exception will be raised if the conversion failed.  NUM2DBL()
can be used to retrieve the double float value in the same way.

You can use the macros
StringValue() and StringValuePtr() to get a char* from a VALUE.
StringValue(var) replaces var's value with the result of "var.to_str()".
StringValuePtr(var) does the same replacement and returns the char*
representation of var.  These macros will skip the replacement if var
is a String.  Notice that the macros take only the lvalue as their
argument, to change the value of var in place.

You can also use the macro named StringValueCStr(). This is just
like StringValuePtr(), but always adds a NUL character at the end of
the result. If the result contains a NUL character, this macro causes
the ArgumentError exception.
StringValuePtr() doesn't guarantee the existence of a NUL at the end
of the result, and the result may contain NUL.

Other data types have corresponding C structures, e.g. struct RArray
for T_ARRAY etc. The VALUE of the type which has the corresponding
structure can be cast to retrieve the pointer to the struct.  The
casting macro will be of the form RXXXX for each data type; for
instance, RARRAY(obj).  See "ruby.h".  However, we do not recommend
to access RXXXX data directly because these data structures are complex.
Use corresponding rb_xxx() functions to access the internal struct.
For example, to access an entry of array, use rb_ary_entry(ary, offset)
and rb_ary_store(ary, offset, obj).

There are some accessing macros for structure members, for example
`RSTRING_LEN(str)' to get the size of the Ruby String object.  The
allocated region can be accessed by `RSTRING_PTR(str)'.

Notice: Do not change the value of the structure directly, unless you
are responsible for the result.  This ends up being the cause of
interesting bugs.

=== Convert C data into VALUE

To convert C data to Ruby values:

FIXNUM ::

  left shift 1 bit, and turn on its least significant bit (LSB).

Other pointer values ::

  cast to VALUE.

You can determine whether a VALUE is a pointer or not by checking its LSB.

Notice: Ruby does not allow arbitrary pointer values to be a VALUE.  They
should be pointers to the structures which Ruby knows about.  The known
structures are defined in <ruby.h>.

To convert C numbers to Ruby values, use these macros:

INT2FIX() :: for integers within 31bits.
INT2NUM() :: for arbitrary sized integers.

INT2NUM() converts an integer into a Bignum if it is out of the FIXNUM
range, but is a bit slower.

=== Manipulating Ruby object

As I already mentioned, it is not recommended to modify an object's
internal structure.  To manipulate objects, use the functions supplied
by the Ruby interpreter. Some (not all) of the useful functions are
listed below:

==== String functions

rb_str_new(const char *ptr, long len) ::

  Creates a new Ruby string.

rb_str_new2(const char *ptr) ::
rb_str_new_cstr(const char *ptr) ::

  Creates a new Ruby string from a C string.  This is equivalent to
  rb_str_new(ptr, strlen(ptr)).

rb_str_new_literal(const char *ptr) ::

  Creates a new Ruby string from a C string literal.

rb_sprintf(const char *format, ...) ::
rb_vsprintf(const char *format, va_list ap) ::

  Creates a new Ruby string with printf(3) format.

  Note: In the format string, "%"PRIsVALUE can be used for Object#to_s
  (or Object#inspect if '+' flag is set) output (and related argument
  must be a VALUE).  Since it conflicts with "%i", for integers in
  format strings, use "%d".

rb_str_append(VALUE str1, VALUE str2) ::

  Appends Ruby string str2 to Ruby string str1.

rb_str_cat(VALUE str, const char *ptr, long len) ::

  Appends len bytes of data from ptr to the Ruby string.

rb_str_cat2(VALUE str, const char* ptr) ::
rb_str_cat_cstr(VALUE str, const char* ptr) ::

  Appends C string ptr to Ruby string str.  This function is
  equivalent to rb_str_cat(str, ptr, strlen(ptr)).

rb_str_catf(VALUE str, const char* format, ...) ::
rb_str_vcatf(VALUE str, const char* format, va_list ap) ::

  Appends C string format and successive arguments to Ruby string
  str according to a printf-like format.  These functions are
  equivalent to rb_str_append(str, rb_sprintf(format, ...)) and
  rb_str_append(str, rb_vsprintf(format, ap)), respectively.

rb_enc_str_new(const char *ptr, long len, rb_encoding *enc) ::
rb_enc_str_new_cstr(const char *ptr, rb_encoding *enc) ::

  Creates a new Ruby string with the specified encoding.

rb_enc_str_new_literal(const char *ptr, rb_encoding *enc) ::

  Creates a new Ruby string from a C string literal with the specified
  encoding.

rb_usascii_str_new(const char *ptr, long len) ::
rb_usascii_str_new_cstr(const char *ptr) ::

  Creates a new Ruby string with encoding US-ASCII.

rb_usascii_str_new_literal(const char *ptr) ::

  Creates a new Ruby string from a C string literal with encoding
  US-ASCII.

rb_utf8_str_new(const char *ptr, long len) ::
rb_utf8_str_new_cstr(const char *ptr) ::

  Creates a new Ruby string with encoding UTF-8.

rb_utf8_str_new_literal(const char *ptr) ::

  Creates a new Ruby string from a C string literal with encoding
  UTF-8.

rb_str_resize(VALUE str, long len) ::

  Resizes a Ruby string to len bytes.  If str is not modifiable, this
  function raises an exception.  The length of str must be set in
  advance.  If len is less than the old length the content beyond
  len bytes is discarded, else if len is greater than the old length
  the content beyond the old length bytes will not be preserved but
  will be garbage.  Note that RSTRING_PTR(str) may change by calling
  this function.

rb_str_set_len(VALUE str, long len) ::

  Sets the length of a Ruby string.  If str is not modifiable, this
  function raises an exception.  This function preserves the content
  up to len bytes, regardless RSTRING_LEN(str).  len must not exceed
  the capacity of str.

rb_str_modify(VALUE str) ::

  Prepares a Ruby string to modify.  If str is not modifiable, this
  function raises an exception, or if the buffer of str is shared,
  this function allocates new buffer to make it unshared.  Always
  you MUST call this function before modifying the contents using
  RSTRING_PTR and/or rb_str_set_len.

==== Array functions

rb_ary_new() ::

  Creates an array with no elements.

rb_ary_new2(long len) ::
rb_ary_new_capa(long len) ::

  Creates an array with no elements, allocating internal buffer
  for len elements.

rb_ary_new3(long n, ...) ::
rb_ary_new_from_args(long n, ...) ::

  Creates an n-element array from the arguments.

rb_ary_new4(long n, VALUE *elts) ::
rb_ary_new_from_values(long n, VALUE *elts) ::

  Creates an n-element array from a C array.

rb_ary_to_ary(VALUE obj) ::

  Converts the object into an array.
  Equivalent to Object#to_ary.

There are many functions to operate an array.  They may dump core if other
types are given.

rb_ary_aref(int argc, const VALUE *argv, VALUE ary) ::

  Equivalent to Array#[].

rb_ary_entry(VALUE ary, long offset) ::

  \ary[offset]

rb_ary_store(VALUE ary, long offset, VALUE obj) ::

  \ary[offset] = obj

rb_ary_subseq(VALUE ary, long beg, long len) ::

  ary[beg, len]

rb_ary_push(VALUE ary, VALUE val) ::
rb_ary_pop(VALUE ary) ::
rb_ary_shift(VALUE ary) ::
rb_ary_unshift(VALUE ary, VALUE val) ::

  ary.push, ary.pop, ary.shift, ary.unshift

rb_ary_cat(VALUE ary, const VALUE *ptr, long len) ::

  Appends len elements of objects from ptr to the array.

== Extending Ruby with C

=== Adding new features to Ruby

You can add new features (classes, methods, etc.) to the Ruby
interpreter.  Ruby provides APIs for defining the following things:

- Classes, Modules
- Methods, singleton methods
- Constants

==== Class and Module Definition

To define a class or module, use the functions below:

  VALUE rb_define_class(const char *name, VALUE super)
  VALUE rb_define_module(const char *name)

These functions return the newly created class or module.  You may
want to save this reference into a variable to use later.

To define nested classes or modules, use the functions below:

  VALUE rb_define_class_under(VALUE outer, const char *name, VALUE super)
  VALUE rb_define_module_under(VALUE outer, const char *name)

==== Method and singleton method definition

To define methods or singleton methods, use these functions:

  void rb_define_method(VALUE klass, const char *name,
                        VALUE (*func)(ANYARGS), int argc)

  void rb_define_singleton_method(VALUE object, const char *name,
                                  VALUE (*func)(ANYARGS), int argc)

The `argc' represents the number of the arguments to the C function,
which must be less than 17.  But I doubt you'll need that many.

If `argc' is negative, it specifies the calling sequence, not number of
the arguments.

If argc is -1, the function will be called as:

  VALUE func(int argc, VALUE *argv, VALUE obj)

where argc is the actual number of arguments, argv is the C array of
the arguments, and obj is the receiver.

