This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
Change mktables to use pack 'J', $addr; to key its inside-out objects.
[perl5.git] / lib / unicore / mktables
CommitLineData
d73e5302 1#!/usr/bin/perl -w
99870f4d
KW
2
3# !!!!!!!!!!!!!! IF YOU MODIFY THIS FILE !!!!!!!!!!!!!!!!!!!!!!!!!
4# Any files created or read by this program should be listed in 'mktables.lst'
5# Use -makelist to regenerate it.
6
23e33b60
KW
7# Needs 'no overloading' to run faster on miniperl. Code commented out at the
8# subroutine objaddr can be used instead to work as far back (untested) as
f998e60c
KW
9# 5.8: needs pack "U". But almost all occurrences of objaddr have been
10# removed in favor of using 'no overloading'. You also would have to go
11# through and replace occurrences like:
051df77b 12# my $addr; { no overloading; $addr = pack 'J', $self; }
f998e60c
KW
13# with
14# my $addr = main::objaddr $self;
6c68572b 15# (or reverse commit 9b01bafde4b022706c3d6f947a0963f821b2e50b
051df77b
NC
16# that instituted the change to main::objaddr, and subsequent commits that
17# changed 0+$self to pack 'J', $self.)
6c68572b 18
23e33b60 19require 5.010_001;
d73e5302 20use strict;
99870f4d 21use warnings;
cf25bb62 22use Carp;
99870f4d
KW
23use File::Find;
24use File::Path;
d07a55ed 25use File::Spec;
99870f4d
KW
26use Text::Tabs;
27
28sub DEBUG () { 0 } # Set to 0 for production; 1 for development
29
30##########################################################################
31#
32# mktables -- create the runtime Perl Unicode files (lib/unicore/.../*.pl),
33# from the Unicode database files (lib/unicore/.../*.txt), It also generates
34# a pod file and a .t file
35#
36# The structure of this file is:
37# First these introductory comments; then
38# code needed for everywhere, such as debugging stuff; then
39# code to handle input parameters; then
40# data structures likely to be of external interest (some of which depend on
41# the input parameters, so follows them; then
42# more data structures and subroutine and package (class) definitions; then
43# the small actual loop to process the input files and finish up; then
44# a __DATA__ section, for the .t tests
45#
46# This program works on all releases of Unicode through at least 5.2. The
47# outputs have been scrutinized most intently for release 5.1. The others
48# have been checked for somewhat more than just sanity. It can handle all
49# existing Unicode character properties in those releases.
50#
99870f4d
KW
51# This program is mostly about Unicode character (or code point) properties.
52# A property describes some attribute or quality of a code point, like if it
53# is lowercase or not, its name, what version of Unicode it was first defined
54# in, or what its uppercase equivalent is. Unicode deals with these disparate
55# possibilities by making all properties into mappings from each code point
56# into some corresponding value. In the case of it being lowercase or not,
57# the mapping is either to 'Y' or 'N' (or various synonyms thereof). Each
58# property maps each Unicode code point to a single value, called a "property
59# value". (Hence each Unicode property is a true mathematical function with
60# exactly one value per code point.)
61#
62# When using a property in a regular expression, what is desired isn't the
63# mapping of the code point to its property's value, but the reverse (or the
64# mathematical "inverse relation"): starting with the property value, "Does a
65# code point map to it?" These are written in a "compound" form:
66# \p{property=value}, e.g., \p{category=punctuation}. This program generates
67# files containing the lists of code points that map to each such regular
68# expression property value, one file per list
69#
70# There is also a single form shortcut that Perl adds for many of the commonly
71# used properties. This happens for all binary properties, plus script,
72# general_category, and block properties.
73#
74# Thus the outputs of this program are files. There are map files, mostly in
75# the 'To' directory; and there are list files for use in regular expression
76# matching, all in subdirectories of the 'lib' directory, with each
77# subdirectory being named for the property that the lists in it are for.
78# Bookkeeping, test, and documentation files are also generated.
79
80my $matches_directory = 'lib'; # Where match (\p{}) files go.
81my $map_directory = 'To'; # Where map files go.
82
83# DATA STRUCTURES
84#
85# The major data structures of this program are Property, of course, but also
86# Table. There are two kinds of tables, very similar to each other.
87# "Match_Table" is the data structure giving the list of code points that have
88# a particular property value, mentioned above. There is also a "Map_Table"
89# data structure which gives the property's mapping from code point to value.
90# There are two structures because the match tables need to be combined in
91# various ways, such as constructing unions, intersections, complements, etc.,
92# and the map ones don't. And there would be problems, perhaps subtle, if
93# a map table were inadvertently operated on in some of those ways.
94# The use of separate classes with operations defined on one but not the other
95# prevents accidentally confusing the two.
96#
97# At the heart of each table's data structure is a "Range_List", which is just
98# an ordered list of "Ranges", plus ancillary information, and methods to
99# operate on them. A Range is a compact way to store property information.
100# Each range has a starting code point, an ending code point, and a value that
101# is meant to apply to all the code points between the two end points,
102# inclusive. For a map table, this value is the property value for those
103# code points. Two such ranges could be written like this:
104# 0x41 .. 0x5A, 'Upper',
105# 0x61 .. 0x7A, 'Lower'
106#
107# Each range also has a type used as a convenience to classify the values.
108# Most ranges in this program will be Type 0, or normal, but there are some
109# ranges that have a non-zero type. These are used only in map tables, and
110# are for mappings that don't fit into the normal scheme of things. Mappings
111# that require a hash entry to communicate with utf8.c are one example;
112# another example is mappings for charnames.pm to use which indicate a name
113# that is algorithmically determinable from its code point (and vice-versa).
114# These are used to significantly compact these tables, instead of listing
115# each one of the tens of thousands individually.
116#
117# In a match table, the value of a range is irrelevant (and hence the type as
118# well, which will always be 0), and arbitrarily set to the null string.
119# Using the example above, there would be two match tables for those two
120# entries, one named Upper would contain the 0x41..0x5A range, and the other
121# named Lower would contain 0x61..0x7A.
122#
123# Actually, there are two types of range lists, "Range_Map" is the one
124# associated with map tables, and "Range_List" with match tables.
125# Again, this is so that methods can be defined on one and not the other so as
126# to prevent operating on them in incorrect ways.
127#
128# Eventually, most tables are written out to files to be read by utf8_heavy.pl
129# in the perl core. All tables could in theory be written, but some are
130# suppressed because there is no current practical use for them. It is easy
131# to change which get written by changing various lists that are near the top
132# of the actual code in this file. The table data structures contain enough
133# ancillary information to allow them to be treated as separate entities for
134# writing, such as the path to each one's file. There is a heading in each
135# map table that gives the format of its entries, and what the map is for all
136# the code points missing from it. (This allows tables to be more compact.)
678f13d5 137#
99870f4d
KW
138# The Property data structure contains one or more tables. All properties
139# contain a map table (except the $perl property which is a
140# pseudo-property containing only match tables), and any properties that
141# are usable in regular expression matches also contain various matching
142# tables, one for each value the property can have. A binary property can
143# have two values, True and False (or Y and N, which are preferred by Unicode
144# terminology). Thus each of these properties will have a map table that
145# takes every code point and maps it to Y or N (but having ranges cuts the
146# number of entries in that table way down), and two match tables, one
147# which has a list of all the code points that map to Y, and one for all the
148# code points that map to N. (For each of these, a third table is also
149# generated for the pseudo Perl property. It contains the identical code
150# points as the Y table, but can be written, not in the compound form, but in
151# a "single" form like \p{IsUppercase}.) Many properties are binary, but some
152# properties have several possible values, some have many, and properties like
153# Name have a different value for every named code point. Those will not,
154# unless the controlling lists are changed, have their match tables written
155# out. But all the ones which can be used in regular expression \p{} and \P{}
156# constructs will. Generally a property will have either its map table or its
157# match tables written but not both. Again, what gets written is controlled
158# by lists which can easily be changed.
678f13d5 159#
99870f4d
KW
160# For information about the Unicode properties, see Unicode's UAX44 document:
161
162my $unicode_reference_url = 'http://www.unicode.org/reports/tr44/';
163
164# As stated earlier, this program will work on any release of Unicode so far.
165# Most obvious problems in earlier data have NOT been corrected except when
166# necessary to make Perl or this program work reasonably. For example, no
167# folding information was given in early releases, so this program uses the
168# substitute of lower case, just so that a regular expression with the /i
169# option will do something that actually gives the right results in many
170# cases. There are also a couple other corrections for version 1.1.5,
171# commented at the point they are made. As an example of corrections that
172# weren't made (but could be) is this statement from DerivedAge.txt: "The
173# supplementary private use code points and the non-character code points were
174# assigned in version 2.0, but not specifically listed in the UCD until
175# versions 3.0 and 3.1 respectively." (To be precise it was 3.0.1 not 3.0.0)
176# More information on Unicode version glitches is further down in these
177# introductory comments.
