1perlxs(3)             User Contributed Perl Documentation            perlxs(3)
2
3
4

NAME

6       perlxs - XS language reference manual
7

DESCRIPTION

9   Introduction
10       XS is an interface description file format used to create an extension
11       interface between Perl and C code (or a C library) which one wishes to
12       use with Perl.  The XS interface is combined with the library to create
13       a new library which can then be either dynamically loaded or statically
14       linked into perl.  The XS interface description is written in the XS
15       language and is the core component of the Perl extension interface.
16
17       Before writing XS, read the "CAVEATS" section below.
18
19       An XSUB forms the basic unit of the XS interface.  After compilation by
20       the xsubpp compiler, each XSUB amounts to a C function definition which
21       will provide the glue between Perl calling conventions and C calling
22       conventions.
23
24       The glue code pulls the arguments from the Perl stack, converts these
25       Perl values to the formats expected by a C function, calls this C
26       function, and then transfers the return values of the C function back
27       to Perl.  Return values here may be a conventional C return value or
28       any C function arguments that may serve as output parameters.  These
29       return values may be passed back to Perl either by putting them on the
30       Perl stack, or by modifying the arguments supplied from the Perl side.
31
32       The above is a somewhat simplified view of what really happens.  Since
33       Perl allows more flexible calling conventions than C, XSUBs may do much
34       more in practice, such as checking input parameters for validity,
35       throwing exceptions (or returning undef/empty list) if the return value
36       from the C function indicates failure, calling different C functions
37       based on numbers and types of the arguments, providing an object-
38       oriented interface, etc.
39
40       Of course, one could write such glue code directly in C.  However, this
41       would be a tedious task, especially if one needs to write glue for
42       multiple C functions, and/or one is not familiar enough with the Perl
43       stack discipline and other such arcana.  XS comes to the rescue here:
44       instead of writing this glue C code in long-hand, one can write a more
45       concise short-hand description of what should be done by the glue, and
46       let the XS compiler xsubpp handle the rest.
47
48       The XS language allows one to describe the mapping between how the C
49       routine is used, and how the corresponding Perl routine is used.  It
50       also allows creation of Perl routines which are directly translated to
51       C code and which are not related to a pre-existing C function.  In
52       cases when the C interface coincides with the Perl interface, the XSUB
53       declaration is almost identical to a declaration of a C function (in
54       K&R style).  In such circumstances, there is another tool called "h2xs"
55       that is able to translate an entire C header file into a corresponding
56       XS file that will provide glue to the functions/macros described in the
57       header file.
58
59       The XS compiler is called xsubpp.  This compiler creates the constructs
60       necessary to let an XSUB manipulate Perl values, and creates the glue
61       necessary to let Perl call the XSUB.  The compiler uses typemaps to
62       determine how to map C function parameters and output values to Perl
63       values and back.  The default typemap (which comes with Perl) handles
64       many common C types.  A supplementary typemap may also be needed to
65       handle any special structures and types for the library being linked.
66       For more information on typemaps, see perlxstypemap.
67
68       A file in XS format starts with a C language section which goes until
69       the first "MODULE =" directive.  Other XS directives and XSUB
70       definitions may follow this line.  The "language" used in this part of
71       the file is usually referred to as the XS language.  xsubpp recognizes
72       and skips POD (see perlpod) in both the C and XS language sections,
73       which allows the XS file to contain embedded documentation.
74
75       See perlxstut for a tutorial on the whole extension creation process.
76
77       Note: For some extensions, Dave Beazley's SWIG system may provide a
78       significantly more convenient mechanism for creating the extension glue
79       code.  See <http://www.swig.org/> for more information.
80
81       For simple bindings to C libraries as well as other machine code
82       libraries, consider instead using the much simpler libffi
83       <http://sourceware.org/libffi/> interface via CPAN modules like
84       FFI::Platypus or FFI::Raw.
85
86   On The Road
87       Many of the examples which follow will concentrate on creating an
88       interface between Perl and the ONC+ RPC bind library functions.  The
89       rpcb_gettime() function is used to demonstrate many features of the XS
90       language.  This function has two parameters; the first is an input
91       parameter and the second is an output parameter.  The function also
92       returns a status value.
93
94               bool_t rpcb_gettime(const char *host, time_t *timep);
95
96       From C this function will be called with the following statements.
97
98            #include <rpc/rpc.h>
99            bool_t status;
100            time_t timep;
101            status = rpcb_gettime( "localhost", &timep );
102
103       If an XSUB is created to offer a direct translation between this
104       function and Perl, then this XSUB will be used from Perl with the
105       following code.  The $status and $timep variables will contain the
106       output of the function.
107
108            use RPC;
109            $status = rpcb_gettime( "localhost", $timep );
110
111       The following XS file shows an XS subroutine, or XSUB, which
112       demonstrates one possible interface to the rpcb_gettime() function.
113       This XSUB represents a direct translation between C and Perl and so
114       preserves the interface even from Perl.  This XSUB will be invoked from
115       Perl with the usage shown above.  Note that the first three #include
116       statements, for "EXTERN.h", "perl.h", and "XSUB.h", will always be
117       present at the beginning of an XS file.  This approach and others will
118       be expanded later in this document.  A #define for
119       "PERL_NO_GET_CONTEXT" should be present to fetch the interpreter
120       context more efficiently, see perlguts for details.
121
122            #define PERL_NO_GET_CONTEXT
123            #include "EXTERN.h"
124            #include "perl.h"
125            #include "XSUB.h"
126            #include <rpc/rpc.h>
127
128            MODULE = RPC  PACKAGE = RPC
129
130            bool_t
131            rpcb_gettime(host,timep)
132                 char *host
133                 time_t &timep
134               OUTPUT:
135                 timep
136
137       Any extension to Perl, including those containing XSUBs, should have a
138       Perl module to serve as the bootstrap which pulls the extension into
139       Perl.  This module will export the extension's functions and variables
140       to the Perl program and will cause the extension's XSUBs to be linked
141       into Perl.  The following module will be used for most of the examples
142       in this document and should be used from Perl with the "use" command as
143       shown earlier.  Perl modules are explained in more detail later in this
144       document.
145
146            package RPC;
147
148            require Exporter;
149            require DynaLoader;
150            @ISA = qw(Exporter DynaLoader);
151            @EXPORT = qw( rpcb_gettime );
152
153            bootstrap RPC;
154            1;
155
156       Throughout this document a variety of interfaces to the rpcb_gettime()
157       XSUB will be explored.  The XSUBs will take their parameters in
158       different orders or will take different numbers of parameters.  In each
159       case the XSUB is an abstraction between Perl and the real C
160       rpcb_gettime() function, and the XSUB must always ensure that the real
161       rpcb_gettime() function is called with the correct parameters.  This
162       abstraction will allow the programmer to create a more Perl-like
163       interface to the C function.
164
165   The Anatomy of an XSUB
166       The simplest XSUBs consist of 3 parts: a description of the return
167       value, the name of the XSUB routine and the names of its arguments, and
168       a description of types or formats of the arguments.
169
170       The following XSUB allows a Perl program to access a C library function
171       called sin().  The XSUB will imitate the C function which takes a
172       single argument and returns a single value.
173
174            double
175            sin(x)
176              double x
177
178       Optionally, one can merge the description of types and the list of
179       argument names, rewriting this as
180
181            double
182            sin(double x)
183
184       This makes this XSUB look similar to an ANSI C declaration.  An
185       optional semicolon is allowed after the argument list, as in
186
187            double
188            sin(double x);
189
190       Parameters with C pointer types can have different semantic: C
191       functions with similar declarations
192
193            bool string_looks_as_a_number(char *s);
194            bool make_char_uppercase(char *c);
195
196       are used in absolutely incompatible manner.  Parameters to these
197       functions could be described to xsubpp like this:
198
199            char *  s
200            char    &c
201
202       Both these XS declarations correspond to the "char*" C type, but they
203       have different semantics, see "The & Unary Operator".
204
205       It is convenient to think that the indirection operator "*" should be
206       considered as a part of the type and the address operator "&" should be
207       considered part of the variable.  See perlxstypemap for more info about
208       handling qualifiers and unary operators in C types.
209
210       The function name and the return type must be placed on separate lines
211       and should be flush left-adjusted.
212
213         INCORRECT                        CORRECT
214
215         double sin(x)                    double
216           double x                       sin(x)
217                                            double x
218
219       The rest of the function description may be indented or left-adjusted.
220       The following example shows a function with its body left-adjusted.
221       Most examples in this document will indent the body for better
222       readability.
223
224         CORRECT
225
226         double
227         sin(x)
228         double x
229
230       More complicated XSUBs may contain many other sections.  Each section
231       of an XSUB starts with the corresponding keyword, such as INIT: or
232       CLEANUP:.  However, the first two lines of an XSUB always contain the
233       same data: descriptions of the return type and the names of the
234       function and its parameters.  Whatever immediately follows these is
235       considered to be an INPUT: section unless explicitly marked with
236       another keyword.  (See "The INPUT: Keyword".)
237
238       An XSUB section continues until another section-start keyword is found.
239
240   The Argument Stack
241       The Perl argument stack is used to store the values which are sent as
242       parameters to the XSUB and to store the XSUB's return value(s).  In
243       reality all Perl functions (including non-XSUB ones) keep their values
244       on this stack all the same time, each limited to its own range of
245       positions on the stack.  In this document the first position on that
246       stack which belongs to the active function will be referred to as
247       position 0 for that function.
248
249       XSUBs refer to their stack arguments with the macro ST(x), where x
250       refers to a position in this XSUB's part of the stack.  Position 0 for
251       that function would be known to the XSUB as ST(0).  The XSUB's incoming
252       parameters and outgoing return values always begin at ST(0).  For many
253       simple cases the xsubpp compiler will generate the code necessary to
254       handle the argument stack by embedding code fragments found in the
255       typemaps.  In more complex cases the programmer must supply the code.
256
257   The RETVAL Variable
258       The RETVAL variable is a special C variable that is declared
259       automatically for you.  The C type of RETVAL matches the return type of
260       the C library function.  The xsubpp compiler will declare this variable
261       in each XSUB with non-"void" return type.  By default the generated C
262       function will use RETVAL to hold the return value of the C library
263       function being called.  In simple cases the value of RETVAL will be
264       placed in ST(0) of the argument stack where it can be received by Perl
265       as the return value of the XSUB.
266
267       If the XSUB has a return type of "void" then the compiler will not
268       declare a RETVAL variable for that function.  When using a PPCODE:
269       section no manipulation of the RETVAL variable is required, the section
270       may use direct stack manipulation to place output values on the stack.
271
272       If PPCODE: directive is not used, "void" return value should be used
273       only for subroutines which do not return a value, even if CODE:
274       directive is used which sets ST(0) explicitly.
275
276       Older versions of this document recommended to use "void" return value
277       in such cases. It was discovered that this could lead to segfaults in
278       cases when XSUB was truly "void". This practice is now deprecated, and
279       may be not supported at some future version. Use the return value "SV
280       *" in such cases. (Currently "xsubpp" contains some heuristic code
281       which tries to disambiguate between "truly-void" and "old-practice-
282       declared-as-void" functions. Hence your code is at mercy of this
283       heuristics unless you use "SV *" as return value.)