If argc is -2, the arguments are passed in a Ruby array. The function
will be called like:

  VALUE func(VALUE obj, VALUE args)

where obj is the receiver, and args is the Ruby array containing
actual arguments.

There are some more functions to define methods. One takes an ID
as the name of method to be defined. See also ID or Symbol below.

  void rb_define_method_id(VALUE klass, ID name,
                           VALUE (*func)(ANYARGS), int argc)

There are two functions to define private/protected methods:

  void rb_define_private_method(VALUE klass, const char *name,
                                VALUE (*func)(ANYARGS), int argc)
  void rb_define_protected_method(VALUE klass, const char *name,
                                  VALUE (*func)(ANYARGS), int argc)

At last, rb_define_module_function defines a module function,
which are private AND singleton methods of the module.
For example, sqrt is a module function defined in the Math module.
It can be called in the following way:

  Math.sqrt(4)

or

  include Math
  sqrt(4)

To define module functions, use:

  void rb_define_module_function(VALUE module, const char *name,
                                 VALUE (*func)(ANYARGS), int argc)

In addition, function-like methods, which are private methods defined
in the Kernel module, can be defined using:

  void rb_define_global_function(const char *name, VALUE (*func)(ANYARGS), int argc)

To define an alias for the method,

  void rb_define_alias(VALUE module, const char* new, const char* old);

To define a reader/writer for an attribute,

  void rb_define_attr(VALUE klass, const char *name, int read, int write)

To define and undefine the `allocate' class method,

  void rb_define_alloc_func(VALUE klass, VALUE (*func)(VALUE klass));
  void rb_undef_alloc_func(VALUE klass);

func has to take the klass as the argument and return a newly
allocated instance.  This instance should be as empty as possible,
without any expensive (including external) resources.

If you are overriding an existing method of any ancestor of your class,
you may rely on:

  VALUE rb_call_super(int argc, const VALUE *argv)

To specify whether keyword arguments are passed when calling super:

  VALUE rb_call_super_kw(int argc, const VALUE *argv, int kw_splat)

+kw_splat+ can have these possible values (used by all methods that accept
+kw_splat+ argument):

RB_NO_KEYWORDS :: Do not pass keywords
RB_PASS_KEYWORDS :: Pass keywords, final argument should be a hash of keywords
RB_PASS_CALLED_KEYWORDS :: Pass keywords if current method was called with
                           keywords, useful for argument delegation

To achieve the receiver of the current scope (if no other way is
available), you can use:

  VALUE rb_current_receiver(void)

==== Constant definition

We have 2 functions to define constants:

  void rb_define_const(VALUE klass, const char *name, VALUE val)
  void rb_define_global_const(const char *name, VALUE val)

The former is to define a constant under specified class/module.  The
latter is to define a global constant.

=== Use Ruby features from C

There are several ways to invoke Ruby's features from C code.

==== Evaluate Ruby programs in a string

The easiest way to use Ruby's functionality from a C program is to
evaluate the string as Ruby program.  This function will do the job:

  VALUE rb_eval_string(const char *str)

Evaluation is done under the current context, thus current local variables
of the innermost method (which is defined by Ruby) can be accessed.

Note that the evaluation can raise an exception. There is a safer
function:

  VALUE rb_eval_string_protect(const char *str, int *state)

It returns nil when an error occurred. Moreover, *state is zero if str was
successfully evaluated, or nonzero otherwise.

==== ID or Symbol

You can invoke methods directly, without parsing the string.  First I
need to explain about ID.  ID is the integer number to represent
Ruby's identifiers such as variable names.  The Ruby data type
corresponding to ID is Symbol.  It can be accessed from Ruby in the
form:

  :Identifier

or

  :"any kind of string"

You can get the ID value from a string within C code by using

  rb_intern(const char *name)
  rb_intern_str(VALUE name)

You can retrieve ID from Ruby object (Symbol or String) given as an
argument by using

  rb_to_id(VALUE symbol)
  rb_check_id(volatile VALUE *name)
  rb_check_id_cstr(const char *name, long len, rb_encoding *enc)

These functions try to convert the argument to a String if it was not
a Symbol nor a String.  The second function stores the converted
result into *name, and returns 0 if the string is not a known symbol.
After this function returned a non-zero value, *name is always a
Symbol or a String, otherwise it is a String if the result is 0.
The third function takes NUL-terminated C string, not Ruby VALUE.

You can retrieve Symbol from Ruby object (Symbol or String) given as
an argument by using

  rb_to_symbol(VALUE name)
  rb_check_symbol(volatile VALUE *namep)
  rb_check_symbol_cstr(const char *ptr, long len, rb_encoding *enc)

These functions are similar to above functions except that these
return a Symbol instead of an ID.

You can convert C ID to Ruby Symbol by using

  VALUE ID2SYM(ID id)

and to convert Ruby Symbol object to ID, use

  ID SYM2ID(VALUE symbol)

==== Invoke Ruby method from C

To invoke methods directly, you can use the function below

  VALUE rb_funcall(VALUE recv, ID mid, int argc, ...)

This function invokes a method on the recv, with the method name
specified by the symbol mid.

==== Accessing the variables and constants

You can access class variables and instance variables using access
functions.  Also, global variables can be shared between both
environments.  There's no way to access Ruby's local variables.

The functions to access/modify instance variables are below:

  VALUE rb_ivar_get(VALUE obj, ID id)
  VALUE rb_ivar_set(VALUE obj, ID id, VALUE val)

id must be the symbol, which can be retrieved by rb_intern().

To access the constants of the class/module:

  VALUE rb_const_get(VALUE obj, ID id)

See also Constant Definition above.

== Information sharing between Ruby and C

=== Ruby constants that can be accessed from C

As stated in section 1.3,
the following Ruby constants can be referred from C.

Qtrue ::
Qfalse ::

  Boolean values.  Qfalse is false in C also (i.e. 0).

Qnil ::

  Ruby nil in C scope.

=== Global variables shared between C and Ruby

Information can be shared between the two environments using shared global
variables.  To define them, you can use functions listed below:

  void rb_define_variable(const char *name, VALUE *var)

This function defines the variable which is shared by both environments.
The value of the global variable pointed to by `var' can be accessed
through Ruby's global variable named `name'.

You can define read-only (from Ruby, of course) variables using the
function below.

  void rb_define_readonly_variable(const char *name, VALUE *var)

You can define hooked variables.  The accessor functions (getter and
setter) are called on access to the hooked variables.

  void rb_define_hooked_variable(const char *name, VALUE *var,
                                 VALUE (*getter)(), void (*setter)())

If you need to supply either setter or getter, just supply 0 for the
hook you don't need.  If both hooks are 0, rb_define_hooked_variable()
works just like rb_define_variable().

The prototypes of the getter and setter functions are as follows:

  VALUE (*getter)(ID id, VALUE *var);
  void (*setter)(VALUE val, ID id, VALUE *var);

Also you can define a Ruby global variable without a corresponding C
variable.  The value of the variable will be set/get only by hooks.

  void rb_define_virtual_variable(const char *name,
                                  VALUE (*getter)(), void (*setter)())

The prototypes of the getter and setter functions are as follows:

  VALUE (*getter)(ID id);
  void (*setter)(VALUE val, ID id);

=== Encapsulate C data into a Ruby object

Sometimes you need to expose your struct in the C world as a Ruby
object.
In a situation like this, making use of the TypedData_XXX macro
family, the pointer to the struct and the Ruby object can be mutually
converted.

--
The old (non-Typed) Data_XXX macro family has been deprecated.
In the future version of Ruby, it is possible old macros will not
work.
++

==== C struct to Ruby object

You can convert sval, a pointer to your struct, into a Ruby object
with the next macro.

  TypedData_Wrap_Struct(klass, data_type, sval)

TypedData_Wrap_Struct() returns a created Ruby object as a VALUE.

The klass argument is the class for the object.  The klass should
derive from rb_cObject, and the allocator must be set by calling
rb_define_alloc_func or rb_undef_alloc_func.

data_type is a pointer to a const rb_data_type_t which describes
how Ruby should manage the struct.

rb_data_type_t is defined like this.  Let's take a look at each
member of the struct.

  typedef struct rb_data_type_struct rb_data_type_t;

  struct rb_data_type_struct {
      const char *wrap_struct_name;
      struct {
          void (*dmark)(void*);
          void (*dfree)(void*);
          size_t (*dsize)(const void *);
          void (*dcompact)(void*);
          void *reserved[1];
      } function;
      const rb_data_type_t *parent;
      void *data;
      VALUE flags;
  };

wrap_struct_name is an identifier of this instance of the struct.
It is basically used for collecting and emitting statistics.
So the identifier must be unique in the process, but doesn't need
to be valid as a C or Ruby identifier.

These dmark / dfree functions are invoked during GC execution.  No
object allocations are allowed during it, so do not allocate ruby
objects inside them.

dmark is a function to mark Ruby objects referred from your struct.
It must mark all references from your struct with rb_gc_mark or
its family if your struct keeps such references.