178#
179# This program works on all properties as of 5.2, though the files for some
678f13d5
KW
180# are suppressed from apparent lack of demand for them. You can change which
181# are output by changing lists in this program.
182#
99870f4d
KW
183# The old version of mktables emphasized the term "Fuzzy" to mean Unocde's
184# loose matchings rules (from Unicode TR18):
185#
186# The recommended names for UCD properties and property values are in
187# PropertyAliases.txt [Prop] and PropertyValueAliases.txt
188# [PropValue]. There are both abbreviated names and longer, more
189# descriptive names. It is strongly recommended that both names be
190# recognized, and that loose matching of property names be used,
191# whereby the case distinctions, whitespace, hyphens, and underbar
192# are ignored.
193# The program still allows Fuzzy to override its determination of if loose
194# matching should be used, but it isn't currently used, as it is no longer
195# needed; the calculations it makes are good enough.
678f13d5 196#
99870f4d
KW
197# SUMMARY OF HOW IT WORKS:
198#
199# Process arguments
200#
201# A list is constructed containing each input file that is to be processed
202#
203# Each file on the list is processed in a loop, using the associated handler
204# code for each:
205# The PropertyAliases.txt and PropValueAliases.txt files are processed
206# first. These files name the properties and property values.
207# Objects are created of all the property and property value names
208# that the rest of the input should expect, including all synonyms.
209# The other input files give mappings from properties to property
210# values. That is, they list code points and say what the mapping
211# is under the given property. Some files give the mappings for
212# just one property; and some for many. This program goes through
213# each file and populates the properties from them. Some properties
214# are listed in more than one file, and Unicode has set up a
215# precedence as to which has priority if there is a conflict. Thus
216# the order of processing matters, and this program handles the
217# conflict possibility by processing the overriding input files
218# last, so that if necessary they replace earlier values.
219# After this is all done, the program creates the property mappings not
220# furnished by Unicode, but derivable from what it does give.
221# The tables of code points that match each property value in each
222# property that is accessible by regular expressions are created.
223# The Perl-defined properties are created and populated. Many of these
224# require data determined from the earlier steps
225# Any Perl-defined synonyms are created, and name clashes between Perl
678f13d5 226# and Unicode are reconciled and warned about.
99870f4d
KW
227# All the properties are written to files
228# Any other files are written, and final warnings issued.
678f13d5 229#
99870f4d
KW
230# For clarity, a number of operators have been overloaded to work on tables:
231# ~ means invert (take all characters not in the set). The more
232# conventional '!' is not used because of the possibility of confusing
233# it with the actual boolean operation.
234# + means union
235# - means subtraction
236# & means intersection
237# The precedence of these is the order listed. Parentheses should be
238# copiously used. These are not a general scheme. The operations aren't
239# defined for a number of things, deliberately, to avoid getting into trouble.
240# Operations are done on references and affect the underlying structures, so
241# that the copy constructors for them have been overloaded to not return a new
242# clone, but the input object itself.
678f13d5 243#
99870f4d
KW
244# The bool operator is deliberately not overloaded to avoid confusion with
245# "should it mean if the object merely exists, or also is non-empty?".
99870f4d
KW
246#
247# WHY CERTAIN DESIGN DECISIONS WERE MADE
678f13d5
KW
248#
249# This program needs to be able to run under miniperl. Therefore, it uses a
250# minimum of other modules, and hence implements some things itself that could
251# be gotten from CPAN
252#
253# This program uses inputs published by the Unicode Consortium. These can
254# change incompatibly between releases without the Perl maintainers realizing
255# it. Therefore this program is now designed to try to flag these. It looks
256# at the directories where the inputs are, and flags any unrecognized files.
257# It keeps track of all the properties in the files it handles, and flags any
258# that it doesn't know how to handle. It also flags any input lines that
259# don't match the expected syntax, among other checks.
260#
261# It is also designed so if a new input file matches one of the known
262# templates, one hopefully just needs to add it to a list to have it
263# processed.
264#
265# As mentioned earlier, some properties are given in more than one file. In
266# particular, the files in the extracted directory are supposedly just
267# reformattings of the others. But they contain information not easily
268# derivable from the other files, including results for Unihan, which this
269# program doesn't ordinarily look at, and for unassigned code points. They
270# also have historically had errors or been incomplete. In an attempt to
271# create the best possible data, this program thus processes them first to
272# glean information missing from the other files; then processes those other
273# files to override any errors in the extracted ones. Much of the design was
274# driven by this need to store things and then possibly override them.
275#
276# It tries to keep fatal errors to a minimum, to generate something usable for
277# testing purposes. It always looks for files that could be inputs, and will
278# warn about any that it doesn't know how to handle (the -q option suppresses
279# the warning).
99870f4d
KW
280#
281# Why have files written out for binary 'N' matches?
282# For binary properties, if you know the mapping for either Y or N; the
678f13d5
KW
283# other is trivial to construct, so could be done at Perl run-time by just
284# complementing the result, instead of having a file for it. That is, if
285# someone types in \p{foo: N}, Perl could translate that to \P{foo: Y} and
286# not need a file. The problem is communicating to Perl that a given
287# property is binary. Perl can't figure it out from looking at the N (or
288# No), as some non-binary properties have these as property values. So
289# rather than inventing a way to communicate this info back to the core,
290# which would have required changes there as well, it was simpler just to
291# add the extra tables.
292#
293# Why is there more than one type of range?
294# This simplified things. There are some very specialized code points that
295# have to be handled specially for output, such as Hangul syllable names.
296# By creating a range type (done late in the development process), it
297# allowed this to be stored with the range, and overridden by other input.
298# Originally these were stored in another data structure, and it became a
299# mess trying to decide if a second file that was for the same property was
300# overriding the earlier one or not.
301#
302# Why are there two kinds of tables, match and map?
303# (And there is a base class shared by the two as well.) As stated above,
304# they actually are for different things. Development proceeded much more
305# smoothly when I (khw) realized the distinction. Map tables are used to
306# give the property value for every code point (actually every code point
307# that doesn't map to a default value). Match tables are used for regular
308# expression matches, and are essentially the inverse mapping. Separating
309# the two allows more specialized methods, and error checks so that one
310# can't just take the intersection of two map tables, for example, as that
311# is nonsensical.
99870f4d
KW
312#
313# There are no match tables generated for matches of the null string. These
678f13d5
KW
314# would like like qr/\p{JSN=}/ currently without modifying the regex code.
315# Perhaps something like them could be added if necessary. The JSN does have
316# a real code point U+110B that maps to the null string, but it is a
317# contributory property, and therefore not output by default. And it's easily
318# handled so far by making the null string the default where it is a
319# possibility.
99870f4d 320#
23e33b60
KW
321# DEBUGGING
322#
678f13d5
KW
323# This program is written so it will run under miniperl. Occasionally changes
324# will cause an error where the backtrace doesn't work well under miniperl.
325# To diagnose the problem, you can instead run it under regular perl, if you
326# have one compiled.
327#
328# There is a good trace facility. To enable it, first sub DEBUG must be set
329# to return true. Then a line like
330#
331# local $to_trace = 1 if main::DEBUG;
332#
333# can be added to enable tracing in its lexical scope or until you insert
334# another line:
335#
336# local $to_trace = 0 if main::DEBUG;
337#
338# then use a line like "trace $a, @b, %c, ...;
339#
340# Some of the more complex subroutines already have trace statements in them.
341# Permanent trace statements should be like:
342#
343# trace ... if main::DEBUG && $to_trace;
344#
345# If there is just one or a few files that you're debugging, you can easily
346# cause most everything else to be skipped. Change the line
347#
348# my $debug_skip = 0;
349#
350# to 1, and every file whose object is in @input_file_objects and doesn't have
351# a, 'non_skip => 1,' in its constructor will be skipped.
352#
99870f4d
KW
353# FUTURE ISSUES
354#
355# The program would break if Unicode were to change its names so that
356# interior white space, underscores, or dashes differences were significant
357# within property and property value names.
358#
359# It might be easier to use the xml versions of the UCD if this program ever
360# would need heavy revision, and the ability to handle old versions was not
361# required.
362#
363# There is the potential for name collisions, in that Perl has chosen names
364# that Unicode could decide it also likes. There have been such collisions in
365# the past, with mostly Perl deciding to adopt the Unicode definition of the
366# name. However in the 5.2 Unicode beta testing, there were a number of such
367# collisions, which were withdrawn before the final release, because of Perl's
368# and other's protests. These all involved new properties which began with
369# 'Is'. Based on the protests, Unicode is unlikely to try that again. Also,
370# many of the Perl-defined synonyms, like Any, Word, etc, are listed in a
371# Unicode document, so they are unlikely to be used by Unicode for another
372# purpose. However, they might try something beginning with 'In', or use any
373# of the other Perl-defined properties. This program will warn you of name
374# collisions, and refuse to generate tables with them, but manual intervention
375# will be required in this event. One scheme that could be implemented, if
376# necessary, would be to have this program generate another file, or add a
377# field to mktables.lst that gives the date of first definition of a property.
