package ExtUtils::Constant;
use vars qw (@ISA $VERSION %XS_Constant %XS_TypeSet @EXPORT_OK %EXPORT_TAGS);
-$VERSION = '0.08';
+$VERSION = '0.14';
=head1 NAME
WriteConstants(
NAME => 'Foo',
NAMES => [qw(FOO BAR BAZ)],
- C_FILE => 'constants.c',
- XS_FILE => 'constants.xs',
);
# Generates wrapper code to make the values of the constants FOO BAR BAZ
# available to perl
Generally one only needs to call the C<WriteConstants> function, and then
- #include "constants.c"
+ #include "const-c.inc"
in the C section of C<Foo.xs>
- INCLUDE constants.xs
+ INCLUDE const-xs.inc
in the XS section of C<Foo.xs>.
A fixed length thing, given as a [pointer, length] pair. If you know the
length of a string at compile time you may use this instead of I<PV>
-=item PVN
+=item SV
A B<mortal> SV.
=cut
-require 5.006; # I think, for [:cntrl:] in REGEXP
-use warnings;
+if ($] >= 5.006) {
+ eval "use warnings; 1" or die $@;
+}
use strict;
+use vars '$is_perl56';
use Carp;
+$is_perl56 = ($] < 5.007 && $] > 5.005_50);
+
use Exporter;
use Text::Wrap;
$Text::Wrap::huge = 'overflow';
%EXPORT_TAGS = ( 'all' => [ qw(
XS_constant constant_types return_clause memEQ_clause C_stringify
- C_constant autoload WriteConstants
+ C_constant autoload WriteConstants WriteMakefileSnippet
) ] );
@EXPORT_OK = ( @{ $EXPORT_TAGS{'all'} } );
+# '' is used as a flag to indicate non-ascii macro names, and hence the need
+# to pass in the utf8 on/off flag.
%XS_Constant = (
+ '' => '',
IV => 'PUSHi(iv)',
UV => 'PUSHu((UV)iv)',
NV => 'PUSHn(nv)',
=item C_stringify NAME
-A function which returns a correctly \ escaped version of the string passed
-suitable for C's "" or ''. It will also be valid as a perl "" string.
+A function which returns a 7 bit ASCII correctly \ escaped version of the
+string passed suitable for C's "" or ''. It will die if passed Unicode
+characters.
=cut
sub C_stringify {
local $_ = shift;
return unless defined $_;
+ # grr 5.6.1
+ confess "Wide character in '$_' intended as a C identifier"
+ if tr/\0-\377// != length;
+ # grr 5.6.1 moreso because its regexps will break on data that happens to
+ # be utf8, which includes my 8 bit test cases.
+ $_ = pack 'C*', unpack 'U*', $_ . pack 'U*' if $is_perl56;
+ s/\\/\\\\/g;
+ s/([\"\'])/\\$1/g; # Grr. fix perl mode.
+ s/\n/\\n/g; # Ensure newlines don't end up in octal
+ s/\r/\\r/g;
+ s/\t/\\t/g;
+ s/\f/\\f/g;
+ s/\a/\\a/g;
+ s/([^\0-\177])/sprintf "\\%03o", ord $1/ge;
+ unless ($] < 5.006) {
+ # This will elicit a warning on 5.005_03 about [: :] being reserved unless
+ # I cheat
+ my $cheat = '([[:^print:]])';
+ s/$cheat/sprintf "\\%03o", ord $1/ge;
+ } else {
+ require POSIX;
+ s/([^A-Za-z0-9_])/POSIX::isprint($1) ? $1 : sprintf "\\%03o", ord $1/ge;
+ }
+ $_;
+}
+
+=item perl_stringify NAME
+
+A function which returns a 7 bit ASCII correctly \ escaped version of the
+string passed suitable for a perl "" string.
+
+=cut
+
+# Hopefully make a happy perl identifier.
+sub perl_stringify {
+ local $_ = shift;
+ return unless defined $_;
s/\\/\\\\/g;
s/([\"\'])/\\$1/g; # Grr. fix perl mode.
s/\n/\\n/g; # Ensure newlines don't end up in octal
s/\t/\\t/g;
s/\f/\\f/g;
s/\a/\\a/g;
- s/([[:cntrl:]])/sprintf "\\%03o", ord $1/ge;
- s/\177/\\177/g; # DEL doesn't seem to be a [:cntrl:]
+ unless ($] < 5.006) {
+ if ($] > 5.007) {
+ s/([^\0-\177])/sprintf "\\x{%X}", ord $1/ge;
+ } else {
+ # Grr 5.6.1. And I don't think I can use utf8; to force the regexp
+ # because 5.005_03 will fail.
+ # This is grim, but I also can't split on //
+ my $copy;
+ foreach my $index (0 .. length ($_) - 1) {
+ my $char = substr ($_, $index, 1);
+ $copy .= ($char le "\177") ? $char : sprintf "\\x{%X}", ord $char;
+ }
+ $_ = $copy;
+ }
+ # This will elicit a warning on 5.005_03 about [: :] being reserved unless
+ # I cheat
+ my $cheat = '([[:^print:]])';
+ s/$cheat/sprintf "\\%03o", ord $1/ge;
+ } else {
+ # Turns out "\x{}" notation only arrived with 5.6
+ s/([^\0-\177])/sprintf "\\x%02X", ord $1/ge;
+ require POSIX;
+ s/([^A-Za-z0-9_])/POSIX::isprint($1) ? $1 : sprintf "\\%03o", ord $1/ge;
+ }
$_;
}
push @lines, "#define PERL_constant_NOTFOUND\t$start\n"; $start++;
push @lines, "#define PERL_constant_NOTDEF\t$start\n"; $start++;
foreach (sort keys %XS_Constant) {
+ next if $_ eq '';
push @lines, "#define PERL_constant_IS$_\t$start\n"; $start++;
}
push @lines, << 'EOT';
#ifndef NVTYPE
typedef double NV; /* 5.6 and later define NVTYPE, and typedef NV to it. */
#endif
+#ifndef aTHX_
+#define aTHX_ /* 5.6 or later define this for threading support. */
+#endif
+#ifndef pTHX_
+#define pTHX_ /* 5.6 or later define this for threading support. */
+#endif
EOT
return join '', @lines;
is used to avoid C<memEQ> for short names, or to generate a comment to
highlight the position of the character in the C<switch> statement.
+If I<CHECKED_AT> is a reference to a scalar, then instead it gives
+the characters pre-checked at the beginning, (and the number of chars by
+which the C variable name has been advanced. These need to be chopped from
+the front of I<NAME>).
+
=cut
sub memEQ_clause {
# Which could actually be a character comparison or even ""
my ($name, $checked_at, $indent) = @_;
$indent = ' ' x ($indent || 4);
+ my $front_chop;
+ if (ref $checked_at) {
+ # regexp won't work on 5.6.1 without use utf8; in turn that won't work
+ # on 5.005_03.
+ substr ($name, 0, length $$checked_at,) = '';
+ $front_chop = C_stringify ($$checked_at);
+ undef $checked_at;
+ }
my $len = length $name;
if ($len < 2) {
return $indent . "if (name[$check] == '$char') {\n";
}
}
- # Could optimise a memEQ on 3 to 2 single character checks here
+ if (($len == 2 and !defined $checked_at)
+ or ($len == 3 and defined ($checked_at) and $checked_at == 2)) {
+ my $char1 = C_stringify (substr $name, 0, 1);
+ my $char2 = C_stringify (substr $name, 1, 1);
+ return $indent . "if (name[0] == '$char1' && name[1] == '$char2') {\n";
+ }
+ if (($len == 3 and defined ($checked_at) and $checked_at == 1)) {
+ my $char1 = C_stringify (substr $name, 0, 1);
+ my $char2 = C_stringify (substr $name, 2, 1);
+ return $indent . "if (name[0] == '$char1' && name[2] == '$char2') {\n";
+ }
+
+ my $pointer = '^';
+ my $have_checked_last = defined ($checked_at) && $len == $checked_at + 1;
+ if ($have_checked_last) {
+ # Checked at the last character, so no need to memEQ it.
+ $pointer = C_stringify (chop $name);
+ $len--;
+ }
+
$name = C_stringify ($name);
my $body = $indent . "if (memEQ(name, \"$name\", $len)) {\n";
- $body .= $indent . "/* ". (' ' x $checked_at) . '^'
- . (' ' x ($len - $checked_at + length $len)) . " */\n"
- if defined $checked_at;
+ # Put a little ^ under the letter we checked at
+ # Screws up for non printable and non-7 bit stuff, but that's too hard to
+ # get right.
+ if (defined $checked_at) {
+ $body .= $indent . "/* ". (' ' x $checked_at) . $pointer
+ . (' ' x ($len - $checked_at + length $len)) . " */\n";
+ } elsif (defined $front_chop) {
+ $body .= $indent . "/* $front_chop"
+ . (' ' x ($len + 1 + length $len)) . " */\n";
+ }
return $body;
}
A function to return a suitable assignment clause. If I<TYPE> is aggregate
(eg I<PVN> expects both pointer and length) then there should be multiple
I<VALUE>s for the components. I<PRE> and I<POST> if defined give snippets
-of C code to preceed and follow the assignment. I<PRE> will be at the start
+of C code to proceed and follow the assignment. I<PRE> will be at the start
of a block, so variables may be defined in it.
=cut
$close = "$indent}\n";
$indent .= " ";
}
- die "Can't generate code for type $type" unless exists $XS_TypeSet{$type};
+ confess "undef \$type" unless defined $type;
+ confess "Can't generate code for type $type" unless exists $XS_TypeSet{$type};
my $typeset = $XS_TypeSet{$type};
if (ref $typeset) {
die "Type $type is aggregate, but only single value given"
return $clause;
}
-=item return_clause VALUE, TYPE, INDENT, MACRO, DEFAULT, PRE, POST, PRE, POST
+=item return_clause
-A function to return a suitable C<#ifdef> clause. I<MACRO> defaults to
-I<VALUE> when not defined. If I<TYPE> is aggregate (eg I<PVN> expects both
-pointer and length) then I<VALUE> should be a reference to an array of
-values in the order expected by the type. C<C_constant> will always call
-this function with I<MACRO> defined, defaulting to the constant's name.
-I<DEFAULT> if defined is an array reference giving default type and and
-value(s) if the clause generated by I<MACRO> doesn't evaluate to true.
-The two pairs I<PRE> and I<POST> if defined give C code snippets to proceed
-and follow the value, and the default value.
+return_clause ITEM, INDENT
+
+A function to return a suitable C<#ifdef> clause. I<ITEM> is a hashref
+(as passed to C<C_constant> and C<match_clause>. I<INDENT> is the number
+of spaces to indent, defaulting to 6.
=cut
-sub return_clause ($$$$$$$$$) {
+sub return_clause ($$) {
##ifdef thingy
# *iv_return = thingy;
# return PERL_constant_ISIV;
##else
# return PERL_constant_NOTDEF;
##endif
- my ($value, $type, $indent, $macro, $default, $pre, $post,
- $def_pre, $def_post) = @_;
+ my ($item, $indent) = @_;
+
+ my ($name, $value, $macro, $default, $pre, $post, $def_pre, $def_post, $type)
+ = @$item{qw (name value macro default pre post def_pre def_post type)};
+ $value = $name unless defined $value;
+ $macro = $name unless defined $macro;
+
$macro = $value unless defined $macro;
$indent = ' ' x ($indent || 6);
+ unless ($type) {
+ # use Data::Dumper; print STDERR Dumper ($item);
+ confess "undef \$type";
+ }
my $clause;
$clause .= "#endif\n";
}
}
- return $clause
+ return $clause;
+}
+
+=pod
+
+XXX document me
+
+=cut
+
+sub match_clause {
+ # $offset defined if we have checked an offset.
+ my ($item, $offset, $indent) = @_;
+ $indent = ' ' x ($indent || 4);
+ my $body = '';
+ my ($no, $yes, $either, $name, $inner_indent);
+ if (ref $item eq 'ARRAY') {
+ ($yes, $no) = @$item;
+ $either = $yes || $no;
+ confess "$item is $either expecting hashref in [0] || [1]"
+ unless ref $either eq 'HASH';
+ $name = $either->{name};
+ } else {
+ confess "$item->{name} has utf8 flag '$item->{utf8}', should be false"
+ if $item->{utf8};
+ $name = $item->{name};
+ $inner_indent = $indent;
+ }
+
+ $body .= memEQ_clause ($name, $offset, length $indent);
+ if ($yes) {
+ $body .= $indent . " if (utf8) {\n";
+ } elsif ($no) {
+ $body .= $indent . " if (!utf8) {\n";
+ }
+ if ($either) {
+ $body .= return_clause ($either, 4 + length $indent);
+ if ($yes and $no) {
+ $body .= $indent . " } else {\n";
+ $body .= return_clause ($no, 4 + length $indent);
+ }
+ $body .= $indent . " }\n";
+ } else {
+ $body .= return_clause ($item, 2 + length $indent);
+ }
+ $body .= $indent . "}\n";
}
=item switch_clause INDENT, NAMELEN, ITEMHASH, ITEM...
$body = wrap ($leader, $follower, $comment) . "\n";
$leader = $follower;
}
- $body .= wrap ($leader, $follower, join (" ", @names) . " */") . "\n";
+ my @safe_names = @names;
+ foreach (@safe_names) {
+ confess sprintf "Name '$_' is length %d, not $namelen", length
+ unless length == $namelen;
+ # Argh. 5.6.1
+ # next unless tr/A-Za-z0-9_//c;
+ next if tr/A-Za-z0-9_// == length;
+ $_ = '"' . perl_stringify ($_) . '"';
+ # Ensure that the enclosing C comment doesn't end
+ # by turning */ into *" . "/
+ s!\*\/!\*"."/!gs;
+ # gcc -Wall doesn't like finding /* inside a comment
+ s!\/\*!/"."\*!gs;
+ }
+ $body .= wrap ($leader, $follower, join (" ", @safe_names) . " */") . "\n";
# Figure out what to switch on.
# (RMS, Spread of jump table, Position, Hashref)
my @best = (1e38, ~0);
- foreach my $i (0 .. ($namelen - 1)) {
+ # Prefer the last character over the others. (As it lets us shortern the
+ # memEQ clause at no cost).
+ foreach my $i ($namelen - 1, 0 .. ($namelen - 2)) {
my ($min, $max) = (~0, 0);
my %spread;
+ if ($is_perl56) {
+ # Need proper Unicode preserving hash keys for bytes in range 128-255
+ # here too, for some reason. grr 5.6.1 yet again.
+ tie %spread, 'ExtUtils::Constant::Aaargh56Hash';
+ }
foreach (@names) {
my $char = substr $_, $i, 1;
my $ord = ord $char;
+ confess "char $ord is out of range" if $ord > 255;
$max = $ord if $ord > $max;
$min = $ord if $ord < $min;
push @{$spread{$char}}, $_;
# the string wins. Because if that passes but the memEQ fails, it may
# only need the start of the string to bin the choice.
# I think. But I'm micro-optimising. :-)
+ # OK. Trump that. Now favour the last character of the string, before the
+ # rest.
my $ss;
$ss += @$_ * @$_ foreach values %spread;
my $rms = sqrt ($ss / keys %spread);
@best = ($rms, $max - $min, $i, \%spread);
}
}
- die "Internal error. Failed to pick a switch point for @names"
+ confess "Internal error. Failed to pick a switch point for @names"
unless defined $best[2];
# use Data::Dumper; print Dumper (@best);
my ($offset, $best) = @best[2,3];
$body .= $indent . "/* Offset $offset gives the best switch position. */\n";
- $body .= $indent . "switch (name[$offset]) {\n";
+
+ my $do_front_chop = $offset == 0 && $namelen > 2;
+ if ($do_front_chop) {
+ $body .= $indent . "switch (*name++) {\n";
+ } else {
+ $body .= $indent . "switch (name[$offset]) {\n";
+ }
foreach my $char (sort keys %$best) {
+ confess sprintf "'$char' is %d bytes long, not 1", length $char
+ if length ($char) != 1;
+ confess sprintf "char %#X is out of range", ord $char if ord ($char) > 255;
$body .= $indent . "case '" . C_stringify ($char) . "':\n";
foreach my $name (sort @{$best->{$char}}) {
my $thisone = $items->{$name};
- my ($value, $macro, $default, $pre, $post, $def_pre, $def_post)
- = @$thisone{qw (value macro default pre post def_pre def_post)};
- $value = $name unless defined $value;
- $macro = $name unless defined $macro;
-
- # We have checked this offset.
- $body .= memEQ_clause ($name, $offset, 2 + length $indent);
- $body .= return_clause ($value, $thisone->{type}, 4 + length $indent,
- $macro, $default, $pre, $post,
- $def_pre, $def_post);
- $body .= $indent . " }\n";
+ # warn "You are here";
+ if ($do_front_chop) {
+ $body .= match_clause ($thisone, \$char, 2 + length $indent);
+ } else {
+ $body .= match_clause ($thisone, $offset, 2 + length $indent);
+ }
}
$body .= $indent . " break;\n";
}
warn "ExtUtils::Constant doesn't know how to handle values of type $_" unless defined $XS_Constant{$_};
}
my $params = {};
+ $params->{''} = 1 if $what->{''};
$params->{IV} = 1 if $what->{IV} || $what->{UV} || $what->{PVN};
$params->{NV} = 1 if $what->{NV};
$params->{PV} = 1 if $what->{PV} || $what->{PVN};
=item dump_names
-dump_names PACKAGE, SUBNAME, DEFAULT_TYPE, TYPES, INDENT, BREAKOUT, ITEM...
+dump_names DEFAULT_TYPE, TYPES, INDENT, OPTIONS, ITEM...
An internal function to generate the embedded perl code that will regenerate
-the constant subroutines. Parameters are the same as for C_constant.
+the constant subroutines. I<DEFAULT_TYPE>, I<TYPES> and I<ITEM>s are the
+same as for C_constant. I<INDENT> is treated as number of spaces to indent
+by. I<OPTIONS> is a hashref of options. Currently only C<declare_types> is
+recognised. If the value is true a C<$types> is always declared in the perl
+code generated, if defined and false never declared, and if undefined C<$types>
+is only declared if the values in I<TYPES> as passed in cannot be inferred from
+I<DEFAULT_TYPES> and the I<ITEM>s.
=cut
sub dump_names {
- my ($package, $subname, $default_type, $what, $indent, $breakout, @items)
- = @_;
- my (@simple, @complex);
+ my ($default_type, $what, $indent, $options, @items) = @_;
+ my $declare_types = $options->{declare_types};
+ $indent = ' ' x ($indent || 0);
+
+ my $result;
+ my (@simple, @complex, %used_types);
foreach (@items) {
- my $type = $_->{type} || $default_type;
- if ($type eq $default_type and 0 == ($_->{name} =~ tr/A-Za-z0-9_//c)
+ my $type;
+ if (ref $_) {
+ $type = $_->{type} || $default_type;
+ if ($_->{utf8}) {
+ # For simplicity always skip the bytes case, and reconstitute this entry
+ # from its utf8 twin.
+ next if $_->{utf8} eq 'no';
+ # Copy the hashref, as we don't want to mess with the caller's hashref.
+ $_ = {%$_};
+ unless ($is_perl56) {
+ utf8::decode ($_->{name});
+ } else {
+ $_->{name} = pack 'U*', unpack 'U0U*', $_->{name};
+ }
+ delete $_->{utf8};
+ }
+ } else {
+ $_ = {name=>$_};
+ $type = $default_type;
+ }
+ $used_types{$type}++;
+ if ($type eq $default_type
+ # grr 5.6.1
+ and length $_->{name} == ($_->{name} =~ tr/A-Za-z0-9_//)
and !defined ($_->{macro}) and !defined ($_->{value})
and !defined ($_->{default}) and !defined ($_->{pre})
and !defined ($_->{post}) and !defined ($_->{def_pre})
push @complex, $_;
}
}
- my $result = <<"EOT";
- /* When generated this function returned values for the list of names given
- in this section of perl code. Rather than manually editing these functions
- to add or remove constants, which would result in this comment and section
- of code becoming inaccurate, we recommend that you edit this section of
- code, and use it to regenerate a new set of constant functions which you
- then use to replace the originals.
-
- Regenerate these constant functions by feeding this entire source file to
- perl -x
-#!$^X -w
-use ExtUtils::Constant qw (constant_types C_constant XS_constant);
-
-EOT
- $result .= 'my $types = {map {($_, 1)} qw(' . join (" ", sort keys %$what)
- . ")};\n";
- $result .= wrap ("my \@names = (qw(",
- " ", join (" ", sort @simple) . ")");
+ if (!defined $declare_types) {
+ # Do they pass in any types we weren't already using?
+ foreach (keys %$what) {
+ next if $used_types{$_};
+ $declare_types++; # Found one in $what that wasn't used.
+ last; # And one is enough to terminate this loop
+ }
+ }
+ if ($declare_types) {
+ $result = $indent . 'my $types = {map {($_, 1)} qw('
+ . join (" ", sort keys %$what) . ")};\n";
+ }
+ $result .= wrap ($indent . "my \@names = (qw(",
+ $indent . " ", join (" ", sort @simple) . ")");
if (@complex) {
foreach my $item (sort {$a->{name} cmp $b->{name}} @complex) {
- my $name = C_stringify $item->{name};
- my $line = ",\n {name=>\"$name\"";
+ my $name = perl_stringify $item->{name};
+ my $line = ",\n$indent {name=>\"$name\"";
$line .= ", type=>\"$item->{type}\"" if defined $item->{type};
foreach my $thing (qw (macro value default pre post def_pre def_post)) {
my $value = $item->{$thing};
if (defined $value) {
if (ref $value) {
$line .= ", $thing=>[\""
- . join ('", "', map {C_stringify $_} @$value) . '"]';
+ . join ('", "', map {perl_stringify $_} @$value) . '"]';
} else {
- $line .= ", $thing=>\"" . C_stringify($value) . "\"";
+ $line .= ", $thing=>\"" . perl_stringify($value) . "\"";
}
}
}
}
$result .= ");\n";
+ $result;
+}
+
+
+=item dogfood
+
+dogfood PACKAGE, SUBNAME, DEFAULT_TYPE, TYPES, INDENT, BREAKOUT, ITEM...
+
+An internal function to generate the embedded perl code that will regenerate
+the constant subroutines. Parameters are the same as for C_constant.
+
+=cut
+
+sub dogfood {
+ my ($package, $subname, $default_type, $what, $indent, $breakout, @items)
+ = @_;
+ my $result = <<"EOT";
+ /* When generated this function returned values for the list of names given
+ in this section of perl code. Rather than manually editing these functions
+ to add or remove constants, which would result in this comment and section
+ of code becoming inaccurate, we recommend that you edit this section of
+ code, and use it to regenerate a new set of constant functions which you
+ then use to replace the originals.
+
+ Regenerate these constant functions by feeding this entire source file to
+ perl -x
+
+#!$^X -w
+use ExtUtils::Constant qw (constant_types C_constant XS_constant);
+
+EOT
+ $result .= dump_names ($default_type, $what, 0, {declare_types=>1}, @items);
$result .= <<'EOT';
print constant_types(); # macro defs
EOT
- $package = C_stringify($package);
+ $package = perl_stringify($package);
$result .=
"foreach (C_constant (\"$package\", '$subname', '$default_type', \$types, ";
# The form of the indent parameter isn't defined. (Yet)
Equivalents of I<pre> and I<post> for the default value.
+=item utf8
+
+Generated internally. Is zero or undefined if name is 7 bit ASCII,
+"no" if the name is 8 bit (and so should only match if SvUTF8() is false),
+"yes" if the name is utf8 encoded.
+
+The internals automatically clone any name with characters 128-255 but none
+256+ (ie one that could be either in bytes or utf8) into a second entry
+which is utf8 encoded.
+
=back
I<PACKAGE> is the name of the package, and is only used in comments inside the
# be a hashref, and pinch %$items from our parent to save recalculation.
($namelen, $items) = @$breakout;
} else {
+ if ($is_perl56) {
+ # Need proper Unicode preserving hash keys.
+ $items = {};
+ tie %$items, 'ExtUtils::Constant::Aaargh56Hash';
+ }
$breakout ||= 3;
$default_type ||= 'IV';
if (!ref $what) {
# Figure out what types we're dealing with, and assign all unknowns to the
# default type
}
- foreach (@items) {
- my $name;
- if (ref $_) {
- my $orig = $_;
+ my @new_items;
+ foreach my $orig (@items) {
+ my ($name, $item);
+ if (ref $orig) {
# Make a copy which is a normalised version of the ref passed in.
- $name = $_->{name};
- my ($type, $macro, $value) = @$_{qw (type macro value)};
+ $name = $orig->{name};
+ my ($type, $macro, $value) = @$orig{qw (type macro value)};
$type ||= $default_type;
$what->{$type} = 1;
- $_ = {name=>$name, type=>$type};
+ $item = {name=>$name, type=>$type};
undef $macro if defined $macro and $macro eq $name;
- $_->{macro} = $macro if defined $macro;
+ $item->{macro} = $macro if defined $macro;
undef $value if defined $value and $value eq $name;
- $_->{value} = $value if defined $value;
+ $item->{value} = $value if defined $value;
foreach my $key (qw(default pre post def_pre def_post)) {
my $value = $orig->{$key};
- $_->{$key} = $value if defined $value;
+ $item->{$key} = $value if defined $value;
# warn "$key $value";
}
} else {
- $name = $_;
- $_ = {name=>$_, type=>$default_type};
+ $name = $orig;
+ $item = {name=>$name, type=>$default_type};
$what->{$default_type} = 1;
}
- warn "ExtUtils::Constant doesn't know how to handle values of type $_ used in macro $name" unless defined $XS_Constant{$_->{type}};
- if (exists $items->{$name}) {
- die "Multiple definitions for macro $name";
+ warn "ExtUtils::Constant doesn't know how to handle values of type $_ used in macro $name" unless defined $XS_Constant{$item->{type}};
+ # tr///c is broken on 5.6.1 for utf8, so my original tr/\0-\177//c
+ # doesn't work. Upgrade to 5.8
+ # if ($name !~ tr/\0-\177//c || $] < 5.005_50) {
+ if ($name =~ tr/\0-\177// == length $name || $] < 5.005_50) {
+ # No characters outside 7 bit ASCII.
+ if (exists $items->{$name}) {
+ die "Multiple definitions for macro $name";
+ }
+ $items->{$name} = $item;
+ } else {
+ # No characters outside 8 bit. This is hardest.
+ if (exists $items->{$name} and ref $items->{$name} ne 'ARRAY') {
+ confess "Unexpected ASCII definition for macro $name";
+ }
+ # Again, 5.6.1 tr broken, so s/5\.6.*/5\.8\.0/;
+ # if ($name !~ tr/\0-\377//c) {
+ if ($name =~ tr/\0-\377// == length $name) {
+# if ($] < 5.007) {
+# $name = pack "C*", unpack "U*", $name;
+# }
+ $item->{utf8} = 'no';
+ $items->{$name}[1] = $item;
+ push @new_items, $item;
+ # Copy item, to create the utf8 variant.
+ $item = {%$item};
+ }
+ # Encode the name as utf8 bytes.
+ unless ($is_perl56) {
+ utf8::encode($name);
+ } else {
+# warn "Was >$name< " . length ${name};
+ $name = pack 'C*', unpack 'C*', $name . pack 'U*';
+# warn "Now '${name}' " . length ${name};
+ }
+ if ($items->{$name}[0]) {
+ die "Multiple definitions for macro $name";
+ }
+ $item->{utf8} = 'yes';
+ $item->{name} = $name;
+ $items->{$name}[0] = $item;
+ # We have need for the utf8 flag.
+ $what->{''} = 1;
}
- $items->{$name} = $_;
+ push @new_items, $item;
}
+ @items = @new_items;
+ # use Data::Dumper; print Dumper @items;
}
my $params = params ($what);
my ($body, @subs) = "static int\n$subname (pTHX_ const char *name";
$body .= ", STRLEN len" unless defined $namelen;
+ $body .= ", int utf8" if $params->{''};
$body .= ", IV *iv_return" if $params->{IV};
$body .= ", NV *nv_return" if $params->{NV};
$body .= ", const char **pv_return" if $params->{PV};
} else {
# We are the top level.
$body .= " /* Initially switch on the length of the name. */\n";
- $body .= dump_names ($package, $subname, $default_type, $what, $indent,
- $breakout, @items);
+ $body .= dogfood ($package, $subname, $default_type, $what, $indent,
+ $breakout, @items);
$body .= " switch (len) {\n";
# Need to group names of the same length
my @by_length;
next unless $by_length[$i]; # None of this length
$body .= " case $i:\n";
if (@{$by_length[$i]} == 1) {
- my $thisone = $by_length[$i]->[0];
- my ($name, $value, $macro, $default, $pre, $post, $def_pre, $def_post)
- = @$thisone{qw (name value macro default pre post def_pre def_post)};
- $value = $name unless defined $value;
- $macro = $name unless defined $macro;
-
- $body .= memEQ_clause ($name);
- $body .= return_clause ($value, $thisone->{type}, undef, $macro,
- $default, $pre, $post, $def_pre, $def_post);
- $body .= " }\n";
+ my $only_thing = $by_length[$i]->[0];
+ if ($only_thing->{utf8}) {
+ if ($only_thing->{utf8} eq 'yes') {
+ # With utf8 on flag item is passed in element 0
+ $body .= match_clause ([$only_thing]);
+ } else {
+ # With utf8 off flag item is passed in element 1
+ $body .= match_clause ([undef, $only_thing]);
+ }
+ } else {
+ $body .= match_clause ($only_thing);
+ }
} elsif (@{$by_length[$i]} < $breakout) {
$body .= switch_clause (4, '', $i, $items, @{$by_length[$i]});
} else {
my $what = {};
foreach (@{$by_length[$i]}) {
$what->{$_->{type}} = 1;
+ $what->{''} = 1 if $_->{utf8};
}
$params = params ($what);
push @subs, C_constant ($package, "${subname}_$i", $default_type, $what,
$indent, [$i, $items], @{$by_length[$i]});
$body .= " return ${subname}_$i (aTHX_ name";
+ $body .= ", utf8" if $params->{''};
$body .= ", iv_return" if $params->{IV};
$body .= ", nv_return" if $params->{NV};
$body .= ", pv_return" if $params->{PV};
the number of parameters passed to the C function C<constant>]
You can call the perl visible subroutine something other than C<constant> if
-you give the parameter I<SUBNAME>. The C subroutine it calls defaults to the
+you give the parameter I<SUBNAME>. The C subroutine it calls defaults to
the name of the perl visible subroutine, unless you give the parameter
I<C_SUBNAME>.
INPUT:
SV * sv;
const char * s = SvPV(sv, len);
+EOT
+ if ($params->{''}) {
+ $xs .= << 'EOT';
+ INPUT:
+ int utf8 = SvUTF8(sv);
+EOT
+ }
+ $xs .= << 'EOT';
PPCODE:
EOT
EOT
}
$xs .= " type = $C_subname(aTHX_ s, len";
+ $xs .= ', utf8' if $params->{''};
$xs .= ', &iv' if $params->{IV};
$xs .= ', &nv' if $params->{NV};
$xs .= ', &pv' if $params->{PV};
EOT
foreach $type (sort keys %XS_Constant) {
+ # '' marks utf8 flag needed.
+ next if $type eq '';
$xs .= "\t/* Uncomment this if you need to return ${type}s\n"
unless $what->{$type};
$xs .= " case PERL_constant_IS$type:\n";
}
+=item WriteMakefileSnippet
+
+WriteMakefileSnippet ATTRIBUTE =E<gt> VALUE [, ...]