--
Note that it is recommended to avoid such a reference.
++

dfree is a function to free the pointer allocation.
If this is RUBY_DEFAULT_FREE, the pointer will be just freed.

dsize calculates memory consumption in bytes by the struct.
Its parameter is a pointer to your struct.
You can pass 0 as dsize if it is hard to implement such a function.
But it is still recommended to avoid 0.

dcompact is invoked when memory compaction took place.
Referred Ruby objects that were marked by rb_gc_mark_movable()
can here be updated per rb_gc_location().

You have to fill reserved with 0.

parent can point to another C type definition that the Ruby object
is inherited from. Then TypedData_Get_Struct() does also accept
derived objects.

You can fill "data" with an arbitrary value for your use.
Ruby does nothing with the member.

flags is a bitwise-OR of the following flag values.
Since they require deep understanding of garbage collector in Ruby,
you can just set 0 to flags if you are not sure.

RUBY_TYPED_FREE_IMMEDIATELY ::

  This flag makes the garbage collector immediately invoke dfree()
  during GC when it need to free your struct.
  You can specify this flag if the dfree never unlocks Ruby's
  internal lock (GVL).

  If this flag is not set, Ruby defers invocation of dfree()
  and invokes dfree() at the same time as finalizers.

RUBY_TYPED_WB_PROTECTED ::

  It shows that implementation of the object supports write barriers.
  If this flag is set, Ruby is better able to do garbage collection
  of the object.

  When it is set, however, you are responsible for putting write
  barriers in all implementations of methods of that object as
  appropriate. Otherwise Ruby might crash while running.

  More about write barriers can be found in "Generational GC" in
  Appendix D.

RUBY_TYPED_FROZEN_SHAREABLE ::

  This flag indicates that the object is shareable object
  if the object is frozen. See Appendix F more details.

  If this flag is not set, the object can not become a shareable
  object by Ractor.make_shareable() method.

You can allocate and wrap the structure in one step.

  TypedData_Make_Struct(klass, type, data_type, sval)

This macro returns an allocated T_DATA object, wrapping the pointer to
the structure, which is also allocated.  This macro works like:

  (sval = ZALLOC(type), TypedData_Wrap_Struct(klass, data_type, sval))

Arguments klass and data_type work like their counterparts in
TypedData_Wrap_Struct().  A pointer to the allocated structure will
be assigned to sval, which should be a pointer of the type specified.

==== Declaratively marking/compacting struct references

In the case where your struct refers to Ruby objects that are simple values,
not wrapped in conditional logic or complex data structures an alternative
approach to marking and reference updating is provided, by declaring offset
references to the VALUES in your struct.

Doing this allows the Ruby GC to support marking these references and GC
compaction without the need to define the +dmark+ and +dcompact+ callbacks.

You must define a static list of VALUE pointers to the offsets within your
struct where the references are located, and set the "data" member to point to
this reference list. The reference list must end with +RUBY_END_REFS+.

Some Macros have been provided to make edge referencing easier:

* <code>RUBY_TYPED_DECL_MARKING</code> =A flag that can be set on the +ruby_data_type_t+ to indicate that references are being declared as edges.

* <code>RUBY_REFERENCES(ref_list_name)</code> - Define _ref_list_name_ as a list of references

* <code>RUBY_REF_END</code> - The end mark of the references list.

* <code>RUBY_REF_EDGE(struct, member)</code> - Declare _member_ as a VALUE edge from _struct_. Use this after +RUBY_REFERENCES_START+

* +RUBY_REFS_LIST_PTR+ - Coerce the reference list into a format that can be
  accepted by the existing +dmark+ interface.

The example below is from Dir (defined in +dir.c+)

  // The struct being wrapped. Notice this contains 3 members of which the second
  // is a VALUE reference to another ruby object.
  struct dir_data {
      DIR *dir;
      const VALUE path;
      rb_encoding *enc;
  }

  // Define a reference list `dir_refs` containing a single entry to `path`.
  // Needs terminating with RUBY_REF_END
  RUBY_REFERENCES(dir_refs) = {
      RUBY_REF_EDGE(dir_data, path),
      RUBY_REF_END
  };

  // Override the "dmark" field with the defined reference list now that we
  // no longer need a marking callback and add RUBY_TYPED_DECL_MARKING to the
  // flags field
  static const rb_data_type_t dir_data_type = {
      "dir",
      {RUBY_REFS_LIST_PTR(dir_refs), dir_free, dir_memsize,},
      0, NULL, RUBY_TYPED_WB_PROTECTED | RUBY_TYPED_FREE_IMMEDIATELY | RUBY_TYPED_DECL_MARKING
  };

Declaring simple references declaratively in this manner allows the GC to both
mark, and move the underlying object, and automatically update the reference to
it during compaction.

==== Ruby object to C struct

To retrieve the C pointer from the T_DATA object, use the macro
TypedData_Get_Struct().

  TypedData_Get_Struct(obj, type, &data_type, sval)

A pointer to the structure will be assigned to the variable sval.

See the example below for details.

== Example - Creating the dbm Extension

OK, here's the example of making an extension library.  This is the
extension to access DBMs.  The full source is included in the ext/
directory in the Ruby's source tree.

=== Make the directory

  % mkdir ext/dbm

Make a directory for the extension library under ext directory.

=== Design the Library

You need to design the library features, before making it.

=== Write the C Code

You need to write C code for your extension library.  If your library
has only one source file, choosing ``LIBRARY.c'' as a file name is
preferred.  On the other hand, in case your library has multiple source
files, avoid choosing ``LIBRARY.c'' for a file name.  It may conflict
with an intermediate file ``LIBRARY.o'' on some platforms.
Note that some functions in mkmf library described below generate
a file ``conftest.c'' for checking with compilation.  You shouldn't
choose ``conftest.c'' as a name of a source file.

Ruby will execute the initializing function named ``Init_LIBRARY'' in
the library.  For example, ``Init_dbm()'' will be executed when loading
the library.

Here's the example of an initializing function.

  #include <ruby.h>
  void
  Init_dbm(void)
  {
      /* define DBM class */
      VALUE cDBM = rb_define_class("DBM", rb_cObject);
      /* Redefine DBM.allocate
      rb_define_alloc_func(cDBM, fdbm_alloc);
      /* DBM includes Enumerable module */
      rb_include_module(cDBM, rb_mEnumerable);

      /* DBM has class method open(): arguments are received as C array */
      rb_define_singleton_method(cDBM, "open", fdbm_s_open, -1);

      /* DBM instance method close(): no args */
      rb_define_method(cDBM, "close", fdbm_close, 0);
      /* DBM instance method []: 1 argument */
      rb_define_method(cDBM, "[]", fdbm_aref, 1);

      /* ... */

      /* ID for a instance variable to store DBM data */
      id_dbm = rb_intern("dbm");
  }

The dbm extension wraps the dbm struct in the C environment using
TypedData_Make_Struct.

  struct dbmdata {
      int  di_size;
      DBM *di_dbm;
  };

  static const rb_data_type_t dbm_type = {
      "dbm",
      {0, free_dbm, memsize_dbm,},
      0, 0,
      RUBY_TYPED_FREE_IMMEDIATELY,
  };

  static VALUE
  fdbm_alloc(VALUE klass)
  {
      struct dbmdata *dbmp;
      /* Allocate T_DATA object and C struct and fill struct with zero bytes */
      return TypedData_Make_Struct(klass, struct dbmdata, &dbm_type, dbmp);
  }

This code wraps the dbmdata structure into a Ruby object.  We avoid
wrapping DBM* directly, because we want to cache size information.
Since Object.allocate allocates an ordinary T_OBJECT type (instead
of T_DATA), it's important to either use rb_define_alloc_func() to
overwrite it or rb_undef_alloc_func() to delete it.

To retrieve the dbmdata structure from a Ruby object, we define the
following macro:

  #define GetDBM(obj, dbmp) do {\
      TypedData_Get_Struct((obj), struct dbmdata, &dbm_type, (dbmp));\
      if ((dbmp) == 0) closed_dbm();\
      if ((dbmp)->di_dbm == 0) closed_dbm();\
  } while (0)

This sort of complicated macro does the retrieving and close checking
for the DBM.

There are three kinds of way to receive method arguments.  First,
methods with a fixed number of arguments receive arguments like this:

  static VALUE
  fdbm_aref(VALUE obj, VALUE keystr)
  {
      struct dbmdata *dbmp;
      GetDBM(obj, dbmp);
      /* Use dbmp to access the key */
      dbm_fetch(dbmp->di_dbm, StringValueCStr(keystr));
      /* ... */
  }

The first argument of the C function is the self, the rest are the
arguments to the method.

Second, methods with an arbitrary number of arguments receive
arguments like this:

  static VALUE
  fdbm_s_open(int argc, VALUE *argv, VALUE klass)
  {
      /* ... */
      if (rb_scan_args(argc, argv, "11", &file, &vmode) == 1) {
          mode = 0666;          /* default value */
      }
      /* ... */
  }

The first argument is the number of method arguments, the second
argument is the C array of the method arguments, and the third
argument is the receiver of the method.