284
285   Returning SVs, AVs and HVs through RETVAL
286       When you're using RETVAL to return an "SV *", there's some magic going
287       on behind the scenes that should be mentioned. When you're manipulating
288       the argument stack using the ST(x) macro, for example, you usually have
289       to pay special attention to reference counts. (For more about reference
290       counts, see perlguts.) To make your life easier, the typemap file
291       automatically makes "RETVAL" mortal when you're returning an "SV *".
292       Thus, the following two XSUBs are more or less equivalent:
293
294         void
295         alpha()
296             PPCODE:
297                 ST(0) = newSVpv("Hello World",0);
298                 sv_2mortal(ST(0));
299                 XSRETURN(1);
300
301         SV *
302         beta()
303             CODE:
304                 RETVAL = newSVpv("Hello World",0);
305             OUTPUT:
306                 RETVAL
307
308       This is quite useful as it usually improves readability. While this
309       works fine for an "SV *", it's unfortunately not as easy to have "AV *"
310       or "HV *" as a return value. You should be able to write:
311
312         AV *
313         array()
314             CODE:
315                 RETVAL = newAV();
316                 /* do something with RETVAL */
317             OUTPUT:
318                 RETVAL
319
320       But due to an unfixable bug (fixing it would break lots of existing
321       CPAN modules) in the typemap file, the reference count of the "AV *" is
322       not properly decremented. Thus, the above XSUB would leak memory
323       whenever it is being called. The same problem exists for "HV *", "CV
324       *", and "SVREF" (which indicates a scalar reference, not a general "SV
325       *").  In XS code on perls starting with perl 5.16, you can override the
326       typemaps for any of these types with a version that has proper handling
327       of refcounts. In your "TYPEMAP" section, do
328
329         AV*   T_AVREF_REFCOUNT_FIXED
330
331       to get the repaired variant. For backward compatibility with older
332       versions of perl, you can instead decrement the reference count
333       manually when you're returning one of the aforementioned types using
334       "sv_2mortal":
335
336         AV *
337         array()
338             CODE:
339                 RETVAL = newAV();
340                 sv_2mortal((SV*)RETVAL);
341                 /* do something with RETVAL */
342             OUTPUT:
343                 RETVAL
344
345       Remember that you don't have to do this for an "SV *". The reference
346       documentation for all core typemaps can be found in perlxstypemap.
347
348   The MODULE Keyword
349       The MODULE keyword is used to start the XS code and to specify the
350       package of the functions which are being defined.  All text preceding
351       the first MODULE keyword is considered C code and is passed through to
352       the output with POD stripped, but otherwise untouched.  Every XS module
353       will have a bootstrap function which is used to hook the XSUBs into
354       Perl.  The package name of this bootstrap function will match the value
355       of the last MODULE statement in the XS source files.  The value of
356       MODULE should always remain constant within the same XS file, though
357       this is not required.
358
359       The following example will start the XS code and will place all
360       functions in a package named RPC.
361
362            MODULE = RPC
363
364   The PACKAGE Keyword
365       When functions within an XS source file must be separated into packages
366       the PACKAGE keyword should be used.  This keyword is used with the
367       MODULE keyword and must follow immediately after it when used.
368
369            MODULE = RPC  PACKAGE = RPC
370
371            [ XS code in package RPC ]
372
373            MODULE = RPC  PACKAGE = RPCB
374
375            [ XS code in package RPCB ]
376
377            MODULE = RPC  PACKAGE = RPC
378
379            [ XS code in package RPC ]
380
381       The same package name can be used more than once, allowing for non-
382       contiguous code. This is useful if you have a stronger ordering
383       principle than package names.
384
385       Although this keyword is optional and in some cases provides redundant
386       information it should always be used.  This keyword will ensure that
387       the XSUBs appear in the desired package.
388
389   The PREFIX Keyword
390       The PREFIX keyword designates prefixes which should be removed from the
391       Perl function names.  If the C function is rpcb_gettime() and the
392       PREFIX value is "rpcb_" then Perl will see this function as gettime().
393
394       This keyword should follow the PACKAGE keyword when used.  If PACKAGE
395       is not used then PREFIX should follow the MODULE keyword.
396
397            MODULE = RPC  PREFIX = rpc_
398
399            MODULE = RPC  PACKAGE = RPCB  PREFIX = rpcb_
400
401   The OUTPUT: Keyword
402       The OUTPUT: keyword indicates that certain function parameters should
403       be updated (new values made visible to Perl) when the XSUB terminates
404       or that certain values should be returned to the calling Perl function.
405       For simple functions which have no CODE: or PPCODE: section, such as
406       the sin() function above, the RETVAL variable is automatically
407       designated as an output value.  For more complex functions the xsubpp
408       compiler will need help to determine which variables are output
409       variables.
410
411       This keyword will normally be used to complement the CODE: keyword.
412       The RETVAL variable is not recognized as an output variable when the
413       CODE: keyword is present.  The OUTPUT: keyword is used in this
414       situation to tell the compiler that RETVAL really is an output
415       variable.
416
417       The OUTPUT: keyword can also be used to indicate that function
418       parameters are output variables.  This may be necessary when a
419       parameter has been modified within the function and the programmer
420       would like the update to be seen by Perl.
421
422            bool_t
423            rpcb_gettime(host,timep)
424                 char *host
425                 time_t &timep
426               OUTPUT:
427                 timep
428
429       The OUTPUT: keyword will also allow an output parameter to be mapped to
430       a matching piece of code rather than to a typemap.
431
432            bool_t
433            rpcb_gettime(host,timep)
434                 char *host
435                 time_t &timep
436               OUTPUT:
437                 timep sv_setnv(ST(1), (double)timep);
438
439       xsubpp emits an automatic SvSETMAGIC() for all parameters in the OUTPUT
440       section of the XSUB, except RETVAL.  This is the usually desired
441       behavior, as it takes care of properly invoking 'set' magic on output
442       parameters (needed for hash or array element parameters that must be
443       created if they didn't exist).  If for some reason, this behavior is
444       not desired, the OUTPUT section may contain a "SETMAGIC: DISABLE" line
445       to disable it for the remainder of the parameters in the OUTPUT
446       section.  Likewise, "SETMAGIC: ENABLE" can be used to reenable it for
447       the remainder of the OUTPUT section.  See perlguts for more details
448       about 'set' magic.
449
450   The NO_OUTPUT Keyword
451       The NO_OUTPUT can be placed as the first token of the XSUB.  This
452       keyword indicates that while the C subroutine we provide an interface
453       to has a non-"void" return type, the return value of this C subroutine
454       should not be returned from the generated Perl subroutine.
455
456       With this keyword present "The RETVAL Variable" is created, and in the
457       generated call to the subroutine this variable is assigned to, but the
458       value of this variable is not going to be used in the auto-generated
459       code.
460
461       This keyword makes sense only if "RETVAL" is going to be accessed by
462       the user-supplied code.  It is especially useful to make a function
463       interface more Perl-like, especially when the C return value is just an
464       error condition indicator.  For example,
465
466         NO_OUTPUT int
467         delete_file(char *name)
468           POSTCALL:
469             if (RETVAL != 0)
470                 croak("Error %d while deleting file '%s'", RETVAL, name);
471
472       Here the generated XS function returns nothing on success, and will
473       die() with a meaningful error message on error.
474
475   The CODE: Keyword
476       This keyword is used in more complicated XSUBs which require special
477       handling for the C function.  The RETVAL variable is still declared,
478       but it will not be returned unless it is specified in the OUTPUT:
479       section.
480
481       The following XSUB is for a C function which requires special handling
482       of its parameters.  The Perl usage is given first.
483
484            $status = rpcb_gettime( "localhost", $timep );
485
486       The XSUB follows.
487
488            bool_t
489            rpcb_gettime(host,timep)
490                 char *host
491                 time_t timep
492               CODE:
493                      RETVAL = rpcb_gettime( host, &timep );
494               OUTPUT:
495                 timep
496                 RETVAL
497
498   The INIT: Keyword
499       The INIT: keyword allows initialization to be inserted into the XSUB
500       before the compiler generates the call to the C function.  Unlike the
501       CODE: keyword above, this keyword does not affect the way the compiler
502       handles RETVAL.
503
504           bool_t
505           rpcb_gettime(host,timep)
506                 char *host
507                 time_t &timep
508               INIT:
509                 printf("# Host is %s\n", host );
510               OUTPUT:
511                 timep
512
513       Another use for the INIT: section is to check for preconditions before
514       making a call to the C function:
515
516           long long
517           lldiv(a,b)
518               long long a
519               long long b
520             INIT:
521               if (a == 0 && b == 0)
522                   XSRETURN_UNDEF;
523               if (b == 0)
524                   croak("lldiv: cannot divide by 0");
525
526   The NO_INIT Keyword
527       The NO_INIT keyword is used to indicate that a function parameter is
528       being used only as an output value.  The xsubpp compiler will normally
529       generate code to read the values of all function parameters from the
530       argument stack and assign them to C variables upon entry to the
531       function.  NO_INIT will tell the compiler that some parameters will be
532       used for output rather than for input and that they will be handled
533       before the function terminates.
534
535       The following example shows a variation of the rpcb_gettime() function.
536       This function uses the timep variable only as an output variable and
537       does not care about its initial contents.
538
539            bool_t
540            rpcb_gettime(host,timep)
541                 char *host
542                 time_t &timep = NO_INIT
543               OUTPUT:
544                 timep
545
546   The TYPEMAP: Keyword
547       Starting with Perl 5.16, you can embed typemaps into your XS code
548       instead of or in addition to typemaps in a separate file.  Multiple
549       such embedded typemaps will be processed in order of appearance in the
550       XS code and like local typemap files take precedence over the default
551       typemap, the embedded typemaps may overwrite previous definitions of
552       TYPEMAP, INPUT, and OUTPUT stanzas.  The syntax for embedded typemaps
553       is
554
555             TYPEMAP: <<HERE
556             ... your typemap code here ...
557             HERE
558
559       where the "TYPEMAP" keyword must appear in the first column of a new
560       line.
561
562       Refer to perlxstypemap for details on writing typemaps.
563
564   Initializing Function Parameters
565       C function parameters are normally initialized with their values from
566       the argument stack (which in turn contains the parameters that were
567       passed to the XSUB from Perl).  The typemaps contain the code segments
568       which are used to translate the Perl values to the C parameters.  The
569       programmer, however, is allowed to override the typemaps and supply
570       alternate (or additional) initialization code.  Initialization code
571       starts with the first "=", ";" or "+" on a line in the INPUT: section.
572       The only exception happens if this ";" terminates the line, then this
573       ";" is quietly ignored.
574
575       The following code demonstrates how to supply initialization code for
576       function parameters.  The initialization code is eval'ed within double
577       quotes by the compiler before it is added to the output so anything
578       which should be interpreted literally [mainly "$", "@", or "\\"] must
579       be protected with backslashes.  The variables $var, $arg, and $type can
580       be used as in typemaps.