378# Each new release of Unicode would use that file as a basis for the next
379# iteration. And the Perl synonym addition code could sort based on the age
380# of the property, so older properties get priority, and newer ones that clash
381# would be refused; hence existing code would not be impacted, and some other
382# synonym would have to be used for the new property. This is ugly, and
383# manual intervention would certainly be easier to do in the short run; lets
384# hope it never comes to this.
678f13d5 385#
99870f4d
KW
386# A NOTE ON UNIHAN
387#
388# This program can generate tables from the Unihan database. But it doesn't
389# by default, letting the CPAN module Unicode::Unihan handle them. Prior to
390# version 5.2, this database was in a single file, Unihan.txt. In 5.2 the
391# database was split into 8 different files, all beginning with the letters
392# 'Unihan'. This program will read those file(s) if present, but it needs to
393# know which of the many properties in the file(s) should have tables created
394# for them. It will create tables for any properties listed in
395# PropertyAliases.txt and PropValueAliases.txt, plus any listed in the
396# @cjk_properties array and the @cjk_property_values array. Thus, if a
397# property you want is not in those files of the release you are building
398# against, you must add it to those two arrays. Starting in 4.0, the
399# Unicode_Radical_Stroke was listed in those files, so if the Unihan database
400# is present in the directory, a table will be generated for that property.
401# In 5.2, several more properties were added. For your convenience, the two
402# arrays are initialized with all the 5.2 listed properties that are also in
403# earlier releases. But these are commented out. You can just uncomment the
404# ones you want, or use them as a template for adding entries for other
405# properties.
406#
407# You may need to adjust the entries to suit your purposes. setup_unihan(),
408# and filter_unihan_line() are the functions where this is done. This program
409# already does some adjusting to make the lines look more like the rest of the
410# Unicode DB; You can see what that is in filter_unihan_line()
411#
412# There is a bug in the 3.2 data file in which some values for the
413# kPrimaryNumeric property have commas and an unexpected comment. A filter
414# could be added for these; or for a particular installation, the Unihan.txt
415# file could be edited to fix them.
416# have to be
417#
678f13d5
KW
418# HOW TO ADD A FILE TO BE PROCESSED
419#
420# A new file from Unicode needs to have an object constructed for it in
421# @input_file_objects, probably at the end or at the end of the extracted
422# ones. The program should warn you if its name will clash with others on
423# restrictive file systems, like DOS. If so, figure out a better name, and
424# add lines to the README.perl file giving that. If the file is a character
425# property, it should be in the format that Unicode has by default
426# standardized for such files for the more recently introduced ones.
427# If so, the Input_file constructor for @input_file_objects can just be the
428# file name and release it first appeared in. If not, then it should be
429# possible to construct an each_line_handler() to massage the line into the
430# standardized form.
431#
432# For non-character properties, more code will be needed. You can look at
433# the existing entries for clues.
434#
435# UNICODE VERSIONS NOTES
436#
437# The Unicode UCD has had a number of errors in it over the versions. And
438# these remain, by policy, in the standard for that version. Therefore it is
439# risky to correct them, because code may be expecting the error. So this
440# program doesn't generally make changes, unless the error breaks the Perl
441# core. As an example, some versions of 2.1.x Jamo.txt have the wrong value
442# for U+1105, which causes real problems for the algorithms for Jamo
443# calculations, so it is changed here.
444#
445# But it isn't so clear cut as to what to do about concepts that are
446# introduced in a later release; should they extend back to earlier releases
447# where the concept just didn't exist? It was easier to do this than to not,
448# so that's what was done. For example, the default value for code points not
449# in the files for various properties was probably undefined until changed by
450# some version. No_Block for blocks is such an example. This program will
451# assign No_Block even in Unicode versions that didn't have it. This has the
452# benefit that code being written doesn't have to special case earlier
453# versions; and the detriment that it doesn't match the Standard precisely for
454# the affected versions.
455#
456# Here are some observations about some of the issues in early versions:
457#
458# The number of code points in \p{alpha} halve in 2.1.9. It turns out that
459# the reason is that the CJK block starting at 4E00 was removed from PropList,
460# and was not put back in until 3.1.0
461#
462# Unicode introduced the synonym Space for White_Space in 4.1. Perl has
463# always had a \p{Space}. In release 3.2 only, they are not synonymous. The
464# reason is that 3.2 introduced U+205F=medium math space, which was not
465# classed as white space, but Perl figured out that it should have been. 4.0
466# reclassified it correctly.
467#
468# Another change between 3.2 and 4.0 is the CCC property value ATBL. In 3.2
469# this was erroneously a synonym for 202. In 4.0, ATB became 202, and ATBL
470# was left with no code points, as all the ones that mapped to 202 stayed
471# mapped to 202. Thus if your program used the numeric name for the class,
472# it would not have been affected, but if it used the mnemonic, it would have
473# been.
474#
475# \p{Script=Hrkt} (Katakana_Or_Hiragana) came in 4.0.1. Before that code
476# points which eventually came to have this script property value, instead
477# mapped to "Unknown". But in the next release all these code points were
478# moved to \p{sc=common} instead.
99870f4d
KW
479#
480# The default for missing code points for BidiClass is complicated. Starting
481# in 3.1.1, the derived file DBidiClass.txt handles this, but this program
482# tries to do the best it can for earlier releases. It is done in
483# process_PropertyAliases()
484#
485##############################################################################
486
487my $UNDEF = ':UNDEF:'; # String to print out for undefined values in tracing
488 # and errors
489my $MAX_LINE_WIDTH = 78;
490
491# Debugging aid to skip most files so as to not be distracted by them when
492# concentrating on the ones being debugged. Add
493# non_skip => 1,
494# to the constructor for those files you want processed when you set this.
495# Files with a first version number of 0 are special: they are always
496# processed regardless of the state of this flag.
497my $debug_skip = 0;
498
499# Set to 1 to enable tracing.
500our $to_trace = 0;
501
502{ # Closure for trace: debugging aid
503 my $print_caller = 1; # ? Include calling subroutine name
504 my $main_with_colon = 'main::';
505 my $main_colon_length = length($main_with_colon);
506
507 sub trace {
508 return unless $to_trace; # Do nothing if global flag not set
509
510 my @input = @_;
511
512 local $DB::trace = 0;
513 $DB::trace = 0; # Quiet 'used only once' message
514
515 my $line_number;
516
517 # Loop looking up the stack to get the first non-trace caller
518 my $caller_line;
519 my $caller_name;
520 my $i = 0;
521 do {
522 $line_number = $caller_line;
523 (my $pkg, my $file, $caller_line, my $caller) = caller $i++;
524 $caller = $main_with_colon unless defined $caller;
525
526 $caller_name = $caller;
527
528 # get rid of pkg
529 $caller_name =~ s/.*:://;
530 if (substr($caller_name, 0, $main_colon_length)
531 eq $main_with_colon)
532 {
533 $caller_name = substr($caller_name, $main_colon_length);
534 }
535
536 } until ($caller_name ne 'trace');
537
538 # If the stack was empty, we were called from the top level
539 $caller_name = 'main' if ($caller_name eq ""
540 || $caller_name eq 'trace');
541
542 my $output = "";
543 foreach my $string (@input) {
544 #print STDERR __LINE__, ": ", join ", ", @input, "\n";
545 if (ref $string eq 'ARRAY' || ref $string eq 'HASH') {
546 $output .= simple_dumper($string);
547 }
548 else {
549 $string = "$string" if ref $string;
550 $string = $UNDEF unless defined $string;
551 chomp $string;
552 $string = '""' if $string eq "";
553 $output .= " " if $output ne ""
554 && $string ne ""
555 && substr($output, -1, 1) ne " "
556 && substr($string, 0, 1) ne " ";
557 $output .= $string;
558 }
559 }
560
99f78760
KW
561 print STDERR sprintf "%4d: ", $line_number if defined $line_number;
562 print STDERR "$caller_name: " if $print_caller;
99870f4d
KW
563 print STDERR $output, "\n";
564 return;
565 }
566}
567
568# This is for a rarely used development feature that allows you to compare two
569# versions of the Unicode standard without having to deal with changes caused
570# by the code points introduced in the later verson. Change the 0 to a SINGLE
571# dotted Unicode release number (e.g. 2.1). Only code points introduced in
572# that release and earlier will be used; later ones are thrown away. You use
573# the version number of the earliest one you want to compare; then run this
574# program on directory structures containing each release, and compare the
575# outputs. These outputs will therefore include only the code points common
576# to both releases, and you can see the changes caused just by the underlying
577# release semantic changes. For versions earlier than 3.2, you must copy a
578# version of DAge.txt into the directory.
579my $string_compare_versions = DEBUG && 0; # e.g., v2.1;
580my $compare_versions = DEBUG
581 && $string_compare_versions
582 && pack "C*", split /\./, $string_compare_versions;
583
584sub uniques {
585 # Returns non-duplicated input values. From "Perl Best Practices:
586 # Encapsulated Cleverness". p. 455 in first edition.