You can use the function rb_scan_args() to check and retrieve the
arguments.  The third argument is a string that specifies how to
capture method arguments and assign them to the following VALUE
references.

You can just check the argument number with rb_check_arity(), this is
handy in the case you want to treat the arguments as a list.

The following is an example of a method that takes arguments by Ruby's
array:

  static VALUE
  thread_initialize(VALUE thread, VALUE args)
  {
      /* ... */
  }

The first argument is the receiver, the second one is the Ruby array
which contains the arguments to the method.

<b>Notice</b>: GC should know about global variables which refer to Ruby's objects,
but are not exported to the Ruby world.  You need to protect them by

  void rb_global_variable(VALUE *var)

or the objects themselves by

  void rb_gc_register_mark_object(VALUE object)

=== Prepare extconf.rb

If the file named extconf.rb exists, it will be executed to generate
Makefile.

extconf.rb is the file for checking compilation conditions etc.  You
need to put

  require 'mkmf'

at the top of the file.  You can use the functions below to check
various conditions.

  append_cppflags(array-of-flags[, opt]): append each flag to $CPPFLAGS if usable
  append_cflags(array-of-flags[, opt]): append each flag to $CFLAGS if usable
  append_ldflags(array-of-flags[, opt]): append each flag to $LDFLAGS if usable
  have_macro(macro[, headers[, opt]]): check whether macro is defined
  have_library(lib[, func[, headers[, opt]]]): check whether library containing function exists
  find_library(lib[, func, *paths]): find library from paths
  have_func(func[, headers[, opt]): check whether function exists
  have_var(var[, headers[, opt]]): check whether variable exists
  have_header(header[, preheaders[, opt]]): check whether header file exists
  find_header(header, *paths): find header from paths
  have_framework(fw): check whether framework exists (for MacOS X)
  have_struct_member(type, member[, headers[, opt]]): check whether struct has member
  have_type(type[, headers[, opt]]): check whether type exists
  find_type(type, opt, *headers): check whether type exists in headers
  have_const(const[, headers[, opt]]): check whether constant is defined
  check_sizeof(type[, headers[, opts]]): check size of type
  check_signedness(type[, headers[, opts]]): check signedness of type
  convertible_int(type[, headers[, opts]]): find convertible integer type
  find_executable(bin[, path]): find executable file path
  create_header(header): generate configured header
  create_makefile(target[, target_prefix]): generate Makefile

See MakeMakefile for full documentation of these functions.

The value of the variables below will affect the Makefile.

  $CFLAGS: included in CFLAGS make variable (such as -O)
  $CPPFLAGS: included in CPPFLAGS make variable (such as -I, -D)
  $LDFLAGS: included in LDFLAGS make variable (such as -L)
  $objs: list of object file names

Compiler/linker flags are not portable usually, you should use
+append_cppflags+, +append_cpflags+ and +append_ldflags+ respectively
instead of appending the above variables directly.

Normally, the object files list is automatically generated by searching
source files, but you must define them explicitly if any sources will
be generated while building.

If a compilation condition is not fulfilled, you should not call
``create_makefile''.  The Makefile will not be generated, compilation will
not be done.

=== Prepare depend (Optional)

If the file named depend exists, Makefile will include that file to
check dependencies.  You can make this file by invoking

  % gcc -MM *.c > depend

It's harmless.  Prepare it.

=== Generate Makefile

Try generating the Makefile by:

  ruby extconf.rb

If the library should be installed under vendor_ruby directory
instead of site_ruby directory, use --vendor option as follows.

  ruby extconf.rb --vendor

You don't need this step if you put the extension library under the ext
directory of the ruby source tree.  In that case, compilation of the
interpreter will do this step for you.

=== Run make

Type

  make

to compile your extension.  You don't need this step either if you have
put the extension library under the ext directory of the ruby source tree.

=== Debug

You may need to rb_debug the extension.  Extensions can be linked
statically by adding the directory name in the ext/Setup file so that
you can inspect the extension with the debugger.

=== Done! Now you have the extension library

You can do anything you want with your library.  The author of Ruby
will not claim any restrictions on your code depending on the Ruby API.
Feel free to use, modify, distribute or sell your program.

== Appendix A. Ruby header and source files overview

=== Ruby header files

Everything under <tt>$repo_root/include/ruby</tt> is installed with
<tt>make install</tt>.
It should be included per <tt>#include <ruby.h></tt> from C extensions.
All symbols are public API with the exception of symbols prefixed with
+rbimpl_+ or +RBIMPL_+. They are implementation details and shouldn't
be used by C extensions.

Only <tt>$repo_root/include/ruby/*.h</tt> whose corresponding macros
are defined in the <tt>$repo_root/include/ruby.h</tt> header are
allowed to be <tt>#include</tt>-d by C extensions.

Header files under <tt>$repo_root/internal/</tt> or directly under the
root <tt>$repo_root/*.h</tt> are not make-installed.
They are internal headers with only internal APIs.

=== Ruby language core

class.c    :: classes and modules
error.c    :: exception classes and exception mechanism
gc.c       :: memory management
load.c     :: library loading
object.c   :: objects
variable.c :: variables and constants

=== Ruby syntax parser

parse.y       :: grammar definition
parse.c       :: automatically generated from parse.y
defs/keywords :: reserved keywords
lex.c         :: automatically generated from keywords

=== Ruby evaluator (a.k.a. YARV)

  compile.c
  eval.c
  eval_error.c
  eval_jump.c
  eval_safe.c
  insns.def           : definition of VM instructions
  iseq.c              : implementation of VM::ISeq
  thread.c            : thread management and context switching
  thread_win32.c      : thread implementation
  thread_pthread.c    : ditto
  vm.c
  vm_dump.c
  vm_eval.c
  vm_exec.c
  vm_insnhelper.c
  vm_method.c

  defs/opt_insns_unif.def  : instruction unification
  defs/opt_operand.def     : definitions for optimization

    -> insn*.inc           : automatically generated
    -> opt*.inc            : automatically generated
    -> vm.inc              : automatically generated

=== Regular expression engine (Onigumo)

  regcomp.c
  regenc.c
  regerror.c
  regexec.c
  regparse.c
  regsyntax.c

=== Utility functions

debug.c    :: debug symbols for C debugger
dln.c      :: dynamic loading
st.c       :: general purpose hash table
strftime.c :: formatting times
util.c     :: misc utilities

=== Ruby interpreter implementation

  dmyext.c
  dmydln.c
  dmyencoding.c
  id.c
  inits.c
  main.c
  ruby.c
  version.c

  gem_prelude.rb
  prelude.rb

=== Class library

array.c      :: Array
bignum.c     :: Bignum
compar.c     :: Comparable
complex.c    :: Complex
cont.c       :: Fiber, Continuation
dir.c        :: Dir
enum.c       :: Enumerable
enumerator.c :: Enumerator
file.c       :: File
hash.c       :: Hash
io.c         :: IO
marshal.c    :: Marshal
math.c       :: Math
numeric.c    :: Numeric, Integer, Fixnum, Float
pack.c       :: Array#pack, String#unpack
proc.c       :: Binding, Proc
process.c    :: Process
random.c     :: random number
range.c      :: Range
rational.c   :: Rational
re.c         :: Regexp, MatchData
signal.c     :: Signal
sprintf.c    :: String#sprintf
string.c     :: String
struct.c     :: Struct
time.c       :: Time

defs/known_errors.def :: Errno::* exception classes
-> known_errors.inc   :: automatically generated

=== Multilingualization

encoding.c  :: Encoding
transcode.c :: Encoding::Converter
enc/*.c     :: encoding classes
enc/trans/* :: codepoint mapping tables

=== goruby interpreter implementation

  goruby.c
  golf_prelude.rb     : goruby specific libraries.
    -> golf_prelude.c : automatically generated

== Appendix B. Ruby extension API reference

=== Types

VALUE ::

  The type for the Ruby object.  Actual structures are defined in ruby.h,
  such as struct RString, etc.  To refer the values in structures, use
  casting macros like RSTRING(obj).

=== Variables and constants

Qnil ::

  nil object

Qtrue ::

  true object (default true value)

Qfalse ::

  false object

=== C pointer wrapping

Data_Wrap_Struct(VALUE klass, void (*mark)(), void (*free)(), void *sval) ::

  Wrap a C pointer into a Ruby object.  If object has references to other
  Ruby objects, they should be marked by using the mark function during
  the GC process.  Otherwise, mark should be 0.  When this object is no
  longer referred by anywhere, the pointer will be discarded by free
  function.

Data_Make_Struct(klass, type, mark, free, sval) ::

  This macro allocates memory using malloc(), assigns it to the variable
  sval, and returns the DATA encapsulating the pointer to memory region.