581
582            bool_t
583            rpcb_gettime(host,timep)
584                 char *host = (char *)SvPVbyte_nolen($arg);
585                 time_t &timep = 0;
586               OUTPUT:
587                 timep
588
589       This should not be used to supply default values for parameters.  One
590       would normally use this when a function parameter must be processed by
591       another library function before it can be used.  Default parameters are
592       covered in the next section.
593
594       If the initialization begins with "=", then it is output in the
595       declaration for the input variable, replacing the initialization
596       supplied by the typemap.  If the initialization begins with ";" or "+",
597       then it is performed after all of the input variables have been
598       declared.  In the ";" case the initialization normally supplied by the
599       typemap is not performed.  For the "+" case, the declaration for the
600       variable will include the initialization from the typemap.  A global
601       variable, %v, is available for the truly rare case where information
602       from one initialization is needed in another initialization.
603
604       Here's a truly obscure example:
605
606            bool_t
607            rpcb_gettime(host,timep)
608                 time_t &timep; /* \$v{timep}=@{[$v{timep}=$arg]} */
609                 char *host + SvOK($v{timep}) ? SvPVbyte_nolen($arg) : NULL;
610               OUTPUT:
611                 timep
612
613       The construct "\$v{timep}=@{[$v{timep}=$arg]}" used in the above
614       example has a two-fold purpose: first, when this line is processed by
615       xsubpp, the Perl snippet "$v{timep}=$arg" is evaluated.  Second, the
616       text of the evaluated snippet is output into the generated C file
617       (inside a C comment)!  During the processing of "char *host" line, $arg
618       will evaluate to ST(0), and $v{timep} will evaluate to ST(1).
619
620   Default Parameter Values
621       Default values for XSUB arguments can be specified by placing an
622       assignment statement in the parameter list.  The default value may be a
623       number, a string or the special string "NO_INIT".  Defaults should
624       always be used on the right-most parameters only.
625
626       To allow the XSUB for rpcb_gettime() to have a default host value the
627       parameters to the XSUB could be rearranged.  The XSUB will then call
628       the real rpcb_gettime() function with the parameters in the correct
629       order.  This XSUB can be called from Perl with either of the following
630       statements:
631
632            $status = rpcb_gettime( $timep, $host );
633
634            $status = rpcb_gettime( $timep );
635
636       The XSUB will look like the code which follows.  A CODE: block is used
637       to call the real rpcb_gettime() function with the parameters in the
638       correct order for that function.
639
640            bool_t
641            rpcb_gettime(timep,host="localhost")
642                 char *host
643                 time_t timep = NO_INIT
644               CODE:
645                      RETVAL = rpcb_gettime( host, &timep );
646               OUTPUT:
647                 timep
648                 RETVAL
649
650   The PREINIT: Keyword
651       The PREINIT: keyword allows extra variables to be declared immediately
652       before or after the declarations of the parameters from the INPUT:
653       section are emitted.
654
655       If a variable is declared inside a CODE: section it will follow any
656       typemap code that is emitted for the input parameters.  This may result
657       in the declaration ending up after C code, which is C syntax error.
658       Similar errors may happen with an explicit ";"-type or "+"-type
659       initialization of parameters is used (see "Initializing Function
660       Parameters").  Declaring these variables in an INIT: section will not
661       help.
662
663       In such cases, to force an additional variable to be declared together
664       with declarations of other variables, place the declaration into a
665       PREINIT: section.  The PREINIT: keyword may be used one or more times
666       within an XSUB.
667
668       The following examples are equivalent, but if the code is using complex
669       typemaps then the first example is safer.
670
671            bool_t
672            rpcb_gettime(timep)
673                 time_t timep = NO_INIT
674               PREINIT:
675                 char *host = "localhost";
676               CODE:
677                 RETVAL = rpcb_gettime( host, &timep );
678               OUTPUT:
679                 timep
680                 RETVAL
681
682       For this particular case an INIT: keyword would generate the same C
683       code as the PREINIT: keyword.  Another correct, but error-prone
684       example:
685
686            bool_t
687            rpcb_gettime(timep)
688                 time_t timep = NO_INIT
689               CODE:
690                 char *host = "localhost";
691                 RETVAL = rpcb_gettime( host, &timep );
692               OUTPUT:
693                 timep
694                 RETVAL
695
696       Another way to declare "host" is to use a C block in the CODE: section:
697
698            bool_t
699            rpcb_gettime(timep)
700                 time_t timep = NO_INIT
701               CODE:
702                 {
703                   char *host = "localhost";
704                   RETVAL = rpcb_gettime( host, &timep );
705                 }
706               OUTPUT:
707                 timep
708                 RETVAL
709
710       The ability to put additional declarations before the typemap entries
711       are processed is very handy in the cases when typemap conversions
712       manipulate some global state:
713
714           MyObject
715           mutate(o)
716               PREINIT:
717                   MyState st = global_state;
718               INPUT:
719                   MyObject o;
720               CLEANUP:
721                   reset_to(global_state, st);
722
723       Here we suppose that conversion to "MyObject" in the INPUT: section and
724       from MyObject when processing RETVAL will modify a global variable
725       "global_state".  After these conversions are performed, we restore the
726       old value of "global_state" (to avoid memory leaks, for example).
727
728       There is another way to trade clarity for compactness: INPUT sections
729       allow declaration of C variables which do not appear in the parameter
730       list of a subroutine.  Thus the above code for mutate() can be
731       rewritten as
732
733           MyObject
734           mutate(o)
735                 MyState st = global_state;
736                 MyObject o;
737               CLEANUP:
738                 reset_to(global_state, st);
739
740       and the code for rpcb_gettime() can be rewritten as
741
742            bool_t
743            rpcb_gettime(timep)
744                 time_t timep = NO_INIT
745                 char *host = "localhost";
746               C_ARGS:
747                 host, &timep
748               OUTPUT:
749                 timep
750                 RETVAL
751
752   The SCOPE: Keyword
753       The SCOPE: keyword allows scoping to be enabled for a particular XSUB.
754       If enabled, the XSUB will invoke ENTER and LEAVE automatically.
755
756       To support potentially complex type mappings, if a typemap entry used
757       by an XSUB contains a comment like "/*scope*/" then scoping will be
758       automatically enabled for that XSUB.
759
760       To enable scoping:
761
762           SCOPE: ENABLE
763
764       To disable scoping:
765
766           SCOPE: DISABLE
767
768   The INPUT: Keyword
769       The XSUB's parameters are usually evaluated immediately after entering
770       the XSUB.  The INPUT: keyword can be used to force those parameters to
771       be evaluated a little later.  The INPUT: keyword can be used multiple
772       times within an XSUB and can be used to list one or more input
773       variables.  This keyword is used with the PREINIT: keyword.
774
775       The following example shows how the input parameter "timep" can be
776       evaluated late, after a PREINIT.
777
778           bool_t
779           rpcb_gettime(host,timep)
780                 char *host
781               PREINIT:
782                 time_t tt;
783               INPUT:
784                 time_t timep
785               CODE:
786                      RETVAL = rpcb_gettime( host, &tt );
787                      timep = tt;
788               OUTPUT:
789                 timep
790                 RETVAL
791
792       The next example shows each input parameter evaluated late.
793
794           bool_t
795           rpcb_gettime(host,timep)
796               PREINIT:
797                 time_t tt;
798               INPUT:
799                 char *host
800               PREINIT:
801                 char *h;
802               INPUT:
803                 time_t timep
804               CODE:
805                      h = host;
806                      RETVAL = rpcb_gettime( h, &tt );
807                      timep = tt;
808               OUTPUT:
809                 timep
810                 RETVAL
811
812       Since INPUT sections allow declaration of C variables which do not
813       appear in the parameter list of a subroutine, this may be shortened to:
814
815           bool_t
816           rpcb_gettime(host,timep)
817                 time_t tt;
818                 char *host;
819                 char *h = host;
820                 time_t timep;
821               CODE:
822                 RETVAL = rpcb_gettime( h, &tt );
823                 timep = tt;
824               OUTPUT:
825                 timep
826                 RETVAL
827
828       (We used our knowledge that input conversion for "char *" is a "simple"
829       one, thus "host" is initialized on the declaration line, and our
830       assignment "h = host" is not performed too early.  Otherwise one would
831       need to have the assignment "h = host" in a CODE: or INIT: section.)
832
833   The IN/OUTLIST/IN_OUTLIST/OUT/IN_OUT Keywords
834       In the list of parameters for an XSUB, one can precede parameter names
835       by the "IN"/"OUTLIST"/"IN_OUTLIST"/"OUT"/"IN_OUT" keywords.  "IN"
836       keyword is the default, the other keywords indicate how the Perl
837       interface should differ from the C interface.
838
839       Parameters preceded by "OUTLIST"/"IN_OUTLIST"/"OUT"/"IN_OUT" keywords
840       are considered to be used by the C subroutine via pointers.
841       "OUTLIST"/"OUT" keywords indicate that the C subroutine does not
842       inspect the memory pointed by this parameter, but will write through
843       this pointer to provide additional return values.
844
845       Parameters preceded by "OUTLIST" keyword do not appear in the usage
846       signature of the generated Perl function.
847
848       Parameters preceded by "IN_OUTLIST"/"IN_OUT"/"OUT" do appear as
849       parameters to the Perl function.  With the exception of
850       "OUT"-parameters, these parameters are converted to the corresponding C
851       type, then pointers to these data are given as arguments to the C
852       function.  It is expected that the C function will write through these
853       pointers.
854
855       The return list of the generated Perl function consists of the C return
856       value from the function (unless the XSUB is of "void" return type or
857       "The NO_OUTPUT Keyword" was used) followed by all the "OUTLIST" and
858       "IN_OUTLIST" parameters (in the order of appearance).  On the return
859       from the XSUB the "IN_OUT"/"OUT" Perl parameter will be modified to
860       have the values written by the C function.
861
862       For example, an XSUB
863
864         void
865         day_month(OUTLIST day, IN unix_time, OUTLIST month)
866           int day
867           int unix_time
868           int month
869
870       should be used from Perl as
871
872         my ($day, $month) = day_month(time);
873
874       The C signature of the corresponding function should be
875
876         void day_month(int *day, int unix_time, int *month);
877
878       The "IN"/"OUTLIST"/"IN_OUTLIST"/"IN_OUT"/"OUT" keywords can be mixed
879       with ANSI-style declarations, as in
880
881         void
882         day_month(OUTLIST int day, int unix_time, OUTLIST int month)
883
884       (here the optional "IN" keyword is omitted).
885
886       The "IN_OUT" parameters are identical with parameters introduced with
887       "The & Unary Operator" and put into the "OUTPUT:" section (see "The
888       OUTPUT: Keyword").  The "IN_OUTLIST" parameters are very similar, the
889       only difference being that the value C function writes through the
890       pointer would not modify the Perl parameter, but is put in the output
891       list.
