package utf8;
use strict;
use warnings;
+use re "/aa"; # So we won't even try to look at above Latin1, potentially
+ # resulting in a recursive call
sub DEBUG () { 0 }
+$|=1 if DEBUG;
sub DESTROY {}
sub croak { require Carp; Carp::croak(@_) }
+# Digits may be separated by a single underscore
+my $digits = qr/ ( [0-9] _? )+ (?!:_) /x;
+
+# A sign can be surrounded by white space
+my $sign = qr/ \s* [+-]? \s* /x;
+
+my $f_float = qr/ $sign $digits+ \. $digits* # e.g., 5.0, 5.
+ | $sign $digits* \. $digits+/x; # 0.7, .7
+
+# A number may be an integer, a rational, or a float with an optional exponent
+# We (shudder) accept a signed denominator
+my $number = qr{ ^ $sign $digits+ $
+ | ^ $sign $digits+ \/ $sign $digits+ $
+ | ^ $f_float (?: [Ee] [+-]? $digits )? $}x;
+
+sub _loose_name ($) {
+ # Given a lowercase property or property-value name, return its
+ # standardized version that is expected for look-up in the 'loose' hashes
+ # in Heavy.pl (hence, this depends on what mktables does). This squeezes
+ # out blanks, underscores and dashes. The complication stems from the
+ # grandfathered-in 'L_', which retains a single trailing underscore.
+
+# integer or float (no exponent)
+my $integer_or_float_re = qr/ ^ -? \d+ (:? \. \d+ )? $ /x;
+
+# Also includes rationals
+my $numeric_re = qr! $integer_or_float_re | ^ -? \d+ / \d+ $ !x;
+ return $_[0] if $_[0] =~ $numeric_re;
+
+ (my $loose = $_[0]) =~ s/[-_ \t]//g;
+
+ return $loose if $loose !~ / ^ (?: is | to )? l $/x;
+ return 'l_' if $_[0] =~ / l .* _ /x; # If original had a trailing '_'
+ return $loose;
+}
+
##
## "SWASH" == "SWATCH HASH". A "swatch" is a swatch of the Unicode landscape.
## It's a data structure that encodes a set of Unicode characters.
sub SWASHNEW {
my ($class, $type, $list, $minbits, $none) = @_;
+ my $user_defined = 0;
local $^D = 0 if $^D;
$class = "" unless defined $class;
##
## Callers of swash_init:
## op.c:pmtrans -- for tr/// and y///
- ## regexec.c:regclass_swash -- for /[]/, \p, and \P
- ## utf8.c:is_utf8_common -- for common Unicode properties
- ## utf8.c:to_utf8_case -- for lc, uc, ucfirst, etc. and //i
+ ## Unicode::UCD::prop_invlist
+ ## Unicode::UCD::prop_invmap
##
## Given a $type, our goal is to fill $list with the set of codepoint
## ranges. If $type is false, $list passed is used.
## $none is undocumented, so I'm (khw) trying to do some documentation
## of it now. It appears to be if there is a mapping in an input file
## that maps to 'XXXX', then that is replaced by $none+1, expressed in
- ## hexadecimal. The only place I found it possibly used was in
- ## S_pmtrans in op.c.
+ ## hexadecimal. It is used somehow in tr///.
##
## To make the parsing of $type clear, this code takes the a rather
## unorthodox approach of last'ing out of the block once we have the
## info we need. Were this to be a subroutine, the 'last' would just
## be a 'return'.
##
+ # If a problem is found $type is returned;
+ # Upon success, a new (or cached) blessed object is returned with
+ # keys TYPE, BITS, EXTRAS, LIST, and NONE with values having the
+ # same meanings as the input parameters.
+ # SPECIALS contains a reference to any special-treatment hash in the
+ # property.
+ # INVERT_IT is non-zero if the result should be inverted before use
+ # USER_DEFINED is non-zero if the result came from a user-defined
my $file; ## file to load data from, and also part of the %Cache key.
- my $ListSorted = 0;
# Change this to get a different set of Unicode tables
my $unicore_dir = 'unicore';
+ my $invert_it = 0;
+ my $list_is_from_mktables = 0; # Is $list returned from a mktables
+ # generated file? If so, we know it's
+ # well behaved.
if ($type)
{
-
# Verify that this isn't a recursive call for this property.
- # Can't use croak, as it may try to recurse here itself.