Data_Get_Struct(data, type, sval) ::

  This macro retrieves the pointer value from DATA, and assigns it to
  the variable sval.

=== Checking VALUE types

RB_TYPE_P(value, type) ::

  Is +value+ an internal type (T_NIL, T_FIXNUM, etc.)?

TYPE(value) ::

  Internal type (T_NIL, T_FIXNUM, etc.)

FIXNUM_P(value) ::

  Is +value+ a Fixnum?

NIL_P(value) ::

  Is +value+ nil?

RB_INTEGER_TYPE_P(value) ::

  Is +value+ an Integer?

RB_FLOAT_TYPE_P(value) ::

  Is +value+ a Float?

void Check_Type(VALUE value, int type) ::

  Ensures +value+ is of the given internal +type+ or raises a TypeError

=== VALUE type conversion

FIX2INT(value), INT2FIX(i) ::

  Fixnum <-> integer

FIX2LONG(value), LONG2FIX(l) ::

  Fixnum <-> long

NUM2INT(value), INT2NUM(i) ::

  Numeric <-> integer

NUM2UINT(value), UINT2NUM(ui) ::

  Numeric <-> unsigned integer

NUM2LONG(value), LONG2NUM(l) ::

  Numeric <-> long

NUM2ULONG(value), ULONG2NUM(ul) ::

  Numeric <-> unsigned long

NUM2LL(value), LL2NUM(ll) ::

  Numeric <-> long long

NUM2ULL(value), ULL2NUM(ull) ::

  Numeric <-> unsigned long long

NUM2OFFT(value), OFFT2NUM(off) ::

  Numeric <-> off_t

NUM2SIZET(value), SIZET2NUM(size) ::

  Numeric <-> size_t

NUM2SSIZET(value), SSIZET2NUM(ssize) ::

  Numeric <-> ssize_t

rb_integer_pack(value, words, numwords, wordsize, nails, flags), rb_integer_unpack(words, numwords, wordsize, nails, flags) ::

  Numeric <-> Arbitrary size integer buffer

NUM2DBL(value) ::

  Numeric -> double

rb_float_new(f) ::

  double -> Float

RSTRING_LEN(str) ::

  String -> length of String data in bytes

RSTRING_PTR(str) ::

  String -> pointer to String data
  Note that the result pointer may not be NUL-terminated

StringValue(value) ::

  Object with \#to_str -> String

StringValuePtr(value) ::

  Object with \#to_str -> pointer to String data

StringValueCStr(value) ::

  Object with \#to_str -> pointer to String data without NUL bytes
  It is guaranteed that the result data is NUL-terminated

rb_str_new2(s) ::

  char * -> String

=== Defining classes and modules

VALUE rb_define_class(const char *name, VALUE super) ::

  Defines a new Ruby class as a subclass of super.

VALUE rb_define_class_under(VALUE module, const char *name, VALUE super) ::

  Creates a new Ruby class as a subclass of super, under the module's
  namespace.

VALUE rb_define_module(const char *name) ::

  Defines a new Ruby module.

VALUE rb_define_module_under(VALUE module, const char *name) ::

  Defines a new Ruby module under the module's namespace.

void rb_include_module(VALUE klass, VALUE module) ::

  Includes module into class.  If class already includes it, just ignored.

void rb_extend_object(VALUE object, VALUE module) ::

  Extend the object with the module's attributes.

=== Defining global variables

void rb_define_variable(const char *name, VALUE *var) ::

  Defines a global variable which is shared between C and Ruby.  If name
  contains a character which is not allowed to be part of the symbol,
  it can't be seen from Ruby programs.

void rb_define_readonly_variable(const char *name, VALUE *var) ::

  Defines a read-only global variable.  Works just like
  rb_define_variable(), except the defined variable is read-only.

void rb_define_virtual_variable(const char *name, VALUE (*getter)(), void (*setter)()) ::

  Defines a virtual variable, whose behavior is defined by a pair of C
  functions.  The getter function is called when the variable is
  referenced.  The setter function is called when the variable is set to a
  value.  The prototype for getter/setter functions are:

    VALUE getter(ID id)
    void setter(VALUE val, ID id)

  The getter function must return the value for the access.

void rb_define_hooked_variable(const char *name, VALUE *var, VALUE (*getter)(), void (*setter)()) ::

  Defines hooked variable.  It's a virtual variable with a C variable.
  The getter is called as

    VALUE getter(ID id, VALUE *var)

  returning a new value.  The setter is called as

    void setter(VALUE val, ID id, VALUE *var)

void rb_global_variable(VALUE *var) ::

  Tells GC to protect C global variable, which holds Ruby value to be marked.

void rb_gc_register_mark_object(VALUE object) ::

  Tells GC to protect the +object+, which may not be referenced anywhere.

=== Constant definition

void rb_define_const(VALUE klass, const char *name, VALUE val) ::

  Defines a new constant under the class/module.

void rb_define_global_const(const char *name, VALUE val) ::

  Defines a global constant.  This is just the same as

    rb_define_const(rb_cObject, name, val)

=== Method definition

rb_define_method(VALUE klass, const char *name, VALUE (*func)(ANYARGS), int argc) ::

  Defines a method for the class.  func is the function pointer.  argc
  is the number of arguments.  if argc is -1, the function will receive
  3 arguments: argc, argv, and self.  if argc is -2, the function will
  receive 2 arguments, self and args, where args is a Ruby array of
  the method arguments.

rb_define_private_method(VALUE klass, const char *name, VALUE (*func)(ANYARGS), int argc) ::

  Defines a private method for the class.  Arguments are same as
  rb_define_method().

rb_define_singleton_method(VALUE klass, const char *name, VALUE (*func)(ANYARGS), int argc) ::

  Defines a singleton method.  Arguments are same as rb_define_method().

rb_check_arity(int argc, int min, int max) ::

  Check the number of arguments, argc is in the range of min..max.  If
  max is UNLIMITED_ARGUMENTS, upper bound is not checked.  If argc is
  out of bounds, an ArgumentError will be raised.

rb_scan_args(int argc, VALUE *argv, const char *fmt, ...) ::

  Retrieve argument from argc and argv to given VALUE references
  according to the format string.  The format can be described in ABNF
  as follows:

    scan-arg-spec  := param-arg-spec [keyword-arg-spec] [block-arg-spec]

    param-arg-spec := pre-arg-spec [post-arg-spec] / post-arg-spec /
                      pre-opt-post-arg-spec
    pre-arg-spec   := num-of-leading-mandatory-args [num-of-optional-args]
    post-arg-spec  := sym-for-variable-length-args
                      [num-of-trailing-mandatory-args]
    pre-opt-post-arg-spec := num-of-leading-mandatory-args num-of-optional-args
                             num-of-trailing-mandatory-args
    keyword-arg-spec := sym-for-keyword-arg
    block-arg-spec := sym-for-block-arg

    num-of-leading-mandatory-args  := DIGIT ; The number of leading
                                            ; mandatory arguments
    num-of-optional-args           := DIGIT ; The number of optional
                                            ; arguments
    sym-for-variable-length-args   := "*"   ; Indicates that variable
                                            ; length arguments are
                                            ; captured as a ruby array
    num-of-trailing-mandatory-args := DIGIT ; The number of trailing
                                            ; mandatory arguments
    sym-for-keyword-arg            := ":"   ; Indicates that keyword
                                            ; argument captured as a hash.
                                            ; If keyword arguments are not
                                            ; provided, returns nil.
    sym-for-block-arg              := "&"   ; Indicates that an iterator
                                            ; block should be captured if
                                            ; given

  For example, "12" means that the method requires at least one
  argument, and at most receives three (1+2) arguments.  So, the format
  string must be followed by three variable references, which are to be
  assigned to captured arguments.  For omitted arguments, variables are
  set to Qnil.  NULL can be put in place of a variable reference, which
  means the corresponding captured argument(s) should be just dropped.

  The number of given arguments, excluding an option hash or iterator
  block, is returned.

rb_scan_args_kw(int kw_splat, int argc, VALUE *argv, const char *fmt, ...) ::

  The same as +rb_scan_args+, except the +kw_splat+ argument specifies whether
  keyword arguments are provided (instead of being determined by the call
  from Ruby to the C function). +kw_splat+ should be one of the following
  values:

  RB_SCAN_ARGS_PASS_CALLED_KEYWORDS :: Same behavior as +rb_scan_args+.
  RB_SCAN_ARGS_KEYWORDS :: The final argument should be a hash treated as
                           keywords.
  RB_SCAN_ARGS_LAST_HASH_KEYWORDS :: Treat a final argument as keywords if it
                                     is a hash, and not as keywords otherwise.

int rb_get_kwargs(VALUE keyword_hash, const ID *table, int required, int optional, VALUE *values) ::

  Retrieves argument VALUEs bound to keywords, which directed by +table+
  into +values+, deleting retrieved entries from +keyword_hash+ along
  the way.  First +required+ number of IDs referred by +table+ are
  mandatory, and succeeding +optional+ (- +optional+ - 1 if
  +optional+ is negative) number of IDs are optional.  If a
  mandatory key is not contained in +keyword_hash+, raises "missing
  keyword" +ArgumentError+.  If an optional key is not present in
  +keyword_hash+, the corresponding element in +values+ is set to +Qundef+.
  If +optional+ is negative, rest of +keyword_hash+ are ignored, otherwise
  raises "unknown keyword" +ArgumentError+.