892
893       The "OUTLIST"/"OUT" parameter differ from "IN_OUTLIST"/"IN_OUT"
894       parameters only by the initial value of the Perl parameter not being
895       read (and not being given to the C function - which gets some garbage
896       instead).  For example, the same C function as above can be interfaced
897       with as
898
899         void day_month(OUT int day, int unix_time, OUT int month);
900
901       or
902
903         void
904         day_month(day, unix_time, month)
905             int &day = NO_INIT
906             int  unix_time
907             int &month = NO_INIT
908           OUTPUT:
909             day
910             month
911
912       However, the generated Perl function is called in very C-ish style:
913
914         my ($day, $month);
915         day_month($day, time, $month);
916
917   The length(NAME) Keyword
918       If one of the input arguments to the C function is the length of a
919       string argument "NAME", one can substitute the name of the length-
920       argument by length(NAME) in the XSUB declaration.  This argument must
921       be omitted when the generated Perl function is called.  E.g.,
922
923         void
924         dump_chars(char *s, short l)
925         {
926           short n = 0;
927           while (n < l) {
928               printf("s[%d] = \"\\%#03o\"\n", n, (int)s[n]);
929               n++;
930           }
931         }
932
933         MODULE = x            PACKAGE = x
934
935         void dump_chars(char *s, short length(s))
936
937       should be called as dump_chars($string).
938
939       This directive is supported with ANSI-type function declarations only.
940
941   Variable-length Parameter Lists
942       XSUBs can have variable-length parameter lists by specifying an
943       ellipsis "(...)" in the parameter list.  This use of the ellipsis is
944       similar to that found in ANSI C.  The programmer is able to determine
945       the number of arguments passed to the XSUB by examining the "items"
946       variable which the xsubpp compiler supplies for all XSUBs.  By using
947       this mechanism one can create an XSUB which accepts a list of
948       parameters of unknown length.
949
950       The host parameter for the rpcb_gettime() XSUB can be optional so the
951       ellipsis can be used to indicate that the XSUB will take a variable
952       number of parameters.  Perl should be able to call this XSUB with
953       either of the following statements.
954
955            $status = rpcb_gettime( $timep, $host );
956
957            $status = rpcb_gettime( $timep );
958
959       The XS code, with ellipsis, follows.
960
961            bool_t
962            rpcb_gettime(timep, ...)
963                 time_t timep = NO_INIT
964               PREINIT:
965                 char *host = "localhost";
966               CODE:
967                 if( items > 1 )
968                      host = (char *)SvPVbyte_nolen(ST(1));
969                 RETVAL = rpcb_gettime( host, &timep );
970               OUTPUT:
971                 timep
972                 RETVAL
973
974   The C_ARGS: Keyword
975       The C_ARGS: keyword allows creating of XSUBS which have different
976       calling sequence from Perl than from C, without a need to write CODE:
977       or PPCODE: section.  The contents of the C_ARGS: paragraph is put as
978       the argument to the called C function without any change.
979
980       For example, suppose that a C function is declared as
981
982           symbolic nth_derivative(int n, symbolic function, int flags);
983
984       and that the default flags are kept in a global C variable
985       "default_flags".  Suppose that you want to create an interface which is
986       called as
987
988           $second_deriv = $function->nth_derivative(2);
989
990       To do this, declare the XSUB as
991
992           symbolic
993           nth_derivative(function, n)
994               symbolic        function
995               int             n
996             C_ARGS:
997               n, function, default_flags
998
999   The PPCODE: Keyword
1000       The PPCODE: keyword is an alternate form of the CODE: keyword and is
1001       used to tell the xsubpp compiler that the programmer is supplying the
1002       code to control the argument stack for the XSUBs return values.
1003       Occasionally one will want an XSUB to return a list of values rather
1004       than a single value.  In these cases one must use PPCODE: and then
1005       explicitly push the list of values on the stack.  The PPCODE: and CODE:
1006       keywords should not be used together within the same XSUB.
1007
1008       The actual difference between PPCODE: and CODE: sections is in the
1009       initialization of "SP" macro (which stands for the current Perl stack
1010       pointer), and in the handling of data on the stack when returning from
1011       an XSUB.  In CODE: sections SP preserves the value which was on entry
1012       to the XSUB: SP is on the function pointer (which follows the last
1013       parameter).  In PPCODE: sections SP is moved backward to the beginning
1014       of the parameter list, which allows "PUSH*()" macros to place output
1015       values in the place Perl expects them to be when the XSUB returns back
1016       to Perl.
1017
1018       The generated trailer for a CODE: section ensures that the number of
1019       return values Perl will see is either 0 or 1 (depending on the
1020       "void"ness of the return value of the C function, and heuristics
1021       mentioned in "The RETVAL Variable").  The trailer generated for a
1022       PPCODE: section is based on the number of return values and on the
1023       number of times "SP" was updated by "[X]PUSH*()" macros.
1024
1025       Note that macros ST(i), "XST_m*()" and "XSRETURN*()" work equally well
1026       in CODE: sections and PPCODE: sections.
1027
1028       The following XSUB will call the C rpcb_gettime() function and will
1029       return its two output values, timep and status, to Perl as a single
1030       list.
1031
1032            void
1033            rpcb_gettime(host)
1034                 char *host
1035               PREINIT:
1036                 time_t  timep;
1037                 bool_t  status;
1038               PPCODE:
1039                 status = rpcb_gettime( host, &timep );
1040                 EXTEND(SP, 2);
1041                 PUSHs(sv_2mortal(newSViv(status)));
1042                 PUSHs(sv_2mortal(newSViv(timep)));
1043
1044       Notice that the programmer must supply the C code necessary to have the
1045       real rpcb_gettime() function called and to have the return values
1046       properly placed on the argument stack.
1047
1048       The "void" return type for this function tells the xsubpp compiler that
1049       the RETVAL variable is not needed or used and that it should not be
1050       created.  In most scenarios the void return type should be used with
1051       the PPCODE: directive.
1052
1053       The EXTEND() macro is used to make room on the argument stack for 2
1054       return values.  The PPCODE: directive causes the xsubpp compiler to
1055       create a stack pointer available as "SP", and it is this pointer which
1056       is being used in the EXTEND() macro.  The values are then pushed onto
1057       the stack with the PUSHs() macro.
1058
1059       Now the rpcb_gettime() function can be used from Perl with the
1060       following statement.
1061
1062            ($status, $timep) = rpcb_gettime("localhost");
1063
1064       When handling output parameters with a PPCODE section, be sure to
1065       handle 'set' magic properly.  See perlguts for details about 'set'
1066       magic.
1067
1068   Returning Undef And Empty Lists
1069       Occasionally the programmer will want to return simply "undef" or an
1070       empty list if a function fails rather than a separate status value.
1071       The rpcb_gettime() function offers just this situation.  If the
1072       function succeeds we would like to have it return the time and if it
1073       fails we would like to have undef returned.  In the following Perl code
1074       the value of $timep will either be undef or it will be a valid time.
1075
1076            $timep = rpcb_gettime( "localhost" );
1077
1078       The following XSUB uses the "SV *" return type as a mnemonic only, and
1079       uses a CODE: block to indicate to the compiler that the programmer has
1080       supplied all the necessary code.  The sv_newmortal() call will
1081       initialize the return value to undef, making that the default return
1082       value.
1083
1084            SV *
1085            rpcb_gettime(host)
1086                 char *  host
1087               PREINIT:
1088                 time_t  timep;
1089                 bool_t x;
1090               CODE:
1091                 ST(0) = sv_newmortal();
1092                 if( rpcb_gettime( host, &timep ) )
1093                      sv_setnv( ST(0), (double)timep);
1094
1095       The next example demonstrates how one would place an explicit undef in
1096       the return value, should the need arise.
1097
1098            SV *
1099            rpcb_gettime(host)
1100                 char *  host
1101               PREINIT:
1102                 time_t  timep;
1103                 bool_t x;
1104               CODE:
1105                 if( rpcb_gettime( host, &timep ) ){
1106                      ST(0) = sv_newmortal();
1107                      sv_setnv( ST(0), (double)timep);
1108                 }
1109                 else{
1110                      ST(0) = &PL_sv_undef;
1111                 }
1112
1113       To return an empty list one must use a PPCODE: block and then not push
1114       return values on the stack.
1115
1116            void
1117            rpcb_gettime(host)
1118                 char *host
1119               PREINIT:
1120                 time_t  timep;
1121               PPCODE:
1122                 if( rpcb_gettime( host, &timep ) )
1123                      PUSHs(sv_2mortal(newSViv(timep)));
1124                 else{
1125                     /* Nothing pushed on stack, so an empty
1126                      * list is implicitly returned. */
1127                 }
1128
1129       Some people may be inclined to include an explicit "return" in the
1130       above XSUB, rather than letting control fall through to the end.  In
1131       those situations "XSRETURN_EMPTY" should be used, instead.  This will
1132       ensure that the XSUB stack is properly adjusted.  Consult perlapi for
1133       other "XSRETURN" macros.
1134
1135       Since "XSRETURN_*" macros can be used with CODE blocks as well, one can
1136       rewrite this example as:
1137
1138            int
1139            rpcb_gettime(host)
1140                 char *host
1141               PREINIT:
1142                 time_t  timep;
1143               CODE:
1144                 RETVAL = rpcb_gettime( host, &timep );
1145                 if (RETVAL == 0)
1146                       XSRETURN_UNDEF;
1147               OUTPUT:
1148                 RETVAL
1149
1150       In fact, one can put this check into a POSTCALL: section as well.
1151       Together with PREINIT: simplifications, this leads to:
1152
1153            int
1154            rpcb_gettime(host)
1155                 char *host
1156                 time_t  timep;
1157               POSTCALL:
1158                 if (RETVAL == 0)
1159                       XSRETURN_UNDEF;
1160
1161   The REQUIRE: Keyword
1162       The REQUIRE: keyword is used to indicate the minimum version of the
1163       xsubpp compiler needed to compile the XS module.  An XS module which
1164       contains the following statement will compile with only xsubpp version
1165       1.922 or greater:
1166
1167               REQUIRE: 1.922
1168
1169   The CLEANUP: Keyword
1170       This keyword can be used when an XSUB requires special cleanup
1171       procedures before it terminates.  When the CLEANUP: keyword is used it
1172       must follow any CODE:, or OUTPUT: blocks which are present in the XSUB.
1173       The code specified for the cleanup block will be added as the last
1174       statements in the XSUB.
1175
1176   The POSTCALL: Keyword
1177       This keyword can be used when an XSUB requires special procedures
1178       executed after the C subroutine call is performed.  When the POSTCALL:
1179       keyword is used it must precede OUTPUT: and CLEANUP: blocks which are
1180       present in the XSUB.
1181
1182       See examples in "The NO_OUTPUT Keyword" and "Returning Undef And Empty
1183       Lists".
1184
1185       The POSTCALL: block does not make a lot of sense when the C subroutine
1186       call is supplied by user by providing either CODE: or PPCODE: section.
1187
1188   The BOOT: Keyword
1189       The BOOT: keyword is used to add code to the extension's bootstrap
1190       function.  The bootstrap function is generated by the xsubpp compiler
1191       and normally holds the statements necessary to register any XSUBs with
1192       Perl.  With the BOOT: keyword the programmer can tell the compiler to
1193       add extra statements to the bootstrap function.