+ # Can't use croak, as it may try to recurse to here itself.
my $class_type = $class . "::$type";
if (grep { $_ eq $class_type } @recursed) {
CORE::die "panic: Infinite recursion in SWASHNEW for '$type'\n";
# regcomp.c surrounds the property name with '__" and '_i' if this
# is to be caseless matching.
- my $caseless = $type =~ s/^__(.*)_i$/$1/;
+ my $caseless = $type =~ s/^(.*)__(.*)_i$/$1$2/;
print STDERR __LINE__, ": type=$type, caseless=$caseless\n" if DEBUG;
## package if no package given
##
- my $caller1 = $type =~ s/(.+)::// ? $1 : caller(1);
+
+ my $caller0 = caller(0);
+ my $caller1 = $type =~ s/(.+):://
+ ? $1
+ : $caller0 eq 'main'
+ ? 'main'
+ : caller(1);
if (defined $caller1 && $type =~ /^I[ns]\w+$/) {
my $prop = "${caller1}::$type";
if $tainted;
no strict 'refs';
$list = &{$prop}($caseless);
+ $user_defined = 1;
last GETFILE;
}
}
}
if (miniperl) {
eval "require '$unicore_dir/Heavy.pl'";
- last GETFILE if $@;
+ if ($@) {
+ print STDERR __LINE__, ": '$@'\n" if DEBUG;
+ pop @recursed if @recursed;
+ return $type;
+ }
}
else {
require "$unicore_dir/Heavy.pl";
print STDERR __LINE__, ": $property\n" if DEBUG;
# Here it is the compound property=table form. The property
- # name is always loosely matched, which means remove any of
- # these:
- $property =~ s/[_\s-]//g;
+ # name is always loosely matched, and always can have an
+ # optional 'is' prefix (which isn't true in the single
+ # form).
+ $property = _loose_name($property) =~ s/^is//r;
# And convert to canonical form. Quit if not valid.
$property = $utf8::loose_property_name_of{$property};
# If the rhs looks like it is a number...
print STDERR __LINE__, ": table=$table\n" if DEBUG;
- if ($table =~ qr{ ^ [ \s 0-9 _ + / . -]+ $ }x) {
- print STDERR __LINE__, ": table=$table\n" if DEBUG;
- # Don't allow leading nor trailing slashes
- if ($table =~ / ^ \/ | \/ $ /x) {
- pop @recursed if @recursed;
- return $type;
- }
+ if ($table =~ $number) {
+ print STDERR __LINE__, ": table=$table\n" if DEBUG;
# Split on slash, in case it is a rational, like \p{1/5}
- my @parts = split qr{ \s* / \s* }x, $table, -1;
+ my @parts = split m{ \s* / \s* }x, $table, -1;
print __LINE__, ": $type\n" if @parts > 2 && DEBUG;
- # Can have maximum of one slash
- if (@parts > 2) {
- pop @recursed if @recursed;
- return $type;
- }
-
foreach my $part (@parts) {
print __LINE__, ": part=$part\n" if DEBUG;
# minus
# Remove underscores between digits.
- $part =~ s/( ?<= [0-9] ) _ (?= [0-9] ) //xg;
+ $part =~ s/(?<= [0-9] ) _ (?= [0-9] ) //xg;
# No leading zeros (but don't make a single '0'
# into a null string)
$part .= '0' if $part eq '-' || $part eq "";
# No trailing zeros after a decimal point
- $part =~ s/ ( \. .*? ) 0+ $ /$1/x;
+ $part =~ s/ ( \. [0-9]*? ) 0+ $ /$1/x;
# Begin with a 0 if a leading decimal point
$part =~ s/ ^ ( -? ) \. /${1}0./x;
print STDERR __LINE__, ": part=$part\n" if DEBUG;
#return $type if $part eq "";
-
- # Result better look like a number. (This test is
- # needed because, for example could have a plus in
- # the middle.)
- if ($part !~ / ^ -? [0-9]+ ( \. [0-9]+)? $ /x) {
- pop @recursed if @recursed;
- return $type;
- }
}
# If a rational...
if ($parts[1] =~ s/^-//) {
# If numerator is also negative, convert the
- # whole thing to positive, or move the minus to
- # the numerator
+ # whole thing to positive, else move the minus
+ # to the numerator
if ($parts[0] !~ s/^-//) {
$parts[0] = '-' . $parts[0];
}
$table = $parts[0];
} else {
- # Here is a floating point numeric_value. Try to
- # convert to rational. First see if is in the list
- # of known ones.
- if (exists $utf8::nv_floating_to_rational{$parts[0]}) {
- $table = $utf8::nv_floating_to_rational{$parts[0]};
- } else {
-
- # Here not in the list. See if is close
- # enough to something in the list. First
- # determine what 'close enough' means. It has
- # to be as tight as what mktables says is the
- # maximum slop, and as tight as how many
- # digits we were passed. That is, if the user
- # said .667, .6667, .66667, etc. we match as
- # many digits as they passed until get to
- # where it doesn't matter any more due to the
- # machine's precision. If they said .6666668,
- # we fail.
- (my $fraction = $parts[0]) =~ s/^.*\.//;
- my $epsilon = 10 ** - (length($fraction));
- if ($epsilon > $utf8::max_floating_slop) {
- $epsilon = $utf8::max_floating_slop;
- }
+ # Here is a floating point numeric_value. Convert
+ # to rational. Get a normalized form, like
+ # 5.00E-01, and look that up in the hash
- # But it can't be tighter than the minimum
- # precision for this machine. If haven't
- # already calculated that minimum, do so now.
- if (! defined $min_floating_slop) {
-
- # Keep going down an order of magnitude
- # until find that adding this quantity to
- # 1 remains 1; but put an upper limit on
- # this so in case this algorithm doesn't
- # work properly on some platform, that we
- # won't loop forever.
- my $count = 0;
- $min_floating_slop = 1;
- while (1+ $min_floating_slop != 1
- && $count++ < 50)
- {
- my $next = $min_floating_slop / 10;
- last if $next == 0; # If underflows,
- # use previous one
- $min_floating_slop = $next;
- print STDERR __LINE__, ": min_float_slop=$min_floating_slop\n" if DEBUG;
- }
-
- # Back off a couple orders of magnitude,
- # just to be safe.
- $min_floating_slop *= 100;
- }
-
- if ($epsilon < $min_floating_slop) {
- $epsilon = $min_floating_slop;
- }
- print STDERR __LINE__, ": fraction=.$fraction; epsilon=$epsilon\n" if DEBUG;
-
- undef $table;
-
- # And for each possible rational in the table,
- # see if it is within epsilon of the input.
- foreach my $official
- (keys %utf8::nv_floating_to_rational)
- {
- print STDERR __LINE__, ": epsilon=$epsilon, official=$official, diff=", abs($parts[0] - $official), "\n" if DEBUG;
- if (abs($parts[0] - $official) < $epsilon) {
- $table =
- $utf8::nv_floating_to_rational{$official};
- last;
- }
- }
+ my $float = sprintf "%.*e",
+ $utf8::e_precision,
+ 0 + $parts[0];
- # Quit if didn't find one.
- if (! defined $table) {
- pop @recursed if @recursed;
- return $type;
- }
+ if (exists $utf8::nv_floating_to_rational{$float}) {
+ $table = $utf8::nv_floating_to_rational{$float};
+ } else {
+ pop @recursed if @recursed;
+ return $type;
}
}
print STDERR __LINE__, ": $property=$table\n" if DEBUG;
# If didn't find it, try again with looser matching by editing
# out the applicable characters on the rhs and looking up
# again.
+ my $strict_property_and_table;
if (! defined $file) {
- $table =~ s/ [_\s-] //xg;
+
+ # This isn't used unless the name begins with 'to'
+ $strict_property_and_table = $property_and_table =~ s/^to//r;
+ $table = _loose_name($table);
$property_and_table = "$prefix$table";
print STDERR __LINE__, ": $property_and_table\n" if DEBUG;
$file = $utf8::loose_to_file_of{$property_and_table};
# Add the constant and go fetch it in.
if (defined $file) {
+
+ # If the file name contains a !, it means to invert. The
+ # 0+ makes sure result is numeric
+ $invert_it = 0 + $file =~ s/!//;
+
if ($utf8::why_deprecated{$file}) {
warnings::warnif('deprecated', "Use of '$type' in \\p{} or \\P{} is deprecated because: $utf8::why_deprecated{$file};");
}
{
$file = $utf8::caseless_equivalent{$property_and_table};
}
- $file= "$unicore_dir/lib/$file.pl";
+
+ # The pseudo-directory '#' means that there really isn't a
+ # file to read, the data is in-line as part of the string;
+ # we extract it below.
+ $file = "$unicore_dir/lib/$file.pl" unless $file =~ m!^#/!;
last GETFILE;
}
print STDERR __LINE__, ": didn't find $property_and_table\n" if DEBUG;
##
- ## See if it's a user-level "To".