  Be warned, handling keyword arguments in the C API is less efficient
  than handling them in Ruby.  Consider using a Ruby wrapper method
  around a non-keyword C function.
  ref: https://bugs.ruby-lang.org/issues/11339

VALUE rb_extract_keywords(VALUE *original_hash) ::

  Extracts pairs whose key is a symbol into a new hash from a hash
  object referred by +original_hash+.  If the original hash contains
  non-symbol keys, then they are copied to another hash and the new hash
  is stored through +original_hash+, else 0 is stored.

=== Invoking Ruby method

VALUE rb_funcall(VALUE recv, ID mid, int narg, ...) ::

  Invokes a method.  To retrieve mid from a method name, use rb_intern().
  Able to call even private/protected methods.

VALUE rb_funcall2(VALUE recv, ID mid, int argc, VALUE *argv) ::
VALUE rb_funcallv(VALUE recv, ID mid, int argc, VALUE *argv) ::

  Invokes a method, passing arguments as an array of values.
  Able to call even private/protected methods.

VALUE rb_funcallv_kw(VALUE recv, ID mid, int argc, VALUE *argv, int kw_splat) ::

  Same as rb_funcallv, using +kw_splat+ to determine whether keyword
  arguments are passed.

VALUE rb_funcallv_public(VALUE recv, ID mid, int argc, VALUE *argv) ::

  Invokes a method, passing arguments as an array of values.
  Able to call only public methods.

VALUE rb_funcallv_public_kw(VALUE recv, ID mid, int argc, VALUE *argv, int kw_splat) ::

  Same as rb_funcallv_public, using +kw_splat+ to determine whether keyword
  arguments are passed.

VALUE rb_funcall_passing_block(VALUE recv, ID mid, int argc, const VALUE* argv) ::

  Same as rb_funcallv_public, except is passes the currently active block as
  the block when calling the method.

VALUE rb_funcall_passing_block_kw(VALUE recv, ID mid, int argc, const VALUE* argv, int kw_splat) ::

  Same as rb_funcall_passing_block, using +kw_splat+ to determine whether
  keyword arguments are passed.

VALUE rb_funcall_with_block(VALUE recv, ID mid, int argc, const VALUE *argv, VALUE passed_procval) ::

  Same as rb_funcallv_public, except +passed_procval+ specifies the block to
  pass to the method.

VALUE rb_funcall_with_block_kw(VALUE recv, ID mid, int argc, const VALUE *argv, VALUE passed_procval, int kw_splat) ::

  Same as rb_funcall_with_block, using +kw_splat+ to determine whether
  keyword arguments are passed.

VALUE rb_eval_string(const char *str) ::

  Compiles and executes the string as a Ruby program.

ID rb_intern(const char *name) ::

  Returns ID corresponding to the name.

char *rb_id2name(ID id) ::

  Returns the name corresponding ID.

char *rb_class2name(VALUE klass) ::

  Returns the name of the class.

int rb_respond_to(VALUE obj, ID id) ::

  Returns true if the object responds to the message specified by id.

=== Instance variables

VALUE rb_iv_get(VALUE obj, const char *name) ::

  Retrieve the value of the instance variable.  If the name is not
  prefixed by `@', that variable shall be inaccessible from Ruby.

VALUE rb_iv_set(VALUE obj, const char *name, VALUE val) ::

  Sets the value of the instance variable.

=== Control structure

VALUE rb_block_call(VALUE recv, ID mid, int argc, VALUE * argv, VALUE (*func) (ANYARGS), VALUE data2) ::

  Calls a method on the recv, with the method name specified by the
  symbol mid, with argc arguments in argv, supplying func as the
  block. When func is called as the block, it will receive the value
  from yield as the first argument, and data2 as the second argument.
  When yielded with multiple values (in C, rb_yield_values(),
  rb_yield_values2() and rb_yield_splat()), data2 is packed as an Array,
  whereas yielded values can be gotten via argc/argv of the third/fourth
  arguments.

VALUE rb_block_call_kw(VALUE recv, ID mid, int argc, VALUE * argv, VALUE (*func) (ANYARGS), VALUE data2, int kw_splat) ::

  Same as rb_funcall_with_block, using +kw_splat+ to determine whether
  keyword arguments are passed.

\[OBSOLETE] VALUE rb_iterate(VALUE (*func1)(), VALUE arg1, VALUE (*func2)(), VALUE arg2) ::

  Calls the function func1, supplying func2 as the block.  func1 will be
  called with the argument arg1.  func2 receives the value from yield as
  the first argument, arg2 as the second argument.

  When rb_iterate is used in 1.9, func1 has to call some Ruby-level method.
  This function is obsolete since 1.9; use rb_block_call instead.

VALUE rb_yield(VALUE val) ::

  Yields val as a single argument to the block.

VALUE rb_yield_values(int n, ...) ::

  Yields +n+ number of arguments to the block, using one C argument per Ruby
  argument.

VALUE rb_yield_values2(int n, VALUE *argv) ::

  Yields +n+ number of arguments to the block, with all Ruby arguments in the
  C argv array.

VALUE rb_yield_values_kw(int n, VALUE *argv, int kw_splat) ::

  Same as rb_yield_values2, using +kw_splat+ to determine whether
  keyword arguments are passed.

VALUE rb_yield_splat(VALUE args) ::

  Same as rb_yield_values2, except arguments are specified by the Ruby
  array +args+.

VALUE rb_yield_splat_kw(VALUE args, int kw_splat) ::

  Same as rb_yield_splat, using +kw_splat+ to determine whether
  keyword arguments are passed.

VALUE rb_rescue(VALUE (*func1)(ANYARGS), VALUE arg1, VALUE (*func2)(ANYARGS), VALUE arg2) ::

  Calls the function func1, with arg1 as the argument.  If an exception
  occurs during func1, it calls func2 with arg2 as the first argument
  and the exception object as the second argument.  The return value
  of rb_rescue() is the return value from func1 if no exception occurs,
  from func2 otherwise.

VALUE rb_ensure(VALUE (*func1)(ANYARGS), VALUE arg1, VALUE (*func2)(ANYARGS), VALUE arg2) ::

  Calls the function func1 with arg1 as the argument, then calls func2
  with arg2 if execution terminated.  The return value from
  rb_ensure() is that of func1 when no exception occurred.

VALUE rb_protect(VALUE (*func) (VALUE), VALUE arg, int *state) ::

  Calls the function func with arg as the argument.  If no exception
  occurred during func, it returns the result of func and *state is zero.
  Otherwise, it returns Qnil and sets *state to nonzero.  If state is
  NULL, it is not set in both cases.
  You have to clear the error info with rb_set_errinfo(Qnil) when
  ignoring the caught exception.

void rb_jump_tag(int state) ::

  Continues the exception caught by rb_protect() and rb_eval_string_protect().
  state must be the returned value from those functions.  This function
  never return to the caller.

void rb_iter_break() ::

  Exits from the current innermost block.  This function never return to
  the caller.

void rb_iter_break_value(VALUE value) ::

  Exits from the current innermost block with the value.  The block will
  return the given argument value.  This function never return to the
  caller.

=== Exceptions and errors

void rb_warn(const char *fmt, ...) ::

  Prints a warning message according to a printf-like format.

void rb_warning(const char *fmt, ...) ::

  Prints a warning message according to a printf-like format, if
  $VERBOSE is true.

void rb_raise(rb_eRuntimeError, const char *fmt, ...) ::

  Raises RuntimeError.  The fmt is a format string just like printf().

void rb_raise(VALUE exception, const char *fmt, ...) ::

  Raises a class exception.  The fmt is a format string just like printf().

void rb_fatal(const char *fmt, ...) ::

  Raises a fatal error, terminates the interpreter.  No exception handling
  will be done for fatal errors, but ensure blocks will be executed.

void rb_bug(const char *fmt, ...) ::

  Terminates the interpreter immediately.  This function should be
  called under the situation caused by the bug in the interpreter.  No
  exception handling nor ensure execution will be done.

Note: In the format string, "%"PRIsVALUE can be used for Object#to_s
(or Object#inspect if '+' flag is set) output (and related argument
must be a VALUE).  Since it conflicts with "%i", for integers in
format strings, use "%d".

=== Threading

As of Ruby 1.9, Ruby supports native 1:1 threading with one kernel
thread per Ruby Thread object.  Currently, there is a GVL (Global VM Lock)
which prevents simultaneous execution of Ruby code which may be released
by the rb_thread_call_without_gvl and rb_thread_call_without_gvl2 functions.
These functions are tricky-to-use and documented in thread.c; do not
use them before reading comments in thread.c.

void rb_thread_schedule(void) ::

  Give the scheduler a hint to pass execution to another thread.