587
588 my %seen;
589 return grep { ! $seen{$_}++ } @_;
590}
591
592$0 = File::Spec->canonpath($0);
593
594my $make_test_script = 0; # ? Should we output a test script
595my $write_unchanged_files = 0; # ? Should we update the output files even if
596 # we don't think they have changed
597my $use_directory = ""; # ? Should we chdir somewhere.
598my $pod_directory; # input directory to store the pod file.
599my $pod_file = 'perluniprops';
600my $t_path; # Path to the .t test file
601my $file_list = 'mktables.lst'; # File to store input and output file names.
602 # This is used to speed up the build, by not
603 # executing the main body of the program if
604 # nothing on the list has changed since the
605 # previous build
606my $make_list = 1; # ? Should we write $file_list. Set to always
607 # make a list so that when the pumpking is
608 # preparing a release, s/he won't have to do
609 # special things
610my $glob_list = 0; # ? Should we try to include unknown .txt files
611 # in the input.
612my $output_range_counts = 1; # ? Should we include the number of code points
613 # in ranges in the output
9ef2b94f
KW
614my $output_names = 0; # ? Should character names be in the output
615my @viacode; # Contains the 1 million character names, if
616 # $output_names is true
617
99870f4d
KW
618# Verbosity levels; 0 is quiet
619my $NORMAL_VERBOSITY = 1;
620my $PROGRESS = 2;
621my $VERBOSE = 3;
622
623my $verbosity = $NORMAL_VERBOSITY;
624
625# Process arguments
626while (@ARGV) {
cf25bb62
JH
627 my $arg = shift @ARGV;
628 if ($arg eq '-v') {
99870f4d
KW
629 $verbosity = $VERBOSE;
630 }
631 elsif ($arg eq '-p') {
632 $verbosity = $PROGRESS;
633 $| = 1; # Flush buffers as we go.
634 }
635 elsif ($arg eq '-q') {
636 $verbosity = 0;
637 }
638 elsif ($arg eq '-w') {
639 $write_unchanged_files = 1; # update the files even if havent changed
640 }
641 elsif ($arg eq '-check') {
6ae7e459
YO
642 my $this = shift @ARGV;
643 my $ok = shift @ARGV;
644 if ($this ne $ok) {
645 print "Skipping as check params are not the same.\n";
646 exit(0);
647 }
00a8df5c 648 }
99870f4d
KW
649 elsif ($arg eq '-P' && defined ($pod_directory = shift)) {
650 -d $pod_directory or croak "Directory '$pod_directory' doesn't exist";
651 }
3df51b85
KW
652 elsif ($arg eq '-maketest' || ($arg eq '-T' && defined ($t_path = shift)))
653 {
99870f4d 654 $make_test_script = 1;
99870f4d
KW
655 }
656 elsif ($arg eq '-makelist') {
657 $make_list = 1;
658 }
659 elsif ($arg eq '-C' && defined ($use_directory = shift)) {
660 -d $use_directory or croak "Unknown directory '$use_directory'";
661 }
662 elsif ($arg eq '-L') {
663
664 # Existence not tested until have chdir'd
665 $file_list = shift;
666 }
667 elsif ($arg eq '-globlist') {
668 $glob_list = 1;
669 }
670 elsif ($arg eq '-c') {
671 $output_range_counts = ! $output_range_counts
672 }
9ef2b94f
KW
673 elsif ($arg eq '-output_names') {
674 $output_names = 1;
675 }
99870f4d
KW
676 else {
677 my $with_c = 'with';
678 $with_c .= 'out' if $output_range_counts; # Complements the state
679 croak <<END;
680usage: $0 [-c|-p|-q|-v|-w] [-C dir] [-L filelist] [ -P pod_dir ]
681 [ -T test_file_path ] [-globlist] [-makelist] [-maketest]
682 [-check A B ]
683 -c : Output comments $with_c number of code points in ranges
684 -q : Quiet Mode: Only output serious warnings.
685 -p : Set verbosity level to normal plus show progress.
686 -v : Set Verbosity level high: Show progress and non-serious
687 warnings
688 -w : Write files regardless
689 -C dir : Change to this directory before proceeding. All relative paths
690 except those specified by the -P and -T options will be done
691 with respect to this directory.
692 -P dir : Output $pod_file file to directory 'dir'.
3df51b85 693 -T path : Create a test script as 'path'; overrides -maketest
99870f4d
KW
694 -L filelist : Use alternate 'filelist' instead of standard one
695 -globlist : Take as input all non-Test *.txt files in current and sub
696 directories
3df51b85
KW
697 -maketest : Make test script 'TestProp.pl' in current (or -C directory),
698 overrides -T
99870f4d 699 -makelist : Rewrite the file list $file_list based on current setup
9ef2b94f
KW
700 -output_names : Output each character's name in the table files; useful for
701 doing what-ifs, looking at diffs; is slow, memory intensive,
702 resulting tables are usable but very large.
99870f4d
KW
703 -check A B : Executes $0 only if A and B are the same
704END
705 }
706}
707
708# Stores the most-recently changed file. If none have changed, can skip the
709# build
710my $youngest = -M $0; # Do this before the chdir!
711
712# Change directories now, because need to read 'version' early.
713if ($use_directory) {
3df51b85 714 if ($pod_directory && ! File::Spec->file_name_is_absolute($pod_directory)) {
99870f4d
KW
715 $pod_directory = File::Spec->rel2abs($pod_directory);
716 }
3df51b85 717 if ($t_path && ! File::Spec->file_name_is_absolute($t_path)) {
99870f4d 718 $t_path = File::Spec->rel2abs($t_path);
00a8df5c 719 }
99870f4d 720 chdir $use_directory or croak "Failed to chdir to '$use_directory':$!";
3df51b85 721 if ($pod_directory && File::Spec->file_name_is_absolute($pod_directory)) {
99870f4d 722 $pod_directory = File::Spec->abs2rel($pod_directory);
02b1aeec 723 }
3df51b85 724 if ($t_path && File::Spec->file_name_is_absolute($t_path)) {
99870f4d 725 $t_path = File::Spec->abs2rel($t_path);
02b1aeec 726 }
00a8df5c
YO
727}
728
99870f4d
KW
729# Get Unicode version into regular and v-string. This is done now because
730# various tables below get populated based on it. These tables are populated
731# here to be near the top of the file, and so easily seeable by those needing
732# to modify things.
733open my $VERSION, "<", "version"
734 or croak "$0: can't open required file 'version': $!\n";
735my $string_version = <$VERSION>;
736close $VERSION;
737chomp $string_version;
738my $v_version = pack "C*", split /\./, $string_version; # v string
739
740# The following are the complete names of properties with property values that
741# are known to not match any code points in some versions of Unicode, but that
742# may change in the future so they should be matchable, hence an empty file is
743# generated for them.
744my @tables_that_may_be_empty = (
745 'Joining_Type=Left_Joining',
746 );
747push @tables_that_may_be_empty, 'Script=Common' if $v_version le v4.0.1;
748push @tables_that_may_be_empty, 'Title' if $v_version lt v2.0.0;
749push @tables_that_may_be_empty, 'Script=Katakana_Or_Hiragana'
750 if $v_version ge v4.1.0;
751
752# The lists below are hashes, so the key is the item in the list, and the
753# value is the reason why it is in the list. This makes generation of
754# documentation easier.
755
756my %why_suppressed; # No file generated for these.
757
758# Files aren't generated for empty extraneous properties. This is arguable.
759# Extraneous properties generally come about because a property is no longer
760# used in a newer version of Unicode. If we generated a file without code
761# points, programs that used to work on that property will still execute
762# without errors. It just won't ever match (or will always match, with \P{}).
763# This means that the logic is now likely wrong. I (khw) think its better to
764# find this out by getting an error message. Just move them to the table
765# above to change this behavior
766my %why_suppress_if_empty_warn_if_not = (
767
768 # It is the only property that has ever officially been removed from the
769 # Standard. The database never contained any code points for it.
770 'Special_Case_Condition' => 'Obsolete',
771
772 # Apparently never official, but there were code points in some versions of
773 # old-style PropList.txt
774 'Non_Break' => 'Obsolete',
775);
776
777# These would normally go in the warn table just above, but they were changed
778# a long time before this program was written, so warnings about them are
779# moot.
780if ($v_version gt v3.2.0) {
781 push @tables_that_may_be_empty,
782 'Canonical_Combining_Class=Attached_Below_Left'
783}
784
785# These are listed in the Property aliases file in 5.2, but Unihan is ignored
786# unless explicitly added.
787if ($v_version ge v5.2.0) {
788 my $unihan = 'Unihan; remove from list if using Unihan';
23e33b60 789 foreach my $table qw (
99870f4d
KW
790 kAccountingNumeric
791 kOtherNumeric
792 kPrimaryNumeric
793 kCompatibilityVariant
794 kIICore
795 kIRG_GSource
796 kIRG_HSource
797 kIRG_JSource
798 kIRG_KPSource
799 kIRG_MSource
800 kIRG_KSource
801 kIRG_TSource
802 kIRG_USource
803 kIRG_VSource
804 kRSUnicode
805 )
806 {
807 $why_suppress_if_empty_warn_if_not{$table} = $unihan;
808 }
ca12659b
NC
809}
810
99870f4d
KW
811# Properties that this program ignores.