- ##
-
- my $caller0 = caller(0);
-
- if (defined $caller0 && $type =~ /^To(?:\w+)$/) {
- my $map = $caller0 . "::" . $type;
-
- if (exists &{$map}) {
- no strict 'refs';
-
- $list = &{$map};
- warnings::warnif('deprecated', "User-defined case-mapping '$type' is deprecated");
- last GETFILE;
- }
- }
-
- ##
## Last attempt -- see if it's a standard "To" name
## (e.g. "ToLower") ToTitle is used by ucfirst().
## The user-level way to access ToDigit() and ToFold()
## is to use Unicode::UCD.
##
- if ($type =~ /^To(Digit|Fold|Lower|Title|Upper)$/) {
- $file = "$unicore_dir/To/$1.pl";
- ## would like to test to see if $file actually exists....
- last GETFILE;
+ # Only check if caller wants non-binary
+ if ($minbits != 1) {
+ if ($property_and_table =~ s/^to//) {
+ # Look input up in list of properties for which we have
+ # mapping files. First do it with the strict approach
+ if (defined ($file = $utf8::strict_property_to_file_of{
+ $strict_property_and_table}))
+ {
+ $type = $utf8::file_to_swash_name{$file};
+ print STDERR __LINE__, ": type set to $type\n"
+ if DEBUG;
+ $file = "$unicore_dir/$file.pl";
+ last GETFILE;
+ }
+ elsif (defined ($file =
+ $utf8::loose_property_to_file_of{$property_and_table}))
+ {
+ $type = $utf8::file_to_swash_name{$file};
+ print STDERR __LINE__, ": type set to $type\n"
+ if DEBUG;
+ $file = "$unicore_dir/$file.pl";
+ last GETFILE;
+ } # If that fails see if there is a corresponding binary
+ # property file
+ elsif (defined ($file =
+ $utf8::loose_to_file_of{$property_and_table}))
+ {
+
+ # Here, there is no map file for the property we
+ # are trying to get the map of, but this is a
+ # binary property, and there is a file for it that
+ # can easily be translated to a mapping, so use
+ # that, treating this as a binary property.
+ # Setting 'minbits' here causes it to be stored as
+ # such in the cache, so if someone comes along
+ # later looking for just a binary, they get it.
+ $minbits = 1;
+
+ # The 0+ makes sure is numeric
+ $invert_it = 0 + $file =~ s/!//;
+ $file = "$unicore_dir/lib/$file.pl"
+ unless $file =~ m!^#/!;
+ last GETFILE;
+ }
+ }
}
##
pop @recursed if @recursed;
return $type;
- }
+ } # end of GETFILE block
if (defined $file) {
print STDERR __LINE__, ": found it (file='$file')\n" if DEBUG;
## If we reach here, it was due to a 'last GETFILE' above
## (exception: user-defined properties and mappings), so we
## have a filename, so now we load it if we haven't already.
- ## If we have, return the cached results. The cache key is the
- ## class and file to load.
- ##
- my $found = $Cache{$class, $file};
- if ($found and ref($found) eq $class) {
- print STDERR __LINE__, ": Returning cached '$file' for \\p{$type}\n" if DEBUG;
- pop @recursed if @recursed;
- return $found;
+
+ # The pseudo-directory '#' means the result isn't really a
+ # file, but is in-line, with semi-colons to be turned into
+ # new-lines. Since it is in-line there is no advantage to
+ # caching the result
+ if ($file =~ s!^#/!!) {
+ $list = $utf8::inline_definitions[$file];
+ }
+ else {
+ # Here, we have an actual file to read in and load, but it
+ # may already have been read-in and cached. The cache key
+ # is the class and file to load, and whether the results
+ # need to be inverted.
+ my $found = $Cache{$class, $file, $invert_it};
+ if ($found and ref($found) eq $class) {
+ print STDERR __LINE__, ": Returning cached swash for '$class,$file,$invert_it' for \\p{$type}\n" if DEBUG;
+ pop @recursed if @recursed;
+ return $found;
+ }
+
+ local $@;
+ local $!;
+ $list = do $file; die $@ if $@;
}
- local $@;
- local $!;
- $list = do $file; die $@ if $@;
+ $list_is_from_mktables = 1;
}
+ } # End of $type is non-null
- $ListSorted = 1; ## we know that these lists are sorted
- }
+ # Here, either $type was null, or we found the requested property and
+ # read it into $list
+
+ my $extras = "";
- my $extras;
my $bits = $minbits;
- if ($list) {
+ # mktables lists don't have extras, like '&utf8::prop', so don't need
+ # to separate them; also lists are already sorted, so don't need to do
+ # that.