=== Input/Output (IO) on a single file descriptor

int rb_io_wait_readable(int fd) ::

  Wait indefinitely for the given FD to become readable, allowing other
  threads to be scheduled.  Returns a true value if a read may be
  performed, false if there is an unrecoverable error.

int rb_io_wait_writable(int fd) ::

  Like rb_io_wait_readable, but for writability.

int rb_wait_for_single_fd(int fd, int events, struct timeval *timeout) ::

  Allows waiting on a single FD for one or multiple events with a
  specified timeout.

  +events+ is a mask of any combination of the following values:

  * RB_WAITFD_IN - wait for readability of normal data
  * RB_WAITFD_OUT - wait for writability
  * RB_WAITFD_PRI - wait for readability of urgent data

  Use a NULL +timeout+ to wait indefinitely.

=== I/O multiplexing

Ruby supports I/O multiplexing based on the select(2) system call.
The Linux select_tut(2) manpage
<http://man7.org/linux/man-pages/man2/select_tut.2.html>
provides a good overview on how to use select(2), and the Ruby API has
analogous functions and data structures to the well-known select API.
Understanding of select(2) is required to understand this section.

typedef struct rb_fdset_t ::

  The data structure which wraps the fd_set bitmap used by select(2).
  This allows Ruby to use FD sets larger than that allowed by
  historic limitations on modern platforms.

void rb_fd_init(rb_fdset_t *) ::

  Initializes the rb_fdset_t, it must be initialized before other rb_fd_*
  operations.  Analogous to calling malloc(3) to allocate an fd_set.

void rb_fd_term(rb_fdset_t *) ::

  Destroys the rb_fdset_t, releasing any memory and resources it used.
  It must be reinitialized using rb_fd_init before future use.
  Analogous to calling free(3) to release memory for an fd_set.

void rb_fd_zero(rb_fdset_t *) ::

  Clears all FDs from the rb_fdset_t, analogous to FD_ZERO(3).

void rb_fd_set(int fd, rb_fdset_t *) ::

  Adds a given FD in the rb_fdset_t, analogous to FD_SET(3).

void rb_fd_clr(int fd, rb_fdset_t *) ::

  Removes a given FD from the rb_fdset_t, analogous to FD_CLR(3).

int rb_fd_isset(int fd, const rb_fdset_t *) ::

  Returns true if a given FD is set in the rb_fdset_t, false if not.
  Analogous to FD_ISSET(3).

int rb_thread_fd_select(int nfds, rb_fdset_t *readfds, rb_fdset_t *writefds, rb_fdset_t *exceptfds, struct timeval *timeout) ::

  Analogous to the select(2) system call, but allows other Ruby
  threads to be scheduled while waiting.

  When only waiting on a single FD, favor rb_io_wait_readable,
  rb_io_wait_writable, or rb_wait_for_single_fd functions since
  they can be optimized for specific platforms (currently, only Linux).

=== Initialize and start the interpreter

The embedding API functions are below (not needed for extension libraries):

void ruby_init() ::

  Initializes the interpreter.

void *ruby_options(int argc, char **argv) ::

  Process command line arguments for the interpreter.
  And compiles the Ruby source to execute.
  It returns an opaque pointer to the compiled source
  or an internal special value.

int ruby_run_node(void *n) ::

  Runs the given compiled source and exits this process.
  It returns EXIT_SUCCESS if successfully runs the source.
  Otherwise, it returns other value.

void ruby_script(char *name) ::

  Specifies the name of the script ($0).

=== Hooks for the interpreter events

void rb_add_event_hook(rb_event_hook_func_t func, rb_event_flag_t events, VALUE data) ::

  Adds a hook function for the specified interpreter events.
  events should be OR'ed value of:

    RUBY_EVENT_LINE
    RUBY_EVENT_CLASS
    RUBY_EVENT_END
    RUBY_EVENT_CALL
    RUBY_EVENT_RETURN
    RUBY_EVENT_C_CALL
    RUBY_EVENT_C_RETURN
    RUBY_EVENT_RAISE
    RUBY_EVENT_ALL

  The definition of rb_event_hook_func_t is below:

    typedef void (*rb_event_hook_func_t)(rb_event_t event, VALUE data,
                                         VALUE self, ID id, VALUE klass)

  The third argument `data' to rb_add_event_hook() is passed to the hook
  function as the second argument, which was the pointer to the current
  NODE in 1.8.  See RB_EVENT_HOOKS_HAVE_CALLBACK_DATA below.

int rb_remove_event_hook(rb_event_hook_func_t func) ::

  Removes the specified hook function.

=== Memory usage

void rb_gc_adjust_memory_usage(ssize_t diff) ::

  Adjusts the amount of registered external memory.  You can tell GC how
  much memory is used by an external library by this function.  Calling
  this function with positive diff means the memory usage is increased;
  new memory block is allocated or a block is reallocated as larger
  size.  Calling this function with negative diff means the memory usage
  is decreased; a memory block is freed or a block is reallocated as
  smaller size.  This function may trigger the GC.

=== Macros for compatibility

Some macros to check API compatibilities are available by default.

NORETURN_STYLE_NEW ::

  Means that NORETURN macro is functional style instead of prefix.

HAVE_RB_DEFINE_ALLOC_FUNC ::

  Means that function rb_define_alloc_func() is provided, that means the
  allocation framework is used.  This is the same as the result of
  have_func("rb_define_alloc_func", "ruby.h").

HAVE_RB_REG_NEW_STR ::

  Means that function rb_reg_new_str() is provided, that creates Regexp
  object from String object.  This is the same as the result of
  have_func("rb_reg_new_str", "ruby.h").

HAVE_RB_IO_T ::

  Means that type rb_io_t is provided.

USE_SYMBOL_AS_METHOD_NAME ::

  Means that Symbols will be returned as method names, e.g.,
  Module#methods, \#singleton_methods and so on.

HAVE_RUBY_*_H ::

  Defined in ruby.h and means corresponding header is available.  For
  instance, when HAVE_RUBY_ST_H is defined you should use ruby/st.h not
  mere st.h.

  Header files corresponding to these macros may be <tt>#include</tt>
  directly from extension libraries.

RB_EVENT_HOOKS_HAVE_CALLBACK_DATA ::

  Means that rb_add_event_hook() takes the third argument `data', to be
  passed to the given event hook function.

=== Defining backward compatible macros for keyword argument functions

Most ruby C extensions are designed to support multiple Ruby versions.
In order to correctly support Ruby 2.7+ in regards to keyword
argument separation, C extensions need to use <code>*_kw</code>
functions.  However, these functions do not exist in Ruby 2.6 and
below, so in those cases macros should be defined to allow you to use
the same code on multiple Ruby versions.  Here are example macros
you can use in extensions that support Ruby 2.6 (or below) when using
the <code>*_kw</code> functions introduced in Ruby 2.7.

  #ifndef RB_PASS_KEYWORDS
  /* Only define macros on Ruby <2.7 */
  #define rb_funcallv_kw(o, m, c, v, kw) rb_funcallv(o, m, c, v)
  #define rb_funcallv_public_kw(o, m, c, v, kw) rb_funcallv_public(o, m, c, v)
  #define rb_funcall_passing_block_kw(o, m, c, v, kw) rb_funcall_passing_block(o, m, c, v)
  #define rb_funcall_with_block_kw(o, m, c, v, b, kw) rb_funcall_with_block(o, m, c, v, b)
  #define rb_scan_args_kw(kw, c, v, s, ...) rb_scan_args(c, v, s, __VA_ARGS__)
  #define rb_call_super_kw(c, v, kw) rb_call_super(c, v)
  #define rb_yield_values_kw(c, v, kw) rb_yield_values2(c, v)
  #define rb_yield_splat_kw(a, kw) rb_yield_splat(a)
  #define rb_block_call_kw(o, m, c, v, f, p, kw) rb_block_call(o, m, c, v, f, p)
  #define rb_fiber_resume_kw(o, c, v, kw) rb_fiber_resume(o, c, v)
  #define rb_fiber_yield_kw(c, v, kw) rb_fiber_yield(c, v)
  #define rb_enumeratorize_with_size_kw(o, m, c, v, f, kw) rb_enumeratorize_with_size(o, m, c, v, f)
  #define SIZED_ENUMERATOR_KW(obj, argc, argv, size_fn, kw_splat) \
      rb_enumeratorize_with_size((obj), ID2SYM(rb_frame_this_func()), \
                                 (argc), (argv), (size_fn))
  #define RETURN_SIZED_ENUMERATOR_KW(obj, argc, argv, size_fn, kw_splat) do { \
          if (!rb_block_given_p())                                            \
              return SIZED_ENUMERATOR(obj, argc, argv, size_fn);              \
      } while (0)
  #define RETURN_ENUMERATOR_KW(obj, argc, argv, kw_splat) RETURN_SIZED_ENUMERATOR(obj, argc, argv, 0)
  #define rb_check_funcall_kw(o, m, c, v, kw) rb_check_funcall(o, m, c, v)
  #define rb_obj_call_init_kw(o, c, v, kw) rb_obj_call_init(o, c, v)
  #define rb_class_new_instance_kw(c, v, k, kw) rb_class_new_instance(c, v, k)
  #define rb_proc_call_kw(p, a, kw) rb_proc_call(p, a)
  #define rb_proc_call_with_block_kw(p, c, v, b, kw) rb_proc_call_with_block(p, c, v, b)
  #define rb_method_call_kw(c, v, m, kw) rb_method_call(c, v, m)
  #define rb_method_call_with_block_kw(c, v, m, b, kw) rb_method_call_with_block(c, v, m, b)
  #define rb_eval_cmd_kwd(c, a, kw) rb_eval_cmd(c, a, 0)
  #endif

== Appendix C. Functions available for use in extconf.rb

See documentation for {mkmf}[rdoc-ref:MakeMakefile].