812my @unimplemented_properties = (
813'Unicode_Radical_Stroke' # Remove if changing to handle this one.
814);
d73e5302 815
99870f4d
KW
816# There are several types of obsolete properties defined by Unicode. These
817# must be hand-edited for every new Unicode release.
818my %why_deprecated; # Generates a deprecated warning message if used.
819my %why_stabilized; # Documentation only
820my %why_obsolete; # Documentation only
821
822{ # Closure
823 my $simple = 'Perl uses the more complete version of this property';
824 my $unihan = 'Unihan properties are by default not enabled in the Perl core. Instead use CPAN: Unicode::Unihan';
825
826 my $other_properties = 'other properties';
827 my $contributory = "Used by Unicode internally for generating $other_properties and not intended to be used stand-alone";
828 my $why_no_expand = "Easily computed, and yet doesn't cover the common encoding forms (UTF-16/8)",
829
830 %why_deprecated = (
831 'Grapheme_Link' => 'Deprecated by Unicode. Use ccc=vr (Canonical_Combining_Class=Virama) instead',
832 'Jamo_Short_Name' => $contributory,
833 'Line_Break=Surrogate' => 'Deprecated by Unicode because surrogates should never appear in well-formed text, and therefore shouldn\'t be the basis for line breaking',
834 'Other_Alphabetic' => $contributory,
835 'Other_Default_Ignorable_Code_Point' => $contributory,
836 'Other_Grapheme_Extend' => $contributory,
837 'Other_ID_Continue' => $contributory,
838 'Other_ID_Start' => $contributory,
839 'Other_Lowercase' => $contributory,
840 'Other_Math' => $contributory,
841 'Other_Uppercase' => $contributory,
842 );
843
844 %why_suppressed = (
845 # There is a lib/unicore/Decomposition.pl (used by normalize.pm) which
846 # contains the same information, but without the algorithmically
847 # determinable Hangul syllables'. This file is not published, so it's
848 # existence is not noted in the comment.
849 'Decomposition_Mapping' => 'Accessible via Unicode::Normalize',
850
851 'ISO_Comment' => 'Apparently no demand for it, but can access it through Unicode::UCD::charinfo. Obsoleted, and code points for it removed in Unicode 5.2',
852 'Unicode_1_Name' => "$simple, and no apparent demand for it, but can access it through Unicode::UCD::charinfo. If there is no later name for a code point, then this one is used instead in charnames",
853
854 'Simple_Case_Folding' => "$simple. Can access this through Unicode::UCD::casefold",
855 'Simple_Lowercase_Mapping' => "$simple. Can access this through Unicode::UCD::charinfo",
856 'Simple_Titlecase_Mapping' => "$simple. Can access this through Unicode::UCD::charinfo",
857 'Simple_Uppercase_Mapping' => "$simple. Can access this through Unicode::UCD::charinfo",
858
859 'Name' => "Accessible via 'use charnames;'",
860 'Name_Alias' => "Accessible via 'use charnames;'",
861
862 # These are sort of jumping the gun; deprecation is proposed for
863 # Unicode version 6.0, but they have never been exposed by Perl, and
864 # likely are soon to be deprecated, so best not to expose them.
865 FC_NFKC_Closure => 'Use NFKC_Casefold instead',
866 Expands_On_NFC => $why_no_expand,
867 Expands_On_NFD => $why_no_expand,
868 Expands_On_NFKC => $why_no_expand,
869 Expands_On_NFKD => $why_no_expand,
870 );
871
872 # The following are suppressed because they were made contributory or
873 # deprecated by Unicode before Perl ever thought about supporting them.
874 foreach my $property ('Jamo_Short_Name', 'Grapheme_Link') {
875 $why_suppressed{$property} = $why_deprecated{$property};
876 }
cf25bb62 877
99870f4d
KW
878 # Customize the message for all the 'Other_' properties
879 foreach my $property (keys %why_deprecated) {
880 next if (my $main_property = $property) !~ s/^Other_//;
881 $why_deprecated{$property} =~ s/$other_properties/the $main_property property (which should be used instead)/;
882 }
883}
884
885if ($v_version ge 4.0.0) {
886 $why_stabilized{'Hyphen'} = 'Use the Line_Break property instead; see www.unicode.org/reports/tr14';
887}
888if ($v_version ge 5.2.0) {
889 $why_obsolete{'ISO_Comment'} = 'Code points for it have been removed';
890}
891
892# Probably obsolete forever
893if ($v_version ge v4.1.0) {
894 $why_suppressed{'Script=Katakana_Or_Hiragana'} = 'Obsolete. All code points previously matched by this have been moved to "Script=Common"';
895}
896
897# This program can create files for enumerated-like properties, such as
898# 'Numeric_Type'. This file would be the same format as for a string
899# property, with a mapping from code point to its value, so you could look up,
900# for example, the script a code point is in. But no one so far wants this
901# mapping, or they have found another way to get it since this is a new
902# feature. So no file is generated except if it is in this list.
903my @output_mapped_properties = split "\n", <<END;
904END
905
906# If you are using the Unihan database, you need to add the properties that
907# you want to extract from it to this table. For your convenience, the
908# properties in the 5.2 PropertyAliases.txt file are listed, commented out
909my @cjk_properties = split "\n", <<'END';
910#cjkAccountingNumeric; kAccountingNumeric
911#cjkOtherNumeric; kOtherNumeric
912#cjkPrimaryNumeric; kPrimaryNumeric
913#cjkCompatibilityVariant; kCompatibilityVariant
914#cjkIICore ; kIICore
915#cjkIRG_GSource; kIRG_GSource
916#cjkIRG_HSource; kIRG_HSource
917#cjkIRG_JSource; kIRG_JSource
918#cjkIRG_KPSource; kIRG_KPSource
919#cjkIRG_KSource; kIRG_KSource
920#cjkIRG_TSource; kIRG_TSource
921#cjkIRG_USource; kIRG_USource
922#cjkIRG_VSource; kIRG_VSource
923#cjkRSUnicode; kRSUnicode ; Unicode_Radical_Stroke; URS
924END
925
926# Similarly for the property values. For your convenience, the lines in the
927# 5.2 PropertyAliases.txt file are listed. Just remove the first BUT NOT both
928# '#' marks
929my @cjk_property_values = split "\n", <<'END';
930## @missing: 0000..10FFFF; cjkAccountingNumeric; NaN
931## @missing: 0000..10FFFF; cjkCompatibilityVariant; <code point>
932## @missing: 0000..10FFFF; cjkIICore; <none>
933## @missing: 0000..10FFFF; cjkIRG_GSource; <none>
934## @missing: 0000..10FFFF; cjkIRG_HSource; <none>
935## @missing: 0000..10FFFF; cjkIRG_JSource; <none>
936## @missing: 0000..10FFFF; cjkIRG_KPSource; <none>
937## @missing: 0000..10FFFF; cjkIRG_KSource; <none>
938## @missing: 0000..10FFFF; cjkIRG_TSource; <none>
939## @missing: 0000..10FFFF; cjkIRG_USource; <none>
940## @missing: 0000..10FFFF; cjkIRG_VSource; <none>
941## @missing: 0000..10FFFF; cjkOtherNumeric; NaN
942## @missing: 0000..10FFFF; cjkPrimaryNumeric; NaN
943## @missing: 0000..10FFFF; cjkRSUnicode; <none>
944END
945
946# The input files don't list every code point. Those not listed are to be
947# defaulted to some value. Below are hard-coded what those values are for
948# non-binary properties as of 5.1. Starting in 5.0, there are
949# machine-parsable comment lines in the files the give the defaults; so this
950# list shouldn't have to be extended. The claim is that all missing entries
951# for binary properties will default to 'N'. Unicode tried to change that in
952# 5.2, but the beta period produced enough protest that they backed off.
953#
954# The defaults for the fields that appear in UnicodeData.txt in this hash must
955# be in the form that it expects. The others may be synonyms.
956my $CODE_POINT = '<code point>';
957my %default_mapping = (
958 Age => "Unassigned",
959 # Bidi_Class => Complicated; set in code
960 Bidi_Mirroring_Glyph => "",
961 Block => 'No_Block',
962 Canonical_Combining_Class => 0,
963 Case_Folding => $CODE_POINT,
964 Decomposition_Mapping => $CODE_POINT,
965 Decomposition_Type => 'None',
966 East_Asian_Width => "Neutral",
967 FC_NFKC_Closure => $CODE_POINT,
968 General_Category => 'Cn',
969 Grapheme_Cluster_Break => 'Other',
970 Hangul_Syllable_Type => 'NA',
971 ISO_Comment => "",
972 Jamo_Short_Name => "",
973 Joining_Group => "No_Joining_Group",
974 # Joining_Type => Complicated; set in code
975 kIICore => 'N', # Is converted to binary
976 #Line_Break => Complicated; set in code
977 Lowercase_Mapping => $CODE_POINT,
978 Name => "",
979 Name_Alias => "",
980 NFC_QC => 'Yes',
981 NFD_QC => 'Yes',
982 NFKC_QC => 'Yes',
983 NFKD_QC => 'Yes',
984 Numeric_Type => 'None',
985 Numeric_Value => 'NaN',
986 Script => ($v_version le 4.1.0) ? 'Common' : 'Unknown',
987 Sentence_Break => 'Other',
988 Simple_Case_Folding => $CODE_POINT,
989 Simple_Lowercase_Mapping => $CODE_POINT,
990 Simple_Titlecase_Mapping => $CODE_POINT,
991 Simple_Uppercase_Mapping => $CODE_POINT,
992 Titlecase_Mapping => $CODE_POINT,
993 Unicode_1_Name => "",
994 Unicode_Radical_Stroke => "",
995 Uppercase_Mapping => $CODE_POINT,
996 Word_Break => 'Other',
997);
998
999# Below are files that Unicode furnishes, but this program ignores, and why
1000my %ignored_files = (
1001 'CJKRadicals.txt' => 'Unihan data',
1002 'Index.txt' => 'An index, not actual data',
1003 'NamedSqProv.txt' => 'Not officially part of the Unicode standard; Append it to NamedSequences.txt if you want to process the contents.',
1004 'NamesList.txt' => 'Just adds commentary',
1005 'NormalizationCorrections.txt' => 'Data is already in other files.',
1006 'Props.txt' => 'Adds nothing to PropList.txt; only in very early releases',
1007 'ReadMe.txt' => 'Just comments',
1008 'README.TXT' => 'Just comments',
1009 'StandardizedVariants.txt' => 'Only for glyph changes, not a Unicode character property. Does not fit into current scheme where one code point is mapped',
1010);
1011
678f13d5 1012### End of externally interesting definitions, except for @input_file_objects
99870f4d
KW
1013
1014my $HEADER=<<"EOF";
1015# !!!!!!! DO NOT EDIT THIS FILE !!!!!!!
3df51b85
KW
1016# This file is machine-generated by $0 from the Unicode
1017# database, Version $string_version. Any changes made here will be lost!
cf25bb62
JH
1018EOF
1019
b6922eda 1020my $INTERNAL_ONLY=<<"EOF";
99870f4d
KW
1021
1022# !!!!!!! INTERNAL PERL USE ONLY !!!!!!!
b6922eda 1023# This file is for internal use by the Perl program only. The format and even
99870f4d
KW
1024# the name or existence of this file are subject to change without notice.
1025# Don't use it directly.
1026EOF
1027
1028my $DEVELOPMENT_ONLY=<<"EOF";
1029# !!!!!!! DEVELOPMENT USE ONLY !!!!!!!
1030# This file contains information artificially constrained to code points
1031# present in Unicode release $string_compare_versions.
1032# IT CANNOT BE RELIED ON. It is for use during development only and should
23e33b60 1033# not be used for production.
b6922eda
KW
1034
1035EOF
1036
99870f4d
KW
1037my $LAST_UNICODE_CODEPOINT_STRING = "10FFFF";
1038my $LAST_UNICODE_CODEPOINT = hex $LAST_UNICODE_CODEPOINT_STRING;
1039my $MAX_UNICODE_CODEPOINTS = $LAST_UNICODE_CODEPOINT + 1;
1040
1041# Matches legal code point. 4-6 hex numbers, If there are 6, the first
1042# two must be 10; if there are 5, the first must not be a 0. Written this way
1043# to decrease backtracking
1044my $code_point_re =
1045 qr/ \b (?: 10[0-9A-F]{4} | [1-9A-F][0-9A-F]{4} | [0-9A-F]{4} ) \b/x;
1046
1047# This matches the beginning of the line in the Unicode db files that give the
1048# defaults for code points not listed (i.e., missing) in the file. The code
1049# depends on this ending with a semi-colon, so it can assume it is a valid
1050# field when the line is split() by semi-colons
1051my $missing_defaults_prefix =
1052 qr/^#\s+\@missing:\s+0000\.\.$LAST_UNICODE_CODEPOINT_STRING\s*;/;
1053
1054# Property types. Unicode has more types, but these are sufficient for our
1055# purposes.
1056my $UNKNOWN = -1; # initialized to illegal value
1057my $NON_STRING = 1; # Either binary or enum
1058my $BINARY = 2;
1059my $ENUM = 3; # Include catalog
1060my $STRING = 4; # Anything else: string or misc
1061
1062# Some input files have lines that give default values for code points not
1063# contained in the file. Sometimes these should be ignored.
1064my $NO_DEFAULTS = 0; # Must evaluate to false
1065my $NOT_IGNORED = 1;
1066my $IGNORED = 2;
1067
1068# Range types. Each range has a type. Most ranges are type 0, for normal,
1069# and will appear in the main body of the tables in the output files, but
1070# there are other types of ranges as well, listed below, that are specially
1071# handled. There are pseudo-types as well that will never be stored as a
1072# type, but will affect the calculation of the type.
1073
1074# 0 is for normal, non-specials
1075my $MULTI_CP = 1; # Sequence of more than code point
1076my $HANGUL_SYLLABLE = 2;
1077my $CP_IN_NAME = 3; # The NAME contains the code point appended to it.
1078my $NULL = 4; # The map is to the null string; utf8.c can't
1079 # handle these, nor is there an accepted syntax
1080 # for them in \p{} constructs
f86864ac 1081my $COMPUTE_NO_MULTI_CP = 5; # Pseudo-type; means that ranges that would
99870f4d
KW
1082 # otherwise be $MULTI_CP type are instead type 0
1083
1084# process_generic_property_file() can accept certain overrides in its input.
1085# Each of these must begin AND end with $CMD_DELIM.
1086my $CMD_DELIM = "\a";
1087my $REPLACE_CMD = 'replace'; # Override the Replace
1088my $MAP_TYPE_CMD = 'map_type'; # Override the Type
1089
1090my $NO = 0;
1091my $YES = 1;
1092
1093# Values for the Replace argument to add_range.
1094# $NO # Don't replace; add only the code points not
1095 # already present.
1096my $IF_NOT_EQUIVALENT = 1; # Replace only under certain conditions; details in
1097 # the comments at the subroutine definition.
1098my $UNCONDITIONALLY = 2; # Replace without conditions.
1099my $MULTIPLE = 4; # Don't replace, but add a duplicate record if
1100 # already there
1101
1102# Flags to give property statuses. The phrases are to remind maintainers that
1103# if the flag is changed, the indefinite article referring to it in the
1104# documentation may need to be as well.
1105my $NORMAL = "";
1106my $SUPPRESSED = 'z'; # The character should never actually be seen, since
1107 # it is suppressed
37e2e78e 1108my $PLACEHOLDER = 'P'; # Implies no pod entry generated
99870f4d
KW
1109my $DEPRECATED = 'D';
1110my $a_bold_deprecated = "a 'B<$DEPRECATED>'";
1111my $A_bold_deprecated = "A 'B<$DEPRECATED>'";
1112my $DISCOURAGED = 'X';
1113my $a_bold_discouraged = "an 'B<$DISCOURAGED>'";
1114my $A_bold_discouraged = "An 'B<$DISCOURAGED>'";
1115my $STRICTER = 'T';
1116my $a_bold_stricter = "a 'B<$STRICTER>'";
1117my $A_bold_stricter = "A 'B<$STRICTER>'";
1118my $STABILIZED = 'S';
1119my $a_bold_stabilized = "an 'B<$STABILIZED>'";
1120my $A_bold_stabilized = "An 'B<$STABILIZED>'";
1121my $OBSOLETE = 'O';
1122my $a_bold_obsolete = "an 'B<$OBSOLETE>'";
1123my $A_bold_obsolete = "An 'B<$OBSOLETE>'";
1124
1125my %status_past_participles = (
1126 $DISCOURAGED => 'discouraged',
1127 $SUPPRESSED => 'should never be generated',
1128 $STABILIZED => 'stabilized',
1129 $OBSOLETE => 'obsolete',
37e2e78e 1130 $DEPRECATED => 'deprecated',
99870f4d
KW
1131);
1132
1133# The format of the values of the map tables:
1134my $BINARY_FORMAT = 'b';
1135my $DECIMAL_FORMAT = 'd';
1136my $FLOAT_FORMAT = 'f';
1137my $INTEGER_FORMAT = 'i';
1138my $HEX_FORMAT = 'x';
1139my $RATIONAL_FORMAT = 'r';
1140my $STRING_FORMAT = 's';
1141
1142my %map_table_formats = (
1143 $BINARY_FORMAT => 'binary',
1144 $DECIMAL_FORMAT => 'single decimal digit',
1145 $FLOAT_FORMAT => 'floating point number',
1146 $INTEGER_FORMAT => 'integer',
1147 $HEX_FORMAT => 'positive hex whole number; a code point',
1148 $RATIONAL_FORMAT => 'rational: an integer or a fraction',
1149 $STRING_FORMAT => 'arbitrary string',
1150);
1151
1152# Unicode didn't put such derived files in a separate directory at first.
1153my $EXTRACTED_DIR = (-d 'extracted') ? 'extracted' : "";
1154my $EXTRACTED = ($EXTRACTED_DIR) ? "$EXTRACTED_DIR/" : "";
1155my $AUXILIARY = 'auxiliary';
1156
1157# Hashes that will eventually go into Heavy.pl for the use of utf8_heavy.pl
1158my %loose_to_file_of; # loosely maps table names to their respective
1159 # files
1160my %stricter_to_file_of; # same; but for stricter mapping.
1161my %nv_floating_to_rational; # maps numeric values floating point numbers to
1162 # their rational equivalent
1163my %loose_property_name_of; # Loosely maps property names to standard form
1164
1165# These constants names and values were taken from the Unicode standard,
1166# version 5.1, section 3.12. They are used in conjunction with Hangul
1167# syllables
1168my $SBase = 0xAC00;
1169my $LBase = 0x1100;
1170my $VBase = 0x1161;
1171my $TBase = 0x11A7;
1172my $SCount = 11172;
1173my $LCount = 19;
1174my $VCount = 21;
1175my $TCount = 28;
1176my $NCount = $VCount * $TCount;
1177
1178# For Hangul syllables; These store the numbers from Jamo.txt in conjunction
1179# with the above published constants.
1180my %Jamo;
1181my %Jamo_L; # Leading consonants
1182my %Jamo_V; # Vowels
1183my %Jamo_T; # Trailing consonants
1184
37e2e78e 1185my @backslash_X_tests; # List of tests read in for testing \X
99870f4d
KW
1186my @unhandled_properties; # Will contain a list of properties found in
1187 # the input that we didn't process.
f86864ac 1188my @match_properties; # Properties that have match tables, to be
99870f4d
KW
1189 # listed in the pod
1190my @map_properties; # Properties that get map files written
1191my @named_sequences; # NamedSequences.txt contents.
1192my %potential_files; # Generated list of all .txt files in the directory
1193 # structure so we can warn if something is being
1194 # ignored.
1195my @files_actually_output; # List of files we generated.
1196my @more_Names; # Some code point names are compound; this is used
1197 # to store the extra components of them.
1198my $MIN_FRACTION_LENGTH = 3; # How many digits of a floating point number at
1199 # the minimum before we consider it equivalent to a
1200 # candidate rational
1201my $MAX_FLOATING_SLOP = 10 ** - $MIN_FRACTION_LENGTH; # And in floating terms
1202
1203# These store references to certain commonly used property objects
1204my $gc;
1205my $perl;
1206my $block;
1207
1208# Are there conflicting names because of beginning with 'In_', or 'Is_'
1209my $has_In_conflicts = 0;
1210my $has_Is_conflicts = 0;
1211
1212sub internal_file_to_platform ($) {
1213 # Convert our file paths which have '/' separators to those of the
1214 # platform.
1215
1216 my $file = shift;
1217 return undef unless defined $file;
1218
1219 return File::Spec->join(split '/', $file);
d07a55ed 1220}
5beb625e 1221
99870f4d
KW
1222sub file_exists ($) { # platform independent '-e'. This program internally
1223 # uses slash as a path separator.
1224 my $file = shift;
1225 return 0 if ! defined $file;
1226 return -e internal_file_to_platform($file);
1227}
5beb625e 1228
99870f4d 1229sub objaddr($) {
23e33b60
KW
1230 # Returns the address of the blessed input object.
1231 # It doesn't check for blessedness because that would do a string eval
1232 # every call, and the program is structured so that this is never called
1233 # for a non-blessed object.
99870f4d 1234
23e33b60 1235 no overloading; # If overloaded, numifying below won't work.
99870f4d
KW
1236
1237 # Numifying a ref gives its address.
051df77b 1238 return pack 'J', $_[0];
99870f4d
KW
1239}
1240
23e33b60
KW
1241# Commented code below should work on Perl 5.8.
1242## This 'require' doesn't necessarily work in miniperl, and even if it does,
1243## the native perl version of it (which is what would operate under miniperl)
1244## is extremely slow, as it does a string eval every call.
1245#my $has_fast_scalar_util = $\18 !~ /miniperl/
1246# && defined eval "require Scalar::Util";
1247#
1248#sub objaddr($) {
1249# # Returns the address of the blessed input object. Uses the XS version if
1250# # available. It doesn't check for blessedness because that would do a
1251# # string eval every call, and the program is structured so that this is
1252# # never called for a non-blessed object.
1253#
1254# return Scalar::Util::refaddr($_[0]) if $has_fast_scalar_util;
1255#
1256# # Check at least that is a ref.
1257# my $pkg = ref($_[0]) or return undef;
1258#
1259# # Change to a fake package to defeat any overloaded stringify
1260# bless $_[0], 'main::Fake';
1261#
1262# # Numifying a ref gives its address.
051df77b 1263# my $addr = pack 'J', $_[0];
23e33b60
KW
1264#
1265# # Return to original class
1266# bless $_[0], $pkg;
1267# return $addr;
1268#}
1269
99870f4d
KW
1270sub max ($$) {
1271 my $a = shift;
1272 my $b = shift;
1273 return $a if $a >= $b;
1274 return $b;
1275}
1276
1277sub min ($$) {
1278 my $a = shift;
1279 my $b = shift;
1280 return $a if $a <= $b;
1281 return $b;
1282}
1283
1284sub clarify_number ($) {
1285 # This returns the input number with underscores inserted every 3 digits
1286 # in large (5 digits or more) numbers. Input must be entirely digits, not
1287 # checked.
1288
1289 my $number = shift;
1290 my $pos = length($number) - 3;
1291 return $number if $pos <= 1;
1292 while ($pos > 0) {
1293 substr($number, $pos, 0) = '_';
1294 $pos -= 3;
5beb625e 1295 }
99870f4d 1296 return $number;
99598c8c
JH
1297}
1298
12ac2576 1299
99870f4d 1300package Carp;
7ebf06b3 1301
99870f4d
KW
1302# These routines give a uniform treatment of messages in this program. They
1303# are placed in the Carp package to cause the stack trace to not include them,
1304# although an alternative would be to use another package and set @CARP_NOT
1305# for it.
12ac2576 1306
99870f4d 1307our $Verbose = 1 if main::DEBUG; # Useful info when debugging
12ac2576 1308
99f78760
KW
1309# This is a work-around suggested by Nicholas Clark to fix a problem with Carp
1310# and overload trying to load Scalar:Util under miniperl. See
1311# http://www.xray.mpe.mpg.de/mailing-lists/perl5-porters/2009-11/msg01057.html
1312undef $overload::VERSION;
1313
99870f4d
KW
1314sub my_carp {
1315 my $message = shift || "";
1316 my $nofold = shift || 0;
7ebf06b3 1317
99870f4d
KW
1318 if ($message) {
1319 $message = main::join_lines($message);
1320 $message =~ s/^$0: *//; # Remove initial program name
1321 $message =~ s/[.;,]+$//; # Remove certain ending punctuation
1322 $message = "\n$0: $message;";
12ac2576 1323
99870f4d
KW
1324 # Fold the message with program name, semi-colon end punctuation
1325 # (which looks good with the message that carp appends to it), and a
1326 # hanging indent for continuation lines.
1327 $message = main::simple_fold($message, "", 4) unless $nofold;
1328 $message =~ s/\n$//; # Remove the trailing nl so what carp
1329 # appends is to the same line
1330 }
12ac2576 1331
99870f4d 1332 return $message if defined wantarray; # If a caller just wants the msg
12ac2576 1333
99870f4d
KW
1334 carp $message;
1335 return;
1336}
7ebf06b3 1337
99870f4d
KW
1338sub my_carp_bug {
1339 # This is called when it is clear that the problem is caused by a bug in
1340 # this program.