+
+A function to generate perl code for Makefile.PL that will regenerate
+the constant subroutines. Parameters are named as passed to C<WriteConstants>,
+with the addition of C<INDENT> to specify the number of leading spaces
+(default 2).
+
+Currently only C<INDENT>, C<NAME>, C<DEFAULT_TYPE>, C<NAMES>, C<C_FILE> and
+C<XS_FILE> are recognised.
+
+=cut
+
+sub WriteMakefileSnippet {
+ my %args = @_;
+ my $indent = $args{INDENT} || 2;
+
+ my $result = <<"EOT";
+ExtUtils::Constant::WriteConstants(
+ NAME => '$args{NAME}',
+ NAMES => \\\@names,
+ DEFAULT_TYPE => '$args{DEFAULT_TYPE}',
+EOT
+ foreach (qw (C_FILE XS_FILE)) {
+ next unless exists $args{$_};
+ $result .= sprintf " %-12s => '%s',\n",
+ $_, $args{$_};
+ }
+ $result .= <<'EOT';
+ );
+EOT
+
+ $result =~ s/^/' 'x$indent/gem;
+ return dump_names ($args{DEFAULT_TYPE}, undef, $indent, undef,
+ @{$args{NAMES}})
+ . $result;
+}
+
=item WriteConstants ATTRIBUTE =E<gt> VALUE [, ...]
Writes a file of C code and a file of XS code which you should C<#include>
and C<INCLUDE> in the C and XS sections respectively of your module's XS
-code. You probaby want to do this in your C<Makefile.PL>, so that you can
+code. You probably want to do this in your C<Makefile.PL>, so that you can
easily edit the list of constants without touching the rest of your module.
The attributes supported are
=item C_FILE
The name of the file to write containing the C code. The default is
-C<constants.c>.
+C<const-c.inc>. The C<-> in the name ensures that the file can't be
+mistaken for anything related to a legitimate perl package name, and
+not naming the file C<.c> avoids having to override Makefile.PL's
+C<.xs> to C<.c> rules.
=item XS_FILE
The name of the file to write containing the XS code. The default is
-C<constants.xs>.
+C<const-xs.inc>.
=item SUBNAME
The perl visible name of the XS subroutine generated which will return the
-constants. The default is C<constant>.
+constants. The default is C<constant>.
=item C_SUBNAME
sub WriteConstants {
my %ARGS =
( # defaults
- C_FILE => 'constants.c',
- XS_FILE => 'constants.xs',
+ C_FILE => 'const-c.inc',
+ XS_FILE => 'const-xs.inc',
SUBNAME => 'constant',
DEFAULT_TYPE => 'IV',
@_);
print $c_fh constant_types(); # macro defs
print $c_fh "\n";
- # indent is still undef. Until anyone implents indent style rules with it.
+ # indent is still undef. Until anyone implements indent style rules with it.
foreach (C_constant ($ARGS{NAME}, $ARGS{C_SUBNAME}, $ARGS{DEFAULT_TYPE},
$types, undef, $ARGS{BREAKOUT_AT}, @{$ARGS{NAMES}})) {
print $c_fh $_, "\n"; # C constant subs
close $xs_fh or warn "Error closing $ARGS{XS_FILE}: $!";
}
+package ExtUtils::Constant::Aaargh56Hash;
+# A support module (hack) to provide sane Unicode hash keys on 5.6.x perl
+use strict;
+require Tie::Hash if $ExtUtils::Constant::is_perl56;
+use vars '@ISA';
+@ISA = 'Tie::StdHash';
+
+#my $a;
+# Storing the values as concatenated BER encoded numbers is actually going to
+# be terser than using UTF8 :-)
+# And the tests are slightly faster. Ops are bad, m'kay
+sub to_key {pack "w*", unpack "U*", ($_[0] . pack "U*")};
+sub from_key {defined $_[0] ? pack "U*", unpack 'w*', $_[0] : undef};
+
+sub STORE { $_[0]->{to_key($_[1])} = $_[2] }
+sub FETCH { $_[0]->{to_key($_[1])} }
+sub FIRSTKEY { my $a = scalar keys %{$_[0]}; from_key (each %{$_[0]}) }
+sub NEXTKEY { from_key (each %{$_[0]}) }
+sub EXISTS { exists $_[0]->{to_key($_[1])} }
+sub DELETE { delete $_[0]->{to_key($_[1])} }
+
+#END {warn "$a accesses";}
1;
__END__
projects / p5sagit/p5-mst-13.2.git / blobdiff