1194
1195       This keyword may be used any time after the first MODULE keyword and
1196       should appear on a line by itself.  The first blank line after the
1197       keyword will terminate the code block.
1198
1199            BOOT:
1200            # The following message will be printed when the
1201            # bootstrap function executes.
1202            printf("Hello from the bootstrap!\n");
1203
1204   The VERSIONCHECK: Keyword
1205       The VERSIONCHECK: keyword corresponds to xsubpp's "-versioncheck" and
1206       "-noversioncheck" options.  This keyword overrides the command line
1207       options.  Version checking is enabled by default.  When version
1208       checking is enabled the XS module will attempt to verify that its
1209       version matches the version of the PM module.
1210
1211       To enable version checking:
1212
1213           VERSIONCHECK: ENABLE
1214
1215       To disable version checking:
1216
1217           VERSIONCHECK: DISABLE
1218
1219       Note that if the version of the PM module is an NV (a floating point
1220       number), it will be stringified with a possible loss of precision
1221       (currently chopping to nine decimal places) so that it may not match
1222       the version of the XS module anymore. Quoting the $VERSION declaration
1223       to make it a string is recommended if long version numbers are used.
1224
1225   The PROTOTYPES: Keyword
1226       The PROTOTYPES: keyword corresponds to xsubpp's "-prototypes" and
1227       "-noprototypes" options.  This keyword overrides the command line
1228       options.  Prototypes are disabled by default.  When prototypes are
1229       enabled, XSUBs will be given Perl prototypes.  This keyword may be used
1230       multiple times in an XS module to enable and disable prototypes for
1231       different parts of the module.  Note that xsubpp will nag you if you
1232       don't explicitly enable or disable prototypes, with:
1233
1234           Please specify prototyping behavior for Foo.xs (see perlxs manual)
1235
1236       To enable prototypes:
1237
1238           PROTOTYPES: ENABLE
1239
1240       To disable prototypes:
1241
1242           PROTOTYPES: DISABLE
1243
1244   The PROTOTYPE: Keyword
1245       This keyword is similar to the PROTOTYPES: keyword above but can be
1246       used to force xsubpp to use a specific prototype for the XSUB.  This
1247       keyword overrides all other prototype options and keywords but affects
1248       only the current XSUB.  Consult "Prototypes" in perlsub for information
1249       about Perl prototypes.
1250
1251           bool_t
1252           rpcb_gettime(timep, ...)
1253                 time_t timep = NO_INIT
1254               PROTOTYPE: $;$
1255               PREINIT:
1256                 char *host = "localhost";
1257               CODE:
1258                         if( items > 1 )
1259                              host = (char *)SvPVbyte_nolen(ST(1));
1260                         RETVAL = rpcb_gettime( host, &timep );
1261               OUTPUT:
1262                 timep
1263                 RETVAL
1264
1265       If the prototypes are enabled, you can disable it locally for a given
1266       XSUB as in the following example:
1267
1268           void
1269           rpcb_gettime_noproto()
1270               PROTOTYPE: DISABLE
1271           ...
1272
1273   The ALIAS: Keyword
1274       The ALIAS: keyword allows an XSUB to have two or more unique Perl names
1275       and to know which of those names was used when it was invoked.  The
1276       Perl names may be fully-qualified with package names.  Each alias is
1277       given an index.  The compiler will setup a variable called "ix" which
1278       contain the index of the alias which was used.  When the XSUB is called
1279       with its declared name "ix" will be 0.
1280
1281       The following example will create aliases FOO::gettime() and
1282       BAR::getit() for this function.
1283
1284           bool_t
1285           rpcb_gettime(host,timep)
1286                 char *host
1287                 time_t &timep
1288               ALIAS:
1289                   FOO::gettime = 1
1290                   BAR::getit = 2
1291               INIT:
1292                 printf("# ix = %d\n", ix );
1293               OUTPUT:
1294                 timep
1295
1296   The OVERLOAD: Keyword
1297       Instead of writing an overloaded interface using pure Perl, you can
1298       also use the OVERLOAD keyword to define additional Perl names for your
1299       functions (like the ALIAS: keyword above).  However, the overloaded
1300       functions must be defined in such a way as to accept the number of
1301       parameters supplied by perl's overload system.  For most overload
1302       methods, it will be three parameters; for the "nomethod" function it
1303       will be four.  However, the bitwise operators "&", "|", "^", and "~"
1304       may be called with three or five arguments (see overload).
1305
1306       If any function has the OVERLOAD: keyword, several additional lines
1307       will be defined in the c file generated by xsubpp in order to register
1308       with the overload magic.
1309
1310       Since blessed objects are actually stored as RV's, it is useful to use
1311       the typemap features to preprocess parameters and extract the actual SV
1312       stored within the blessed RV.  See the sample for T_PTROBJ_SPECIAL
1313       below.
1314
1315       To use the OVERLOAD: keyword, create an XS function which takes three
1316       input parameters (or use the C-style '...' definition) like this:
1317
1318           SV *
1319           cmp (lobj, robj, swap)
1320           My_Module_obj    lobj
1321           My_Module_obj    robj
1322           IV               swap
1323           OVERLOAD: cmp <=>
1324           { /* function defined here */}
1325
1326       In this case, the function will overload both of the three way
1327       comparison operators.  For all overload operations using non-alpha
1328       characters, you must type the parameter without quoting, separating
1329       multiple overloads with whitespace.  Note that "" (the stringify
1330       overload) should be entered as \"\" (i.e. escaped).
1331
1332       Since, as mentioned above, bitwise operators may take extra arguments,
1333       you may want to use something like "(lobj, robj, swap, ...)" (with
1334       literal "...") as your parameter list.
1335
1336   The FALLBACK: Keyword
1337       In addition to the OVERLOAD keyword, if you need to control how Perl
1338       autogenerates missing overloaded operators, you can set the FALLBACK
1339       keyword in the module header section, like this:
1340
1341           MODULE = RPC  PACKAGE = RPC
1342
1343           FALLBACK: TRUE
1344           ...
1345
1346       where FALLBACK can take any of the three values TRUE, FALSE, or UNDEF.
1347       If you do not set any FALLBACK value when using OVERLOAD, it defaults
1348       to UNDEF.  FALLBACK is not used except when one or more functions using
1349       OVERLOAD have been defined.  Please see "fallback" in overload for more
1350       details.
1351
1352   The INTERFACE: Keyword
1353       This keyword declares the current XSUB as a keeper of the given calling
1354       signature.  If some text follows this keyword, it is considered as a
1355       list of functions which have this signature, and should be attached to
1356       the current XSUB.
1357
1358       For example, if you have 4 C functions multiply(), divide(), add(),
1359       subtract() all having the signature:
1360
1361           symbolic f(symbolic, symbolic);
1362
1363       you can make them all to use the same XSUB using this:
1364
1365           symbolic
1366           interface_s_ss(arg1, arg2)
1367               symbolic        arg1
1368               symbolic        arg2
1369           INTERFACE:
1370               multiply divide
1371               add subtract
1372
1373       (This is the complete XSUB code for 4 Perl functions!)  Four generated
1374       Perl function share names with corresponding C functions.
1375
1376       The advantage of this approach comparing to ALIAS: keyword is that
1377       there is no need to code a switch statement, each Perl function (which
1378       shares the same XSUB) knows which C function it should call.
1379       Additionally, one can attach an extra function remainder() at runtime
1380       by using
1381
1382           CV *mycv = newXSproto("Symbolic::remainder",
1383                                 XS_Symbolic_interface_s_ss, __FILE__, "$$");
1384           XSINTERFACE_FUNC_SET(mycv, remainder);
1385
1386       say, from another XSUB.  (This example supposes that there was no
1387       INTERFACE_MACRO: section, otherwise one needs to use something else
1388       instead of "XSINTERFACE_FUNC_SET", see the next section.)
1389
1390   The INTERFACE_MACRO: Keyword
1391       This keyword allows one to define an INTERFACE using a different way to
1392       extract a function pointer from an XSUB.  The text which follows this
1393       keyword should give the name of macros which would extract/set a
1394       function pointer.  The extractor macro is given return type, "CV*", and
1395       "XSANY.any_dptr" for this "CV*".  The setter macro is given cv, and the
1396       function pointer.
1397
1398       The default value is "XSINTERFACE_FUNC" and "XSINTERFACE_FUNC_SET".  An
1399       INTERFACE keyword with an empty list of functions can be omitted if
1400       INTERFACE_MACRO keyword is used.
1401
1402       Suppose that in the previous example functions pointers for multiply(),
1403       divide(), add(), subtract() are kept in a global C array "fp[]" with
1404       offsets being "multiply_off", "divide_off", "add_off", "subtract_off".
1405       Then one can use
1406
1407           #define XSINTERFACE_FUNC_BYOFFSET(ret,cv,f) \
1408               ((XSINTERFACE_CVT_ANON(ret))fp[CvXSUBANY(cv).any_i32])
1409           #define XSINTERFACE_FUNC_BYOFFSET_set(cv,f) \
1410               CvXSUBANY(cv).any_i32 = CAT2( f, _off )
1411
1412       in C section,
1413
1414           symbolic
1415           interface_s_ss(arg1, arg2)
1416               symbolic        arg1
1417               symbolic        arg2
1418             INTERFACE_MACRO:
1419               XSINTERFACE_FUNC_BYOFFSET
1420               XSINTERFACE_FUNC_BYOFFSET_set
1421             INTERFACE:
1422               multiply divide
1423               add subtract
1424
1425       in XSUB section.
1426
1427   The INCLUDE: Keyword
1428       This keyword can be used to pull other files into the XS module.  The
1429       other files may have XS code.  INCLUDE: can also be used to run a
1430       command to generate the XS code to be pulled into the module.
1431
1432       The file Rpcb1.xsh contains our rpcb_gettime() function:
1433
1434           bool_t
1435           rpcb_gettime(host,timep)
1436                 char *host
1437                 time_t &timep
1438               OUTPUT:
1439                 timep
1440
1441       The XS module can use INCLUDE: to pull that file into it.
1442
1443           INCLUDE: Rpcb1.xsh
1444
1445       If the parameters to the INCLUDE: keyword are followed by a pipe ("|")
1446       then the compiler will interpret the parameters as a command. This
1447       feature is mildly deprecated in favour of the "INCLUDE_COMMAND:"
1448       directive, as documented below.
1449
1450           INCLUDE: cat Rpcb1.xsh |
1451
1452       Do not use this to run perl: "INCLUDE: perl |" will run the perl that
1453       happens to be the first in your path and not necessarily the same perl
1454       that is used to run "xsubpp". See "The INCLUDE_COMMAND: Keyword".
1455
1456   The INCLUDE_COMMAND: Keyword
1457       Runs the supplied command and includes its output into the current XS
1458       document. "INCLUDE_COMMAND" assigns special meaning to the $^X token in
1459       that it runs the same perl interpreter that is running "xsubpp":
1460
1461           INCLUDE_COMMAND: cat Rpcb1.xsh
1462
1463           INCLUDE_COMMAND: $^X -e ...
1464
1465   The CASE: Keyword
1466       The CASE: keyword allows an XSUB to have multiple distinct parts with
1467       each part acting as a virtual XSUB.  CASE: is greedy and if it is used
1468       then all other XS keywords must be contained within a CASE:.  This
1469       means nothing may precede the first CASE: in the XSUB and anything
1470       following the last CASE: is included in that case.
1471
1472       A CASE: might switch via a parameter of the XSUB, via the "ix" ALIAS:
1473       variable (see "The ALIAS: Keyword"), or maybe via the "items" variable
1474       (see "Variable-length Parameter Lists").  The last CASE: becomes the
1475       default case if it is not associated with a conditional.  The following
1476       example shows CASE switched via "ix" with a function rpcb_gettime()
1477       having an alias x_gettime().  When the function is called as
1478       rpcb_gettime() its parameters are the usual "(char *host, time_t
1479       *timep)", but when the function is called as x_gettime() its parameters
1480       are reversed, "(time_t *timep, char *host)".
1481
1482           long
1483           rpcb_gettime(a,b)
1484             CASE: ix == 1
1485               ALIAS:
1486                 x_gettime = 1
1487               INPUT:
1488                 # 'a' is timep, 'b' is host
1489                 char *b
1490                 time_t a = NO_INIT
1491               CODE:
1492                      RETVAL = rpcb_gettime( b, &a );
1493               OUTPUT:
1494                 a
1495                 RETVAL
1496             CASE:
1497                 # 'a' is host, 'b' is timep
1498                 char *a
1499                 time_t &b = NO_INIT
1500               OUTPUT:
1501                 b
1502                 RETVAL
1503
1504       That function can be called with either of the following statements.
1505       Note the different argument lists.
1506
1507               $status = rpcb_gettime( $host, $timep );
1508
1509               $status = x_gettime( $timep, $host );
1510
1511   The EXPORT_XSUB_SYMBOLS: Keyword
1512       The EXPORT_XSUB_SYMBOLS: keyword is likely something you will never
1513       need.  In perl versions earlier than 5.16.0, this keyword does nothing.
1514       Starting with 5.16, XSUB symbols are no longer exported by default.
1515       That is, they are "static" functions. If you include
1516
1517         EXPORT_XSUB_SYMBOLS: ENABLE
1518
1519       in your XS code, the XSUBs following this line will not be declared
1520       "static".  You can later disable this with
1521
1522         EXPORT_XSUB_SYMBOLS: DISABLE
1523
1524       which, again, is the default that you should probably never change.
1525       You cannot use this keyword on versions of perl before 5.16 to make
1526       XSUBs "static".
1527
1528   The & Unary Operator
1529       The "&" unary operator in the INPUT: section is used to tell xsubpp
1530       that it should convert a Perl value to/from C using the C type to the
1531       left of "&", but provide a pointer to this value when the C function is
1532       called.
1533
1534       This is useful to avoid a CODE: block for a C function which takes a
1535       parameter by reference.  Typically, the parameter should be not a
1536       pointer type (an "int" or "long" but not an "int*" or "long*").
1537
1538       The following XSUB will generate incorrect C code.  The xsubpp compiler
1539       will turn this into code which calls rpcb_gettime() with parameters
1540       "(char *host, time_t timep)", but the real rpcb_gettime() wants the
1541       "timep" parameter to be of type "time_t*" rather than "time_t".
1542
1543           bool_t
1544           rpcb_gettime(host,timep)
1545                 char *host
1546                 time_t timep
1547               OUTPUT:
1548                 timep
1549
1550       That problem is corrected by using the "&" operator.  The xsubpp
1551       compiler will now turn this into code which calls rpcb_gettime()
1552       correctly with parameters "(char *host, time_t *timep)".  It does this
1553       by carrying the "&" through, so the function call looks like
1554       "rpcb_gettime(host, &timep)".
1555
1556           bool_t
1557           rpcb_gettime(host,timep)
1558                 char *host
1559                 time_t &timep
1560               OUTPUT:
1561                 timep
1562
1563   Inserting POD, Comments and C Preprocessor Directives
1564       C preprocessor directives are allowed within BOOT:, PREINIT: INIT:,
1565       CODE:, PPCODE:, POSTCALL:, and CLEANUP: blocks, as well as outside the
1566       functions.  Comments are allowed anywhere after the MODULE keyword.
1567       The compiler will pass the preprocessor directives through untouched
1568       and will remove the commented lines. POD documentation is allowed at
1569       any point, both in the C and XS language sections. POD must be
1570       terminated with a "=cut" command; "xsubpp" will exit with an error if
1571       it does not. It is very unlikely that human generated C code will be
1572       mistaken for POD, as most indenting styles result in whitespace in
1573       front of any line starting with "=". Machine generated XS files may
1574       fall into this trap unless care is taken to ensure that a space breaks
1575       the sequence "\n=".
1576
1577       Comments can be added to XSUBs by placing a "#" as the first non-
1578       whitespace of a line.  Care should be taken to avoid making the comment
1579       look like a C preprocessor directive, lest it be interpreted as such.
1580       The simplest way to prevent this is to put whitespace in front of the
1581       "#".
1582
1583       If you use preprocessor directives to choose one of two versions of a
1584       function, use
1585
1586           #if ... version1
1587           #else /* ... version2  */
1588           #endif
1589
1590       and not
1591
1592           #if ... version1
1593           #endif
1594           #if ... version2
1595           #endif
1596
1597       because otherwise xsubpp will believe that you made a duplicate
1598       definition of the function.  Also, put a blank line before the
1599       #else/#endif so it will not be seen as part of the function body.
1600
1601   Using XS With C++
1602       If an XSUB name contains "::", it is considered to be a C++ method.
1603       The generated Perl function will assume that its first argument is an
1604       object pointer.  The object pointer will be stored in a variable called
1605       THIS.  The object should have been created by C++ with the new()
1606       function and should be blessed by Perl with the sv_setref_pv() macro.
1607       The blessing of the object by Perl can be handled by a typemap.  An
1608       example typemap is shown at the end of this section.
1609
1610       If the return type of the XSUB includes "static", the method is
1611       considered to be a static method.  It will call the C++ function using
1612       the class::method() syntax.  If the method is not static the function
1613       will be called using the THIS->method() syntax.
1614
1615       The next examples will use the following C++ class.
1616
1617            class color {
1618                 public:
1619                 color();
1620                 ~color();
1621                 int blue();
1622                 void set_blue( int );
1623
1624                 private:
1625                 int c_blue;
1626            };
1627
1628       The XSUBs for the blue() and set_blue() methods are defined with the
1629       class name but the parameter for the object (THIS, or "self") is
1630       implicit and is not listed.
1631
1632            int
1633            color::blue()
1634
1635            void
1636            color::set_blue( val )
1637                 int val
1638
1639       Both Perl functions will expect an object as the first parameter.  In
1640       the generated C++ code the object is called "THIS", and the method call
1641       will be performed on this object.  So in the C++ code the blue() and
1642       set_blue() methods will be called as this:
1643
1644            RETVAL = THIS->blue();
1645
1646            THIS->set_blue( val );
1647
1648       You could also write a single get/set method using an optional
1649       argument:
1650
1651            int
1652            color::blue( val = NO_INIT )
1653                int val
1654                PROTOTYPE $;$
1655                CODE:
1656                    if (items > 1)
1657                        THIS->set_blue( val );
1658                    RETVAL = THIS->blue();
1659                OUTPUT:
1660                    RETVAL
1661
1662       If the function's name is DESTROY then the C++ "delete" function will
1663       be called and "THIS" will be given as its parameter.  The generated C++
1664       code for
1665
1666            void
1667            color::DESTROY()
1668
1669       will look like this:
1670
1671            color *THIS = ...;  // Initialized as in typemap
1672
1673            delete THIS;
1674
1675       If the function's name is new then the C++ "new" function will be
1676       called to create a dynamic C++ object.  The XSUB will expect the class
1677       name, which will be kept in a variable called "CLASS", to be given as
1678       the first argument.
1679
1680            color *
1681            color::new()
1682
1683       The generated C++ code will call "new".
1684
1685            RETVAL = new color();
1686
1687       The following is an example of a typemap that could be used for this
1688       C++ example.
1689
1690           TYPEMAP
1691           color *  O_OBJECT
1692
1693           OUTPUT
1694           # The Perl object is blessed into 'CLASS', which should be a
1695           # char* having the name of the package for the blessing.
1696           O_OBJECT
1697               sv_setref_pv( $arg, CLASS, (void*)$var );
1698
1699           INPUT
1700           O_OBJECT
1701               if( sv_isobject($arg) && (SvTYPE(SvRV($arg)) == SVt_PVMG) )
1702                   $var = ($type)SvIV((SV*)SvRV( $arg ));
1703               else{
1704                   warn(\"${Package}::$func_name() -- \"
1705                       \"$var is not a blessed SV reference\");
1706                   XSRETURN_UNDEF;
1707               }
1708
1709   Interface Strategy
1710       When designing an interface between Perl and a C library a straight
1711       translation from C to XS (such as created by "h2xs -x") is often
1712       sufficient.  However, sometimes the interface will look very C-like and
1713       occasionally nonintuitive, especially when the C function modifies one
1714       of its parameters, or returns failure inband (as in "negative return
1715       values mean failure").  In cases where the programmer wishes to create
1716       a more Perl-like interface the following strategy may help to identify
1717       the more critical parts of the interface.
1718
1719       Identify the C functions with input/output or output parameters.  The
1720       XSUBs for these functions may be able to return lists to Perl.
1721
1722       Identify the C functions which use some inband info as an indication of
1723       failure.  They may be candidates to return undef or an empty list in
1724       case of failure.  If the failure may be detected without a call to the
1725       C function, you may want to use an INIT: section to report the failure.
1726       For failures detectable after the C function returns one may want to
1727       use a POSTCALL: section to process the failure.  In more complicated
1728       cases use CODE: or PPCODE: sections.
1729
1730       If many functions use the same failure indication based on the return
1731       value, you may want to create a special typedef to handle this
1732       situation.  Put
1733
1734         typedef int negative_is_failure;
1735
1736       near the beginning of XS file, and create an OUTPUT typemap entry for
1737       "negative_is_failure" which converts negative values to "undef", or
1738       maybe croak()s.  After this the return value of type
1739       "negative_is_failure" will create more Perl-like interface.
1740
1741       Identify which values are used by only the C and XSUB functions
1742       themselves, say, when a parameter to a function should be a contents of
1743       a global variable.  If Perl does not need to access the contents of the
1744       value then it may not be necessary to provide a translation for that
1745       value from C to Perl.
1746
1747       Identify the pointers in the C function parameter lists and return
1748       values.  Some pointers may be used to implement input/output or output
1749       parameters, they can be handled in XS with the "&" unary operator, and,
1750       possibly, using the NO_INIT keyword.  Some others will require handling
1751       of types like "int *", and one needs to decide what a useful Perl
1752       translation will do in such a case.  When the semantic is clear, it is
1753       advisable to put the translation into a typemap file.
1754
1755       Identify the structures used by the C functions.  In many cases it may
1756       be helpful to use the T_PTROBJ typemap for these structures so they can
1757       be manipulated by Perl as blessed objects.  (This is handled
1758       automatically by "h2xs -x".)
1759
1760       If the same C type is used in several different contexts which require
1761       different translations, "typedef" several new types mapped to this C
1762       type, and create separate typemap entries for these new types.  Use
1763       these types in declarations of return type and parameters to XSUBs.
1764
1765   Perl Objects And C Structures
1766       When dealing with C structures one should select either T_PTROBJ or
1767       T_PTRREF for the XS type.  Both types are designed to handle pointers
1768       to complex objects.  The T_PTRREF type will allow the Perl object to be
1769       unblessed while the T_PTROBJ type requires that the object be blessed.
1770       By using T_PTROBJ one can achieve a form of type-checking because the
1771       XSUB will attempt to verify that the Perl object is of the expected
1772       type.
1773
1774       The following XS code shows the getnetconfigent() function which is
1775       used with ONC+ TIRPC.  The getnetconfigent() function will return a
1776       pointer to a C structure and has the C prototype shown below.  The
1777       example will demonstrate how the C pointer will become a Perl
1778       reference.  Perl will consider this reference to be a pointer to a
1779       blessed object and will attempt to call a destructor for the object.  A
1780       destructor will be provided in the XS source to free the memory used by
1781       getnetconfigent().  Destructors in XS can be created by specifying an
1782       XSUB function whose name ends with the word DESTROY.  XS destructors
1783       can be used to free memory which may have been malloc'd by another
1784       XSUB.
1785
1786            struct netconfig *getnetconfigent(const char *netid);
1787
1788       A "typedef" will be created for "struct netconfig".  The Perl object
1789       will be blessed in a class matching the name of the C type, with the
1790       tag "Ptr" appended, and the name should not have embedded spaces if it
1791       will be a Perl package name.  The destructor will be placed in a class
1792       corresponding to the class of the object and the PREFIX keyword will be
1793       used to trim the name to the word DESTROY as Perl will expect.
1794
1795            typedef struct netconfig Netconfig;
1796
1797            MODULE = RPC  PACKAGE = RPC
1798
1799            Netconfig *
1800            getnetconfigent(netid)
1801                 char *netid
1802
1803            MODULE = RPC  PACKAGE = NetconfigPtr  PREFIX = rpcb_
1804
1805            void
1806            rpcb_DESTROY(netconf)
1807                 Netconfig *netconf
1808               CODE:
1809                 printf("Now in NetconfigPtr::DESTROY\n");
1810                 free( netconf );
1811
1812       This example requires the following typemap entry.  Consult
1813       perlxstypemap for more information about adding new typemaps for an
1814       extension.
1815
1816            TYPEMAP
1817            Netconfig *  T_PTROBJ
1818
1819       This example will be used with the following Perl statements.
1820
1821            use RPC;
1822            $netconf = getnetconfigent("udp");
1823
1824       When Perl destroys the object referenced by $netconf it will send the
1825       object to the supplied XSUB DESTROY function.  Perl cannot determine,
1826       and does not care, that this object is a C struct and not a Perl
1827       object.  In this sense, there is no difference between the object
1828       created by the getnetconfigent() XSUB and an object created by a normal
1829       Perl subroutine.
1830
1831   Safely Storing Static Data in XS
1832       Starting with Perl 5.8, a macro framework has been defined to allow
1833       static data to be safely stored in XS modules that will be accessed
1834       from a multi-threaded Perl.
1835
1836       Although primarily designed for use with multi-threaded Perl, the
1837       macros have been designed so that they will work with non-threaded Perl
1838       as well.
1839
1840       It is therefore strongly recommended that these macros be used by all
1841       XS modules that make use of static data.
1842
1843       The easiest way to get a template set of macros to use is by specifying
1844       the "-g" ("--global") option with h2xs (see h2xs).
1845
1846       Below is an example module that makes use of the macros.
1847
1848           #define PERL_NO_GET_CONTEXT
1849           #include "EXTERN.h"
1850           #include "perl.h"
1851           #include "XSUB.h"
1852
1853           /* Global Data */
1854
1855           #define MY_CXT_KEY "BlindMice::_guts" XS_VERSION
1856
1857           typedef struct {
1858               int count;
1859               char name[3][100];
1860           } my_cxt_t;
1861
1862           START_MY_CXT
1863
1864           MODULE = BlindMice           PACKAGE = BlindMice
1865
1866           BOOT:
1867           {
1868               MY_CXT_INIT;
1869               MY_CXT.count = 0;
1870               strcpy(MY_CXT.name[0], "None");
1871               strcpy(MY_CXT.name[1], "None");
1872               strcpy(MY_CXT.name[2], "None");
1873           }
1874
1875           int
1876           newMouse(char * name)
1877               PREINIT:
1878                 dMY_CXT;
1879               CODE:
1880                 if (MY_CXT.count >= 3) {
1881                     warn("Already have 3 blind mice");
1882                     RETVAL = 0;
1883                 }
1884                 else {
1885                     RETVAL = ++ MY_CXT.count;
1886                     strcpy(MY_CXT.name[MY_CXT.count - 1], name);
1887                 }
1888               OUTPUT:
1889                 RETVAL
1890
1891           char *
1892           get_mouse_name(index)
1893                 int index
1894               PREINIT:
1895                 dMY_CXT;
1896               CODE:
1897                 if (index > MY_CXT.count)
1898                   croak("There are only 3 blind mice.");
1899                 else
1900                   RETVAL = MY_CXT.name[index - 1];
1901               OUTPUT:
1902                 RETVAL
1903
1904           void
1905           CLONE(...)
1906               CODE:
1907                 MY_CXT_CLONE;
1908
1909       MY_CXT REFERENCE
1910
1911       MY_CXT_KEY
1912            This macro is used to define a unique key to refer to the static
1913            data for an XS module. The suggested naming scheme, as used by
1914            h2xs, is to use a string that consists of the module name, the
1915            string "::_guts" and the module version number.
1916
1917                #define MY_CXT_KEY "MyModule::_guts" XS_VERSION
1918
1919       typedef my_cxt_t
1920            This struct typedef must always be called "my_cxt_t". The other
1921            "CXT*" macros assume the existence of the "my_cxt_t" typedef name.
1922
1923            Declare a typedef named "my_cxt_t" that is a structure that
1924            contains all the data that needs to be interpreter-local.
1925
1926                typedef struct {
1927                    int some_value;
1928                } my_cxt_t;
1929
1930       START_MY_CXT
1931            Always place the START_MY_CXT macro directly after the declaration
1932            of "my_cxt_t".
1933
1934       MY_CXT_INIT
1935            The MY_CXT_INIT macro initializes storage for the "my_cxt_t"
1936            struct.
1937
1938            It must be called exactly once, typically in a BOOT: section. If
1939            you are maintaining multiple interpreters, it should be called
1940            once in each interpreter instance, except for interpreters cloned
1941            from existing ones.  (But see "MY_CXT_CLONE" below.)
1942
1943       dMY_CXT
1944            Use the dMY_CXT macro (a declaration) in all the functions that
1945            access MY_CXT.
1946
1947       MY_CXT
1948            Use the MY_CXT macro to access members of the "my_cxt_t" struct.
1949            For example, if "my_cxt_t" is
1950
1951                typedef struct {
1952                    int index;
1953                } my_cxt_t;
1954
1955            then use this to access the "index" member
1956
1957                dMY_CXT;
1958                MY_CXT.index = 2;
1959
1960       aMY_CXT/pMY_CXT
1961            "dMY_CXT" may be quite expensive to calculate, and to avoid the
1962            overhead of invoking it in each function it is possible to pass
1963            the declaration onto other functions using the "aMY_CXT"/"pMY_CXT"
1964            macros, eg
1965
1966                void sub1() {
1967                    dMY_CXT;
1968                    MY_CXT.index = 1;
1969                    sub2(aMY_CXT);
1970                }
1971
1972                void sub2(pMY_CXT) {
1973                    MY_CXT.index = 2;
1974                }
1975
1976            Analogously to "pTHX", there are equivalent forms for when the
1977            macro is the first or last in multiple arguments, where an
1978            underscore represents a comma, i.e.  "_aMY_CXT", "aMY_CXT_",
1979            "_pMY_CXT" and "pMY_CXT_".
1980
1981       MY_CXT_CLONE
1982            By default, when a new interpreter is created as a copy of an
1983            existing one (eg via "threads->create()"), both interpreters share
1984            the same physical my_cxt_t structure. Calling "MY_CXT_CLONE"
1985            (typically via the package's CLONE() function), causes a byte-for-
1986            byte copy of the structure to be taken, and any future dMY_CXT
1987            will cause the copy to be accessed instead.
1988
1989       MY_CXT_INIT_INTERP(my_perl)
1990       dMY_CXT_INTERP(my_perl)
1991            These are versions of the macros which take an explicit
1992            interpreter as an argument.
1993
1994       Note that these macros will only work together within the same source
1995       file; that is, a dMY_CTX in one source file will access a different
1996       structure than a dMY_CTX in another source file.
1997
1998   Thread-aware system interfaces
1999       Starting from Perl 5.8, in C/C++ level Perl knows how to wrap
2000       system/library interfaces that have thread-aware versions (e.g.
2001       getpwent_r()) into frontend macros (e.g. getpwent()) that correctly
2002       handle the multithreaded interaction with the Perl interpreter.  This
2003       will happen transparently, the only thing you need to do is to
2004       instantiate a Perl interpreter.
2005
2006       This wrapping happens always when compiling Perl core source (PERL_CORE
2007       is defined) or the Perl core extensions (PERL_EXT is defined).  When
2008       compiling XS code outside of the Perl core, the wrapping does not take
2009       place before Perl 5.28.  Starting in that release you can
2010
2011        #define PERL_REENTRANT
2012
2013       in your code to enable the wrapping.  It is advisable to do so if you
2014       are using such functions, as intermixing the "_r"-forms (as Perl
2015       compiled for multithreaded operation will do) and the "_r"-less forms
2016       is neither well-defined (inconsistent results, data corruption, or even
2017       crashes become more likely), nor is it very portable.  Unfortunately,
2018       not all systems have all the "_r" forms, but using this "#define" gives
2019       you whatever protection that Perl is aware is available on each system.
2020

EXAMPLES

2022       File "RPC.xs": Interface to some ONC+ RPC bind library functions.
2023
2024            #define PERL_NO_GET_CONTEXT
2025            #include "EXTERN.h"
2026            #include "perl.h"
2027            #include "XSUB.h"
2028
2029            /* Note: On glibc 2.13 and earlier, this needs be <rpc/rpc.h> */
2030            #include <tirpc/rpc.h>
2031
2032            typedef struct netconfig Netconfig;
2033
2034            MODULE = RPC  PACKAGE = RPC
2035
2036            SV *
2037            rpcb_gettime(host="localhost")
2038                 char *host
2039               PREINIT:
2040                 time_t  timep;
2041               CODE:
2042                 ST(0) = sv_newmortal();
2043                 if( rpcb_gettime( host, &timep ) )
2044                      sv_setnv( ST(0), (double)timep );
2045
2046            Netconfig *
2047            getnetconfigent(netid="udp")
2048                 char *netid
2049
2050            MODULE = RPC  PACKAGE = NetconfigPtr  PREFIX = rpcb_
2051
2052            void
2053            rpcb_DESTROY(netconf)
2054                 Netconfig *netconf
2055               CODE:
2056                 printf("NetconfigPtr::DESTROY\n");
2057                 free( netconf );
2058
2059       File "typemap": Custom typemap for RPC.xs. (cf. perlxstypemap)
2060
2061            TYPEMAP
2062            Netconfig *  T_PTROBJ
2063
2064       File "RPC.pm": Perl module for the RPC extension.
2065
2066            package RPC;
2067
2068            require Exporter;
2069            require DynaLoader;
2070            @ISA = qw(Exporter DynaLoader);
2071            @EXPORT = qw(rpcb_gettime getnetconfigent);
2072
2073            bootstrap RPC;
2074            1;
2075
2076       File "rpctest.pl": Perl test program for the RPC extension.
2077
2078            use RPC;
2079
2080            $netconf = getnetconfigent();
2081            $a = rpcb_gettime();
2082            print "time = $a\n";
2083            print "netconf = $netconf\n";
2084
2085            $netconf = getnetconfigent("tcp");
2086            $a = rpcb_gettime("poplar");
2087            print "time = $a\n";
2088            print "netconf = $netconf\n";
2089
2090       In Makefile.PL add -ltirpc and -I/usr/include/tirpc.
2091

CAVEATS

2093       XS code has full access to system calls including C library functions.
2094       It thus has the capability of interfering with things that the Perl
2095       core or other modules have set up, such as signal handlers or file
2096       handles.  It could mess with the memory, or any number of harmful
2097       things.  Don't.
2098
2099       Some modules have an event loop, waiting for user-input.  It is highly
2100       unlikely that two such modules would work adequately together in a
2101       single Perl application.
2102
2103       In general, the perl interpreter views itself as the center of the
2104       universe as far as the Perl program goes.  XS code is viewed as a help-
2105       mate, to accomplish things that perl doesn't do, or doesn't do fast
2106       enough, but always subservient to perl.  The closer XS code adheres to
2107       this model, the less likely conflicts will occur.
2108
2109       One area where there has been conflict is in regards to C locales.
2110       (See perllocale.)  perl, with one exception and unless told otherwise,
2111       sets up the underlying locale the program is running in to the locale
2112       passed into it from the environment.  This is an important difference
2113       from a generic C language program, where the underlying locale is the
2114       "C" locale unless the program changes it.  As of v5.20, this underlying
2115       locale is completely hidden from pure Perl code outside the lexical
2116       scope of "use locale" except for a couple of function calls in the
2117       POSIX module which of necessity use it.  But the underlying locale,
2118       with that one exception is exposed to XS code, affecting all C library
2119       routines whose behavior is locale-dependent.  Your XS code better not
2120       assume that the underlying locale is "C".  The exception is the
2121       "LC_NUMERIC" locale category, and the reason it is an exception is that
2122       experience has shown that it can be problematic for XS code, whereas we
2123       have not had reports of problems with the other locale categories.  And
2124       the reason for this one category being problematic is that the
2125       character used as a decimal point can vary.  Many European languages
2126       use a comma, whereas English, and hence Perl are expecting a dot
2127       (U+002E: FULL STOP).  Many modules can handle only the radix character
2128       being a dot, and so perl attempts to make it so.  Up through Perl
2129       v5.20, the attempt was merely to set "LC_NUMERIC" upon startup to the
2130       "C" locale.  Any setlocale() otherwise would change it; this caused
2131       some failures.  Therefore, starting in v5.22, perl tries to keep
2132       "LC_NUMERIC" always set to "C" for XS code.
2133
2134       To summarize, here's what to expect and how to handle locales in XS
2135       code:
2136
2137       Non-locale-aware XS code
2138           Keep in mind that even if you think your code is not locale-aware,
2139           it may call a library function that is.  Hopefully the man page for
2140           such a function will indicate that dependency, but the
2141           documentation is imperfect.
2142
2143           The current locale is exposed to XS code except possibly
2144           "LC_NUMERIC" (explained in the next paragraph).  There have not
2145           been reports of problems with the other categories.  Perl
2146           initializes things on start-up so that the current locale is the
2147           one which is indicated by the user's environment in effect at that
2148           time.  See "ENVIRONMENT" in perllocale.
2149
2150           However, up through v5.20, Perl initialized things on start-up so
2151           that "LC_NUMERIC" was set to the "C" locale.  But if any code
2152           anywhere changed it, it would stay changed.  This means that your
2153           module can't count on "LC_NUMERIC" being something in particular,
2154           and you can't expect floating point numbers (including version
2155           strings) to have dots in them.  If you don't allow for a non-dot,
2156           your code could break if anyone anywhere changed the locale.  For
2157           this reason, v5.22 changed the behavior so that Perl tries to keep
2158           "LC_NUMERIC" in the "C" locale except around the operations
2159           internally where it should be something else.  Misbehaving XS code
2160           will always be able to change the locale anyway, but the most
2161           common instance of this is checked for and handled.
2162
2163       Locale-aware XS code
2164           If the locale from the user's environment is desired, there should
2165           be no need for XS code to set the locale except for "LC_NUMERIC",
2166           as perl has already set the others up.  XS code should avoid
2167           changing the locale, as it can adversely affect other, unrelated,
2168           code and may not be thread-safe.  To minimize problems, the macros
2169           "STORE_LC_NUMERIC_SET_TO_NEEDED" in perlapi,
2170           "STORE_LC_NUMERIC_FORCE_TO_UNDERLYING" in perlapi, and
2171           "RESTORE_LC_NUMERIC" in perlapi should be used to affect any needed
2172           change.
2173
2174           But, starting with Perl v5.28, locales are thread-safe on platforms
2175           that support this functionality.  Windows has this starting with
2176           Visual Studio 2005.  Many other modern platforms support the
2177           thread-safe POSIX 2008 functions.  The C "#define"
2178           "USE_THREAD_SAFE_LOCALE" will be defined iff this build is using
2179           these.  From Perl-space, the read-only variable "${SAFE_LOCALES}"
2180           is 1 if either the build is not threaded, or if
2181           "USE_THREAD_SAFE_LOCALE" is defined; otherwise it is 0.
2182
2183           The way this works under-the-hood is that every thread has a choice
2184           of using a locale specific to it (this is the Windows and POSIX
2185           2008 functionality), or the global locale that is accessible to all
2186           threads (this is the functionality that has always been there).
2187           The implementations for Windows and POSIX are completely different.
2188           On Windows, the runtime can be set up so that the standard
2189           setlocale(3) function either only knows about the global locale or
2190           the locale for this thread.  On POSIX, "setlocale" always deals
2191           with the global locale, and other functions have been created to
2192           handle per-thread locales.  Perl makes this transparent to perl-
2193           space code.  It continues to use POSIX::setlocale(), and the
2194           interpreter translates that into the per-thread functions.
2195
2196           All other locale-sensitive functions automatically use the per-
2197           thread locale, if that is turned on, and failing that, the global
2198           locale.  Thus calls to "setlocale" are ineffective on POSIX systems
2199           for the current thread if that thread is using a per-thread locale.
2200           If perl is compiled for single-thread operation, it does not use
2201           the per-thread functions, so "setlocale" does work as expected.
2202
2203           If you have loaded the "POSIX" module you can use the methods given
2204           in perlcall to call "POSIX::setlocale" to safely change or query
2205           the locale (on systems where it is safe to do so), or you can use
2206           the new 5.28 function "Perl_setlocale" in perlapi instead, which is
2207           a drop-in replacement for the system setlocale(3), and handles
2208           single-threaded and multi-threaded applications transparently.
2209
2210           There are some locale-related library calls that still aren't
2211           thread-safe because they return data in a buffer global to all
2212           threads.  In the past, these didn't matter as locales weren't
2213           thread-safe at all.  But now you have to be aware of them in case
2214           your module is called in a multi-threaded application.  The known
2215           ones are
2216
2217            asctime()
2218            ctime()
2219            gcvt() [POSIX.1-2001 only (function removed in POSIX.1-2008)]
2220            getdate()
2221            wcrtomb() if its final argument is NULL
2222            wcsrtombs() if its final argument is NULL
2223            wcstombs()
2224            wctomb()
2225
2226           Some of these shouldn't really be called in a Perl application, and
2227           for others there are thread-safe versions of these already
2228           implemented:
2229
2230            asctime_r()
2231            ctime_r()
2232            Perl_langinfo()
2233
2234           The "_r" forms are automatically used, starting in Perl 5.28, if
2235           you compile your code, with
2236
2237            #define PERL_REENTRANT
2238
2239           See also "Perl_langinfo" in perlapi.  You can use the methods given
2240           in perlcall, to get the best available locale-safe versions of
2241           these
2242
2243            POSIX::localeconv()
2244            POSIX::wcstombs()
2245            POSIX::wctomb()
2246
2247           And note, that some items returned by "Localeconv" are available
2248           through "Perl_langinfo" in perlapi.
2249
2250           The others shouldn't be used in a threaded application.
2251
2252           Some modules may call a non-perl library that is locale-aware.
2253           This is fine as long as it doesn't try to query or change the
2254           locale using the system "setlocale".  But if these do call the
2255           system "setlocale", those calls may be ineffective.  Instead,
2256           "Perl_setlocale" works in all circumstances.  Plain setlocale is
2257           ineffective on multi-threaded POSIX 2008 systems.  It operates only
2258           on the global locale, whereas each thread has its own locale,
2259           paying no attention to the global one.  Since converting these non-
2260           Perl libraries to "Perl_setlocale" is out of the question, there is
2261           a new function in v5.28 "switch_to_global_locale" that will switch
2262           the thread it is called from so that any system "setlocale" calls
2263           will have their desired effect.  The function "sync_locale" must be
2264           called before returning to perl.
2265
2266           This thread can change the locale all it wants and it won't affect
2267           any other thread, except any that also have been switched to the
2268           global locale.  This means that a multi-threaded application can
2269           have a single thread using an alien library without a problem; but
2270           no more than a single thread can be so-occupied.  Bad results
2271           likely will happen.
2272
2273           In perls without multi-thread locale support, some alien libraries,
2274           such as "Gtk" change locales.  This can cause problems for the Perl
2275           core and other modules.  For these, before control is returned to
2276           perl, starting in v5.20.1, calling the function sync_locale() from
2277           XS should be sufficient to avoid most of these problems.  Prior to
2278           this, you need a pure Perl statement that does this:
2279
2280            POSIX::setlocale(LC_ALL, POSIX::setlocale(LC_ALL));
2281
2282           or use the methods given in perlcall.
2283

XS VERSION

2285       This document covers features supported by "ExtUtils::ParseXS" (also
2286       known as "xsubpp") 3.13_01.
2287

AUTHOR

2289       Originally written by Dean Roehrich <roehrich@cray.com>.
2290
2291       Maintained since 1996 by The Perl Porters <perl5-porters@perl.org>.
2292
2293
2294
2295perl v5.36.0                      2023-01-20                         perlxs(3)
Impressum