== Appendix D. Generational GC

Ruby 2.1 introduced a generational garbage collector (called RGenGC).
RGenGC (mostly) keeps compatibility.

Generally, the use of the technique called write barriers is required in
extension libraries for generational GC
(https://en.wikipedia.org/wiki/Garbage_collection_%28computer_science%29).
RGenGC works fine without write barriers in extension libraries.

If your library adheres to the following tips, performance can
be further improved. Especially, the "Don't touch pointers directly" section is
important.

=== Incompatibility

You can't write RBASIC(obj)->klass field directly because it is const
value now.

Basically you should not write this field because MRI expects it to be
an immutable field, but if you want to do it in your extension you can
use the following functions:

VALUE rb_obj_hide(VALUE obj) ::

  Clear RBasic::klass field. The object will be an internal object.
  ObjectSpace::each_object can't find this object.

VALUE rb_obj_reveal(VALUE obj, VALUE klass) ::

  Reset RBasic::klass to be klass.
  We expect the `klass' is hidden class by rb_obj_hide().

=== Write barriers

RGenGC doesn't require write barriers to support generational GC.
However, caring about write barrier can improve the performance of
RGenGC. Please check the following tips.

==== Don't touch pointers directly

In MRI (include/ruby/ruby.h), some macros to acquire pointers to the
internal data structures are supported such as RARRAY_PTR(),
RSTRUCT_PTR() and so on.

DO NOT USE THESE MACROS and instead use the corresponding C-APIs such as
rb_ary_aref(), rb_ary_store() and so on.

==== Consider whether to insert write barriers

You don't need to care about write barriers if you only use built-in
types.

If you support T_DATA objects, you may consider using write barriers.

Inserting write barriers into T_DATA objects only works with the
following type objects: (a) long-lived objects, (b) when a huge number
of objects are generated and \(c) container-type objects that have
references to other objects. If your extension provides such a type of
T_DATA objects, consider inserting write barriers.

(a): short-lived objects don't become old generation objects.
(b): only a few oldgen objects don't have performance impact.
\(c): only a few references don't have performance impact.

Inserting write barriers is a very difficult hack, it is easy to
introduce critical bugs. And inserting write barriers has several areas
of overhead. Basically we don't recommend you insert write barriers.
Please carefully consider the risks.

==== Combine with built-in types

Please consider utilizing built-in types. Most built-in types support
write barrier, so you can use them to avoid manually inserting write
barriers.

For example, if your T_DATA has references to other objects, then you
can move these references to Array. A T_DATA object only has a reference
to an array object. Or you can also use a Struct object to gather a
T_DATA object (without any references) and an that Array contains
references.

With use of such techniques, you don't need to insert write barriers
anymore.

==== Insert write barriers

\[AGAIN] Inserting write barriers is a very difficult hack, and it is
easy to introduce critical bugs. And inserting write barriers has
several areas of overhead. Basically we don't recommend you insert write
barriers. Please carefully consider the risks.

Before inserting write barriers, you need to know about RGenGC algorithm
(gc.c will help you). Macros and functions to insert write barriers are
available in include/ruby/ruby.h. An example is available in iseq.c.

For a complete guide for RGenGC and write barriers, please refer to
<https://bugs.ruby-lang.org/projects/ruby-master/wiki/RGenGC>.

== Appendix E. RB_GC_GUARD to protect from premature GC

C Ruby currently uses conservative garbage collection, thus VALUE
variables must remain visible on the stack or registers to ensure any
associated data remains usable.  Optimizing C compilers are not designed
with conservative garbage collection in mind, so they may optimize away
the original VALUE even if the code depends on data associated with that
VALUE.

The following example illustrates the use of RB_GC_GUARD to ensure
the contents of sptr remain valid while the second invocation of
rb_str_new_cstr is running.

  VALUE s, w;
  const char *sptr;

  s = rb_str_new_cstr("hello world!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!");
  sptr = RSTRING_PTR(s);
  w = rb_str_new_cstr(sptr + 6); /* Possible GC invocation */

  RB_GC_GUARD(s); /* ensure s (and thus sptr) do not get GC-ed */

In the above example, RB_GC_GUARD must be placed _after_ the last use of
sptr.  Placing RB_GC_GUARD before dereferencing sptr would be of no use.
RB_GC_GUARD is only effective on the VALUE data type, not converted C
data types.

RB_GC_GUARD would not be necessary at all in the above example if
non-inlined function calls are made on the `s' VALUE after sptr is
dereferenced.  Thus, in the above example, calling any un-inlined
function on `s' such as:

  rb_str_modify(s);

Will ensure `s' stays on the stack or register to prevent a
GC invocation from prematurely freeing it.

Using the RB_GC_GUARD macro is preferable to using the "volatile"
keyword in C.  RB_GC_GUARD has the following advantages:

1. the intent of the macro use is clear

2. RB_GC_GUARD only affects its call site, "volatile" generates some
   extra code every time the variable is used, hurting optimization.

3. "volatile" implementations may be buggy/inconsistent in some
   compilers and architectures. RB_GC_GUARD is customizable for broken
   systems/compilers without negatively affecting other systems.

== Appendix F. Ractor support

Ractor(s) are the parallel execution mechanism introduced in Ruby 3.0. All
ractors can run in parallel on a different OS thread (using an underlying system
provided thread), so the C extension should be thread-safe. A C extension that
can run in multiple ractors is called "Ractor-safe".

Ractor safety around C extensions has the following properties:
1. By default, all C extensions are recognized as Ractor-unsafe.
2. Ractor-unsafe C-methods may only be called from the main Ractor.  If invoked
   by a non-main Ractor, then a Ractor::UnsafeError is raised.
3. If an extension desires to be marked as Ractor-safe the extension should
   call rb_ext_ractor_safe(true) at the Init_ function for the extension, and
   all defined methods will be marked as Ractor-safe.

To make a "Ractor-safe" C extension, we need to check the following points:

(1) Do not share unshareable objects between ractors

For example, C's global variable can lead sharing an unshareable objects
between ractors.

  VALUE g_var;
  VALUE set(VALUE self, VALUE v){ return g_var = v; }
  VALUE get(VALUE self){ return g_var; }

set() and get() pair can share an unshareable objects using g_var, and
it is Ractor-unsafe.

Not only using global variables directly, some indirect data structure
such as global st_table can share the objects, so please take care.

Note that class and module objects are shareable objects, so you can
keep the code "cFoo = rb_define_class(...)" with C's global variables.

(2) Check the thread-safety of the extension

An extension should be thread-safe. For example, the following code is
not thread-safe:

  bool g_called = false;
  VALUE call(VALUE self) {
    if (g_called) rb_raise("recursive call is not allowed.");
    g_called = true;
    VALUE ret = do_something();
    g_called = false;
    return ret;
  }

because g_called global variable should be synchronized by other
ractor's threads. To avoid such data-race, some synchronization should
be used. Check include/ruby/thread_native.h and include/ruby/atomic.h.

With Ractors, all objects given as method parameters and the receiver (self)
are guaranteed to be from the current Ractor or to be shareable. As a
consequence, it is easier to make code ractor-safe than to make code generally
thread-safe. For example, we don't need to lock an array object to access the
element of it.

(3) Check the thread-safety of any used library

If the extension relies on an external library, such as a function foo() from
a library libfoo, the function libfoo foo() should be thread safe.

(4) Make an object shareable

This is not required to make an extension Ractor-safe.

If an extension provides special objects defined by rb_data_type_t,
consider these objects can become shareable or not.

RUBY_TYPED_FROZEN_SHAREABLE flag indicates that these objects can be
shareable objects if the object is frozen. This means that if the object
is frozen, the mutation of wrapped data is not allowed.

(5) Others

There are possibly other points or requirements which must be considered in the
making of a Ractor-safe extension.  This document will be extended as they are
discovered.

:enddoc: Local variables:
:enddoc: fill-column: 70
:enddoc: end: