1 package ExtUtils::Constant;
5 ExtUtils::Constant - generate XS code to import C header constants
9 use ExtUtils::Constant qw (constant_types C_constant XS_constant);
10 print constant_types(); # macro defs
11 foreach (C_constant ("Foo", undef, "IV", undef, undef, undef,
13 print $_, "\n"; # C constant subs
15 print "MODULE = Foo PACKAGE = Foo\n";
16 print XS_constant ("Foo", {NV => 1, IV => 1}); # XS for Foo::constant
20 ExtUtils::Constant facilitates generating C and XS wrapper code to allow
21 perl modules to AUTOLOAD constants defined in C library header files.
22 It is principally used by the C<h2xs> utility, on which this code is based.
23 It doesn't contain the routines to scan header files to extract these
28 Generally one only needs to call the 3 functions shown in the synopsis,
29 C<constant_types()>, C<C_constant> and C<XS_constant>.
31 Currently this module understands the following types. h2xs may only know
32 a subset. The sizes of the numeric types are chosen by the C<Configure>
33 script at compile time.
39 signed integer, at least 32 bits.
43 unsigned integer, the same size as I<IV>
47 floating point type, probably C<double>, possibly C<long double>
51 NUL terminated string, length will be determined with C<strlen>
55 A fixed length thing, given as a [pointer, length] pair. If you know the
56 length of a string at compile time you may use this instead of I<PV>
60 Truth. (C<PL_sv_yes>) The value is not needed (and ignored).
64 Defined Falsehood. (C<PL_sv_no>) The value is not needed (and ignored).
68 C<undef>. The value of the macro is not needed.
78 require 5.006; # I think, for [:cntrl:] in REGEXP
84 use vars qw (@ISA $VERSION %XS_Constant %XS_TypeSet @EXPORT_OK %EXPORT_TAGS);
86 $Text::Wrap::huge = 'overflow';
87 $Text::Wrap::columns = 80;
92 %EXPORT_TAGS = ( 'all' => [ qw(
93 XS_constant constant_types return_clause memEQ_clause C_stringify
97 @EXPORT_OK = ( @{ $EXPORT_TAGS{'all'} } );
101 UV => 'PUSHu((UV)iv)',
103 PV => 'PUSHp(pv, strlen(pv))',
104 PVN => 'PUSHp(pv, iv)',
105 YES => 'PUSHs(&PL_sv_yes)',
106 NO => 'PUSHs(&PL_sv_no)',
107 UNDEF => '', # implicit undef
111 IV => '*iv_return =',
112 UV => '*iv_return = (IV)',
113 NV => '*nv_return =',
114 PV => '*pv_return =',
115 PVN => ['*pv_return =', '*iv_return = (IV)'],
122 =item C_stringify NAME
124 A function which returns a correctly \ escaped version of the string passed
125 suitable for C's "" or ''. It will also be valid as a perl "" string.
129 # Hopefully make a happy C identifier.
132 return unless defined $_;
134 s/([\"\'])/\\$1/g; # Grr. fix perl mode.
135 s/\n/\\n/g; # Ensure newlines don't end up in octal
140 s/([[:cntrl:]])/sprintf "\\%03o", ord $1/ge;
141 s/\177/\\177/g; # DEL doesn't seem to be a [:cntrl:]
147 A function returning a single scalar with C<#define> definitions for the
148 constants used internally between the generated C and XS functions.
152 sub constant_types () {
155 push @lines, "#define PERL_constant_NOTFOUND\t$start\n"; $start++;
156 push @lines, "#define PERL_constant_NOTDEF\t$start\n"; $start++;
157 foreach (sort keys %XS_Constant) {
158 push @lines, "#define PERL_constant_IS$_\t$start\n"; $start++;
160 push @lines, << 'EOT';
163 typedef double NV; /* 5.6 and later define NVTYPE, and typedef NV to it. */
167 return join '', @lines;
170 =item memEQ_clause NAME, CHECKED_AT, INDENT
172 A function to return a suitable C C<if> statement to check whether I<NAME>
173 is equal to the C variable C<name>. If I<CHECKED_AT> is defined, then it
174 is used to avoid C<memEQ> for short names, or to generate a comment to
175 highlight the position of the character in the C<switch> statement.
180 # if (memEQ(name, "thingy", 6)) {
181 # Which could actually be a character comparison or even ""
182 my ($name, $checked_at, $indent) = @_;
183 $indent = ' ' x ($indent || 4);
184 my $len = length $name;
187 return $indent . "{\n" if (defined $checked_at and $checked_at == 0);
188 # We didn't switch, drop through to the code for the 2 character string
191 if ($len < 3 and defined $checked_at) {
193 if ($checked_at == 1) {
195 } elsif ($checked_at == 0) {
198 if (defined $check) {
199 my $char = C_stringify (substr $name, $check, 1);
200 return $indent . "if (name[$check] == '$char') {\n";
203 # Could optimise a memEQ on 3 to 2 single character checks here
204 $name = C_stringify ($name);
205 my $body = $indent . "if (memEQ(name, \"$name\", $len)) {\n";
206 $body .= $indent . "/* ". (' ' x $checked_at) . '^'
207 . (' ' x ($len - $checked_at + length $len)) . " */\n"
208 if defined $checked_at;
212 =item assign INDENT, TYPE, VALUE...
214 A function to return a suitable assignment clause. If I<TYPE> is aggregate
215 (eg I<PVN> expects both pointer and length) then there should be multiple
216 I<VALUE>s for the components.
220 # Hmm. value undef to to NOTDEF? value () to do NOTFOUND?
226 die "Can't generate code for type $type" unless exists $XS_TypeSet{$type};
227 my $typeset = $XS_TypeSet{$type};
229 die "Type $type is aggregate, but only single value given"
231 foreach (0 .. $#$typeset) {
232 $clause .= $indent . "$typeset->[$_] $_[$_];\n";
234 } elsif (defined $typeset) {
235 die "Aggregate value given for type $type"
237 $clause .= $indent . "$typeset $_[0];\n";
239 $clause .= "${indent}return PERL_constant_IS$type;\n";
243 =item return_clause VALUE, TYPE, INDENT, MACRO, DEFAULT
245 A function to return a suitable C<#ifdef> clause. I<MACRO> defaults to
246 I<VALUE> when not defined. If I<TYPE> is aggregate (eg I<PVN> expects both
247 pointer and length) then I<VALUE> should be a reference to an array of
248 values in the order expected by the type. C<C_constant> will always call
249 this function with I<MACRO> defined, defaulting to the constant's name.
250 I<DEFAULT> if defined is an array reference giving default type and and
251 value(s) if the clause generated by I<MACRO> doesn't evaluate to true.
255 sub return_clause ($$$$$) {
257 # *iv_return = thingy;
258 # return PERL_constant_ISIV;
260 # return PERL_constant_NOTDEF;
262 my ($value, $type, $indent, $macro, $default) = @_;
263 $macro = $value unless defined $macro;
264 $indent = ' ' x ($indent || 6);
270 $clause = $macro->[0];
272 $clause = "#ifdef $macro\n";
275 # *iv_return = thingy;
276 # return PERL_constant_ISIV;
277 $clause .= assign ($indent, $type, ref $value ? @$value : $value);
280 $clause .= "#else\n";
282 # return PERL_constant_NOTDEF;
283 if (!defined $default) {
284 $clause .= "${indent}return PERL_constant_NOTDEF;\n";
286 $clause .= assign ($indent, ref $default ? @$default : $default);
291 $clause .= $macro->[1];
293 $clause .= "#endif\n";
300 An internal function. I<WHAT> should be a hashref of types the constant
301 function will return. I<params> returns the list of flags C<$use_iv, $use_nv,
302 $use_pv> to show which combination of pointers will be needed in the C
309 foreach (sort keys %$what) {
310 warn "ExtUtils::Constant doesn't know how to handle values of type $_" unless defined $XS_Constant{$_};
312 my $use_iv = $what->{IV} || $what->{UV} || $what->{PVN};
313 my $use_nv = $what->{NV};
314 my $use_pv = $what->{PV} || $what->{PVN};
315 return ($use_iv, $use_nv, $use_pv);
320 dump_names PACKAGE, SUBNAME, DEFAULT_TYPE, TYPES, INDENT, ITEM...
322 An internal function to generate the embedded perl code that will regenerate
323 the constant subroutines. Parameters are the same as for C_constant, except
324 that there is no NAMELEN.
329 my ($package, $subname, $default_type, $what, $indent, @items) = @_;
330 my (@simple, @complex);
332 my $type = $_->{type} || $default_type;
333 if ($type eq $default_type and 0 == ($_->{name} =~ tr/A-Za-z0-9_//c)
334 and !defined ($_->{macro}) and !defined ($_->{value})
335 and !defined ($_->{default})) {
336 # It's the default type, and the name consists only of A-Za-z0-9_
337 push @simple, $_->{name};
342 my $result = <<"EOT";
343 /* When generated this function returned values for the list of names given
344 in this section of perl code. Rather than manually editing these functions
345 to add or remove constants, which would result in this comment and section
346 of code becoming inaccurate, we recommend that you edit this section of
347 code, and use it to regenerate a new set of constant functions which you
348 then use to replace the originals.
350 Regenerate these constant functions by feeding this entire source file to
354 use ExtUtils::Constant qw (constant_types C_constant XS_constant);
357 $result .= 'my $types = {' . join (", ", map "$_ => 1", sort keys %$what)
359 $result .= wrap ("my \@names = (qw(",
360 " ", join (" ", sort @simple) . ")");
362 foreach my $item (sort {$a->{name} cmp $b->{name}} @complex) {
363 my $name = C_stringify $item->{name};
364 my ($macro, $value, $default) = @$item{qw (macro value default)};
365 my $line = ",\n {name=>\"$name\"";
366 $line .= ", type=>\"$item->{type}\"" if defined $item->{type};
367 if (defined $macro) {
369 $line .= ', macro=>["'. join ('", "', map {C_stringify $_} @$macro)
372 $line .= ", macro=>\"" . C_stringify($macro) . "\"";
375 if (defined $value) {
377 $line .= ', value=>["'. join ('", "', map {C_stringify $_} @$value)
380 $line .= ", value=>\"" . C_stringify($value) . "\"";
383 if (defined $default) {
385 $line .= ', default=>["'. join ('", "', map {C_stringify $_}
389 $line .= ", default=>\"" . C_stringify($default) . "\"";
393 # Ensure that the enclosing C comment doesn't end
394 # by turning */ into *" . "/
395 $line =~ s!\*\/!\*" . "/!gs;
396 # gcc -Wall doesn't like finding /* inside a comment
397 $line =~ s!\/\*!/" . "\*!gs;
405 print constant_types(); # macro defs
407 $package = C_stringify($package);
409 "foreach (C_constant (\"$package\", '$subname', '$default_type', \$types, ";
410 # The form of the indent parameter isn't defined. (Yet)
411 if (defined $indent) {
412 require Data::Dumper;
413 $Data::Dumper::Terse=1;
414 chomp ($indent = Data::Dumper::Dumper ($indent));
419 $result .= ', undef, @names) ) {
420 print $_, "\n"; # C constant subs
422 print "#### XS Section:\n";
423 print XS_constant ("' . $package . '", $types);
434 C_constant PACKAGE, SUBNAME, DEFAULT_TYPE, TYPES, INDENT, NAMELEN, ITEM...
436 A function that returns a B<list> of C subroutine definitions that return
437 the value and type of constants when passed the name by the XS wrapper.
438 I<ITEM...> gives a list of constant names. Each can either be a string,
439 which is taken as a C macro name, or a reference to a hash with the following
446 The name of the constant, as seen by the perl code.
450 The type of the constant (I<IV>, I<NV> etc)
454 A C expression for the value of the constant, or a list of C expressions if
455 the type is aggregate. This defaults to the I<name> if not given.
459 The C pre-processor macro to use in the C<#ifdef>. This defaults to the
460 I<name>, and is mainly used if I<value> is an C<enum>. If a reference an
461 array is passed then the first element is used in place of the C<#ifdef>
462 line, and the second element in place of the C<#endif>. This allows
463 pre-processor constructions such as
471 to be used to determine if a constant is to be defined.
475 Default value to use (instead of C<croak>ing with "your vendor has not
476 defined...") to return if the macro isn't defined. Specify a reference to
477 an array with type followed by value(s).
481 I<PACKAGE> is the name of the package, and is only used in comments inside the
484 The next 5 arguments can safely be given as C<undef>, and are mainly used
485 for recursion. I<SUBNAME> defaults to C<constant> if undefined.
487 I<DEFAULT_TYPE> is the type returned by C<ITEM>s that don't specify their
488 type. In turn it defaults to I<IV>. I<TYPES> should be given either as a comma
489 separated list of types that the C subroutine C<constant> will generate or as
490 a reference to a hash. I<DEFAULT_TYPE> will be added to the list if not
491 present, as will any types given in the list of I<ITEM>s. The resultant list
492 should be the same list of types that C<XS_constant> is given. [Otherwise
493 C<XS_constant> and C<C_constant> may differ in the number of parameters to the
494 constant function. I<INDENT> is currently unused and ignored. In future it may
495 be used to pass in information used to change the C indentation style used.]
496 The best way to maintain consistency is to pass in a hash reference and let
497 this function update it.
499 I<NAMELEN> if defined signals that all the I<name>s of the I<ITEM>s are of
500 this length, and that the constant name passed in by perl is checked and
501 also of this length. It is used during recursion, and should be C<undef>
502 unless the caller has checked all the lengths during code generation, and
503 the generated subroutine is only to be called with a name of this length.
508 my ($package, $subname, $default_type, $what, $indent, $namelen, @items) = @_;
510 $subname ||= 'constant';
511 # I'm not using this. But a hashref could be used for full formatting without
514 $default_type ||= 'IV';
516 # Convert line of the form IV,UV,NV to hash
517 $what = {map {$_ => 1} split /,\s*/, ($what || '')};
518 # Figure out what types we're dealing with, and assign all unknowns to the
525 # Make a copy which is a normalised version of the ref passed in.
527 my ($type, $macro, $value, $default) = @$_{qw (type macro value default)};
528 $type ||= $default_type;
530 $_ = {name=>$name, type=>$type};
532 undef $macro if defined $macro and $macro eq $name;
533 $_->{macro} = $macro if defined $macro;
534 undef $value if defined $value and $value eq $name;
535 $_->{value} = $value if defined $value;
536 $_->{default} = $default if defined $default;
539 $_ = {name=>$_, type=>$default_type};
540 $what->{$default_type} = 1;
542 warn "ExtUtils::Constant doesn't know how to handle values of type $_ used in macro $name" unless defined $XS_Constant{$_->{type}};
543 if (exists $items{$name}) {
544 die "Multiple definitions for macro $name";
548 my ($use_iv, $use_nv, $use_pv) = params ($what);
550 my ($body, @subs) = "static int\n$subname (const char *name";
551 $body .= ", STRLEN len" unless defined $namelen;
552 $body .= ", IV *iv_return" if $use_iv;
553 $body .= ", NV *nv_return" if $use_nv;
554 $body .= ", const char **pv_return" if $use_pv;
557 if (defined $namelen) {
558 # We are a child subroutine. Print the simple description
559 my @names = sort map {$_->{name}} @items;
561 /* When generated this function returned values for the list of names given
562 here. However, subsequent manual editing may have added or removed some.
564 . wrap (" ", " ", join (" ", @names) . " */") . "\n";
565 # Figure out what to switch on.
566 # (RMS, Spread of jump table, Position, Hashref)
567 my @best = (1e38, ~0);
568 foreach my $i (0 .. ($namelen - 1)) {
569 my ($min, $max) = (~0, 0);
572 my $char = substr $_, $i, 1;
574 $max = $ord if $ord > $max;
575 $min = $ord if $ord < $min;
576 push @{$spread{$char}}, $_;
579 # I'm going to pick the character to split on that minimises the root
580 # mean square of the number of names in each case. Normally this should
581 # be the one with the most keys, but it may pick a 7 where the 8 has
582 # one long linear search. I'm not sure if RMS or just sum of squares is
584 # $max and $min are for the tie-breaker if the root mean squares match.
585 # Assuming that the compiler may be building a jump table for the
586 # switch() then try to minimise the size of that jump table.
587 # Finally use < not <= so that if it still ties the earliest part of
588 # the string wins. Because if that passes but the memEQ fails, it may
589 # only need the start of the string to bin the choice.
590 # I think. But I'm micro-optimising. :-)
592 $ss += @$_ * @$_ foreach values %spread;
593 my $rms = sqrt ($ss / keys %spread);
594 if ($rms < $best[0] || ($rms == $best[0] && ($max - $min) < $best[1])) {
595 @best = ($rms, $max - $min, $i, \%spread);
598 die "Internal error. Failed to pick a switch point for @names"
599 unless defined $best[2];
600 # use Data::Dumper; print Dumper (@best);
601 my ($offset, $best) = @best[2,3];
602 $body .= " /* Names all of length $namelen. */\n";
604 $body .= " /* Offset $offset gives the best switch position. */\n";
605 $body .= " switch (name[$offset]) {\n";
606 foreach my $char (sort keys %$best) {
607 $body .= " case '" . C_stringify ($char) . "':\n";
608 foreach my $name (sort @{$best->{$char}}) {
609 my $thisone = $items{$name};
610 my ($value, $macro, $default) = @$thisone{qw (value macro default)};
611 $value = $name unless defined $value;
612 $macro = $name unless defined $macro;
614 $body .= memEQ_clause ($name, $offset); # We have checked this offset.
615 $body .= return_clause ($value, $thisone->{type}, undef, $macro,
619 $body .= " break;\n";
623 # We are the top level.
624 $body .= " /* Initially switch on the length of the name. */\n";
625 $body .= dump_names ($package, $subname, $default_type, $what, $indent,
627 $body .= " switch (len) {\n";
628 # Need to group names of the same length
631 push @{$by_length[length $_->{name}]}, $_;
633 foreach my $i (0 .. $#by_length) {
634 next unless $by_length[$i]; # None of this length
635 $body .= " case $i:\n";
636 if (@{$by_length[$i]} == 1) {
637 my $thisone = $by_length[$i]->[0];
638 my ($name, $value, $macro, $default)
639 = @$thisone{qw (name value macro default)};
640 $value = $name unless defined $value;
641 $macro = $name unless defined $macro;
643 $body .= memEQ_clause ($name);
644 $body .= return_clause ($value, $thisone->{type}, undef, $macro,
648 push @subs, C_constant ($package, "${subname}_$i", $default_type,
649 $what, $indent, $i, @{$by_length[$i]});
650 $body .= " return ${subname}_$i (name";
651 $body .= ", iv_return" if $use_iv;
652 $body .= ", nv_return" if $use_nv;
653 $body .= ", pv_return" if $use_pv;
656 $body .= " break;\n";
660 $body .= " return PERL_constant_NOTFOUND;\n}\n";
661 return (@subs, $body);
664 =item XS_constant PACKAGE, TYPES, SUBNAME, C_SUBNAME
666 A function to generate the XS code to implement the perl subroutine
667 I<PACKAGE>::constant used by I<PACKAGE>::AUTOLOAD to load constants.
668 This XS code is a wrapper around a C subroutine usually generated by
669 C<C_constant>, and usually named C<constant>.
671 I<TYPES> should be given either as a comma separated list of types that the
672 C subroutine C<constant> will generate or as a reference to a hash. It should
673 be the same list of types as C<C_constant> was given.
674 [Otherwise C<XS_constant> and C<C_constant> may have different ideas about
675 the number of parameters passed to the C function C<constant>]
677 You can call the perl visible subroutine something other than C<constant> if
678 you give the parameter I<SUBNAME>. The C subroutine it calls defaults to the
679 the name of the perl visible subroutine, unless you give the parameter
688 my $C_subname = shift;
689 $subname ||= 'constant';
690 $C_subname ||= $subname;
693 # Convert line of the form IV,UV,NV to hash
694 $what = {map {$_ => 1} split /,\s*/, ($what)};
696 my ($use_iv, $use_nv, $use_pv) = params ($what);
704 dXSTARG; /* Faster if we have it. */
715 $xs .= " /* IV\t\tiv;\tUncomment this if you need to return IVs */\n";
720 $xs .= " /* NV\t\tnv;\tUncomment this if you need to return NVs */\n";
723 $xs .= " const char *pv;\n";
726 " /* const char\t*pv;\tUncomment this if you need to return PVs */\n";
732 const char * s = SvPV(sv, len);
736 if ($use_iv xor $use_nv) {
738 /* Change this to $C_subname(s, len, &iv, &nv);
739 if you need to return both NVs and IVs */
742 $xs .= " type = $C_subname(s, len";
743 $xs .= ', &iv' if $use_iv;
744 $xs .= ', &nv' if $use_nv;
745 $xs .= ', &pv' if $use_pv;
749 /* Return 1 or 2 items. First is error message, or undef if no error.
750 Second, if present, is found value */
752 case PERL_constant_NOTFOUND:
753 sv = sv_2mortal(newSVpvf("%s is not a valid $package macro", s));
756 case PERL_constant_NOTDEF:
757 sv = sv_2mortal(newSVpvf(
758 "Your vendor has not defined $package macro %s used", s));
763 foreach $type (sort keys %XS_Constant) {
764 $xs .= "\t/* Uncomment this if you need to return ${type}s\n"
765 unless $what->{$type};
767 case PERL_constant_IS$type:
773 unless ($what->{$type}) {
774 chop $xs; # Yes, another need for chop not chomp.
780 sv = sv_2mortal(newSVpvf(
781 "Unexpected return type %d while processing $package macro %s used",
791 =item autoload PACKAGE, VERSION, AUTOLOADER
793 A function to generate the AUTOLOAD subroutine for the module I<PACKAGE>
794 I<VERSION> is the perl version the code should be backwards compatible with.
795 It defaults to the version of perl running the subroutine. If I<AUTOLOADER>
796 is true, the AUTOLOAD subroutine falls back on AutoLoader::AUTOLOAD for all
797 names that the constant() routine doesn't recognise.
801 # ' # Grr. syntax highlighters that don't grok pod.
804 my ($module, $compat_version, $autoloader) = @_;
805 $compat_version ||= $];
806 croak "Can't maintain compatibility back as far as version $compat_version"
807 if $compat_version < 5;
808 my $func = "sub AUTOLOAD {\n"
809 . " # This AUTOLOAD is used to 'autoload' constants from the constant()\n"
811 $func .= " If a constant is not found then control is passed\n"
812 . " # to the AUTOLOAD in AutoLoader." if $autoloader;
816 . " my \$constname;\n";
818 " our \$AUTOLOAD;\n" if ($compat_version >= 5.006);
821 (\$constname = \$AUTOLOAD) =~ s/.*:://;
822 croak "&${module}::constant not defined" if \$constname eq 'constant';
823 my (\$error, \$val) = constant(\$constname);
829 if ($error =~ /is not a valid/) {
830 $AutoLoader::AUTOLOAD = $AUTOLOAD;
831 goto &AutoLoader::AUTOLOAD;
839 " if (\$error) { croak \$error; }\n";
845 # Fixed between 5.005_53 and 5.005_61
846 #XXX if ($] >= 5.00561) {
847 #XXX *$AUTOLOAD = sub () { $val };
850 *$AUTOLOAD = sub { $val };
867 Nicholas Clark <nick@ccl4.org> based on the code in C<h2xs> by Larry Wall and