+ if ($list && ! $list_is_from_mktables) {
my $taint = substr($list,0,0); # maintain taint
- my @tmp = split(/^/m, $list);
- my %seen;
- no warnings;
- $extras = join '', $taint, grep /^[^0-9a-fA-F]/, @tmp;
- $list = join '', $taint,
- map { $_->[1] }
- sort { $a->[0] <=> $b->[0] }
- map { /^([0-9a-fA-F]+)/; [ CORE::hex($1), $_ ] }
- grep { /^([0-9a-fA-F]+)/ and not $seen{$1}++ } @tmp; # XXX doesn't do ranges right
+
+ # Separate the extras from the code point list, and make sure
+ # user-defined properties and tr/// are well-behaved for
+ # downstream code.
+ if ($user_defined || $none) {
+ my @tmp = split(/^/m, $list);
+ my %seen;
+ no warnings;
+
+ # The extras are anything that doesn't begin with a hex digit.
+ $extras = join '', $taint, grep /^[^0-9a-fA-F]/, @tmp;
+
+ # Remove the extras, and sort the remaining entries by the
+ # numeric value of their beginning hex digits, removing any
+ # duplicates.
+ $list = join '', $taint,
+ map { $_->[1] }
+ sort { $a->[0] <=> $b->[0] }
+ map { /^([0-9a-fA-F]+)/ && !$seen{$1}++ ? [ CORE::hex($1), $_ ] : () }
+ @tmp; # XXX doesn't do ranges right
+ }
+ else {
+ # mktables has gone to some trouble to make non-user defined
+ # properties well-behaved, so we can skip the effort we do for
+ # user-defined ones. Any extras are at the very beginning of
+ # the string.
+
+ # This regex splits out the first lines of $list into $1 and
+ # strips them off from $list, until we get one that begins
+ # with a hex number, alone on the line, or followed by a tab.
+ # Either portion may be empty.
+ $list =~ s/ \A ( .*? )
+ (?: \z | (?= ^ [0-9a-fA-F]+ (?: \t | $) ) )
+ //msx;
+
+ $extras = "$taint$1";
+ }
}
if ($none) {
elsif ($c =~ /^([0-9a-fA-F]+)/) {
$subobj = utf8->SWASHNEW("", $c, $minbits, 0);
}
+ print STDERR __LINE__, ": returned from getting sub object for $name\n" if DEBUG;
if (! ref $subobj) {
pop @recursed if @recursed && $type;
return $subobj;
}
push @extras, $name => $subobj;
$bits = $subobj->{BITS} if $bits < $subobj->{BITS};
+ $user_defined = $subobj->{USER_DEFINED}
+ if $subobj->{USER_DEFINED};
}
}
}
}
if (DEBUG) {
- print STDERR __LINE__, ": CLASS = $class, TYPE => $type, BITS => $bits, NONE => $none";
+ print STDERR __LINE__, ": CLASS = $class, TYPE => $type, BITS => $bits, NONE => $none, INVERT_IT => $invert_it, USER_DEFINED => $user_defined";
print STDERR "\nLIST =>\n$list" if defined $list;
print STDERR "\nEXTRAS =>\n$extras" if defined $extras;
print STDERR "\n";
EXTRAS => $extras,
LIST => $list,
NONE => $none,
+ USER_DEFINED => $user_defined,
@extras,
} => $class;
if ($file) {
- $Cache{$class, $file} = $SWASH;
+ $Cache{$class, $file, $invert_it} = $SWASH;
+ if ($type
+ && exists $utf8::SwashInfo{$type}
+ && exists $utf8::SwashInfo{$type}{'specials_name'})
+ {
+ my $specials_name = $utf8::SwashInfo{$type}{'specials_name'};
+ no strict "refs";
+ print STDERR "\nspecials_name => $specials_name\n" if DEBUG;
+ $SWASH->{'SPECIALS'} = \%$specials_name;
+ }
+ $SWASH->{'INVERT_IT'} = $invert_it;
}
pop @recursed if @recursed && $type;
}
}
-# Now SWASHGET is recasted into a C function S_swash_get (see utf8.c).
+# Now SWASHGET is recasted into a C function S_swatch_get (see utf8.c).
1;