7ebf06b3 1341
99870f4d
KW
1342 my $message = shift;
1343 $message =~ s/^$0: *//;
1344 $message = my_carp("Bug in $0. Please report it by running perlbug or if that is unavailable, by sending email to perbug\@perl.org:\n$message");
1345 carp $message;
1346 return;
1347}
7ebf06b3 1348
99870f4d
KW
1349sub carp_too_few_args {
1350 if (@_ != 2) {
1351 my_carp_bug("Wrong number of arguments: to 'carp_too_few_arguments'. No action taken.");
1352 return;
12ac2576 1353 }
7ebf06b3 1354
99870f4d
KW
1355 my $args_ref = shift;
1356 my $count = shift;
7ebf06b3 1357
99870f4d
KW
1358 my_carp_bug("Need at least $count arguments to "
1359 . (caller 1)[3]
1360 . ". Instead got: '"
1361 . join ', ', @$args_ref
1362 . "'. No action taken.");
1363 return;
12ac2576
JP
1364}
1365
99870f4d
KW
1366sub carp_extra_args {
1367 my $args_ref = shift;
1368 my_carp_bug("Too many arguments to 'carp_extra_args': (" . join(', ', @_) . "); Extras ignored.") if @_;
12ac2576 1369
99870f4d
KW
1370 unless (ref $args_ref) {
1371 my_carp_bug("Argument to 'carp_extra_args' ($args_ref) must be a ref. Not checking arguments.");
1372 return;
1373 }
1374 my ($package, $file, $line) = caller;
1375 my $subroutine = (caller 1)[3];
cf25bb62 1376
99870f4d
KW
1377 my $list;
1378 if (ref $args_ref eq 'HASH') {
1379 foreach my $key (keys %$args_ref) {
1380 $args_ref->{$key} = $UNDEF unless defined $args_ref->{$key};
cf25bb62 1381 }
99870f4d 1382 $list = join ', ', each %{$args_ref};
cf25bb62 1383 }
99870f4d
KW
1384 elsif (ref $args_ref eq 'ARRAY') {
1385 foreach my $arg (@$args_ref) {
1386 $arg = $UNDEF unless defined $arg;
1387 }
1388 $list = join ', ', @$args_ref;
1389 }
1390 else {
1391 my_carp_bug("Can't cope with ref "
1392 . ref($args_ref)
1393 . " . argument to 'carp_extra_args'. Not checking arguments.");
1394 return;
1395 }
1396
1397 my_carp_bug("Unrecognized parameters in options: '$list' to $subroutine. Skipped.");
1398 return;
d73e5302
JH
1399}
1400
99870f4d
KW
1401package main;
1402
1403{ # Closure
1404
1405 # This program uses the inside-out method for objects, as recommended in
1406 # "Perl Best Practices". This closure aids in generating those. There
1407 # are two routines. setup_package() is called once per package to set
1408 # things up, and then set_access() is called for each hash representing a
1409 # field in the object. These routines arrange for the object to be
1410 # properly destroyed when no longer used, and for standard accessor
1411 # functions to be generated. If you need more complex accessors, just
1412 # write your own and leave those accesses out of the call to set_access().
1413 # More details below.
1414
1415 my %constructor_fields; # fields that are to be used in constructors; see
1416 # below
1417
1418 # The values of this hash will be the package names as keys to other
1419 # hashes containing the name of each field in the package as keys, and
1420 # references to their respective hashes as values.
1421 my %package_fields;
1422
1423 sub setup_package {
1424 # Sets up the package, creating standard DESTROY and dump methods
1425 # (unless already defined). The dump method is used in debugging by
1426 # simple_dumper().
1427 # The optional parameters are:
1428 # a) a reference to a hash, that gets populated by later
1429 # set_access() calls with one of the accesses being
1430 # 'constructor'. The caller can then refer to this, but it is
1431 # not otherwise used by these two routines.
1432 # b) a reference to a callback routine to call during destruction
1433 # of the object, before any fields are actually destroyed
1434
1435 my %args = @_;
1436 my $constructor_ref = delete $args{'Constructor_Fields'};
1437 my $destroy_callback = delete $args{'Destroy_Callback'};
1438 Carp::carp_extra_args(\@_) if main::DEBUG && %args;
1439
1440 my %fields;
1441 my $package = (caller)[0];
1442
1443 $package_fields{$package} = \%fields;
1444 $constructor_fields{$package} = $constructor_ref;
1445
1446 unless ($package->can('DESTROY')) {
1447 my $destroy_name = "${package}::DESTROY";
1448 no strict "refs";
1449
1450 # Use typeglob to give the anonymous subroutine the name we want
1451 *$destroy_name = sub {
1452 my $self = shift;
051df77b 1453 my $addr; { no overloading; $addr = pack 'J', $self; }
99870f4d
KW
1454
1455 $self->$destroy_callback if $destroy_callback;
1456 foreach my $field (keys %{$package_fields{$package}}) {
1457 #print STDERR __LINE__, ": Destroying ", ref $self, " ", sprintf("%04X", $addr), ": ", $field, "\n";
1458 delete $package_fields{$package}{$field}{$addr};
1459 }
1460 return;
1461 }
1462 }
1463
1464 unless ($package->can('dump')) {
1465 my $dump_name = "${package}::dump";
1466 no strict "refs";
1467 *$dump_name = sub {
1468 my $self = shift;
1469 return dump_inside_out($self, $package_fields{$package}, @_);
1470 }
1471 }
1472 return;
1473 }
1474
1475 sub set_access {
1476 # Arrange for the input field to be garbage collected when no longer
1477 # needed. Also, creates standard accessor functions for the field
1478 # based on the optional parameters-- none if none of these parameters:
1479 # 'addable' creates an 'add_NAME()' accessor function.
1480 # 'readable' or 'readable_array' creates a 'NAME()' accessor
1481 # function.
1482 # 'settable' creates a 'set_NAME()' accessor function.
1483 # 'constructor' doesn't create an accessor function, but adds the
1484 # field to the hash that was previously passed to
1485 # setup_package();
1486 # Any of the accesses can be abbreviated down, so that 'a', 'ad',
1487 # 'add' etc. all mean 'addable'.
1488 # The read accessor function will work on both array and scalar
1489 # values. If another accessor in the parameter list is 'a', the read
1490 # access assumes an array. You can also force it to be array access
1491 # by specifying 'readable_array' instead of 'readable'
1492 #
1493 # A sort-of 'protected' access can be set-up by preceding the addable,
1494 # readable or settable with some initial portion of 'protected_' (but,
1495 # the underscore is required), like 'p_a', 'pro_set', etc. The
1496 # "protection" is only by convention. All that happens is that the
1497 # accessor functions' names begin with an underscore. So instead of
1498 # calling set_foo, the call is _set_foo. (Real protection could be
1499 # accomplished by having a new subroutine, end_package called at the
1500 # end of each package, and then storing the __LINE__ ranges and
1501 # checking them on every accessor. But that is way overkill.)
1502
1503 # We create anonymous subroutines as the accessors and then use
1504 # typeglobs to assign them to the proper package and name
1505
1506 my $name = shift; # Name of the field
1507 my $field = shift; # Reference to the inside-out hash containing the
1508 # field
1509
1510 my $package = (caller)[0];
1511
1512 if (! exists $package_fields{$package}) {
1513 croak "$0: Must call 'setup_package' before 'set_access'";
1514 }
d73e5302 1515
99870f4d
KW
1516 # Stash the field so DESTROY can get it.
1517 $package_fields{$package}{$name} = $field;
cf25bb62 1518
99870f4d
KW
1519 # Remaining arguments are the accessors. For each...
1520 foreach my $access (@_) {
1521 my $access = lc $access;
cf25bb62 1522
99870f4d 1523 my $protected = "";
cf25bb62 1524
99870f4d
KW
1525 # Match the input as far as it goes.
1526 if ($access =~ /^(p[^_]*)_/) {
1527 $protected = $1;
1528 if (substr('protected_', 0, length $protected)
1529 eq $protected)
1530 {
1531
1532 # Add 1 for the underscore not included in $protected
1533 $access = substr($access, length($protected) + 1);
1534 $protected = '_';
1535 }
1536 else {
1537 $protected = "";
1538 }
1539 }
1540
1541 if (substr('addable', 0, length $access) eq $access) {
1542 my $subname = "${package}::${protected}add_$name";
1543 no strict "refs";
1544
1545 # add_ accessor. Don't add if already there, which we
1546 # determine using 'eq' for scalars and '==' otherwise.
1547 *$subname = sub {
1548 use strict "refs";
1549 return Carp::carp_too_few_args(\@_, 2) if main::DEBUG && @_ < 2;
1550 my $self = shift;
1551 my $value = shift;
051df77b 1552 my $addr; { no overloading; $addr = pack 'J', $self; }
99870f4d
KW
1553 Carp::carp_extra_args(\@_) if main::DEBUG && @_;
1554 if (ref $value) {
f998e60c 1555 return if grep { $value == $_ } @{$field->{$addr}};
99870f4d
KW
1556 }
1557 else {
f998e60c 1558 return if grep { $value eq $_ } @{$field->{$addr}};
99870f4d 1559 }
f998e60c 1560 push @{$field->{$addr}}, $value;
99870f4d
KW
1561 return;
1562 }
1563 }
1564 elsif (substr('constructor', 0, length $access) eq $access) {
1565 if ($protected) {
1566 Carp::my_carp_bug("Can't set-up 'protected' constructors")
1567 }
1568 else {
1569 $constructor_fields{$package}{$name} = $field;
1570 }
1571 }
1572 elsif (substr('readable_array', 0, length $access) eq $access) {
1573
1574 # Here has read access. If one of the other parameters for
1575 # access is array, or this one specifies array (by being more
1576 # than just 'readable_'), then create a subroutine that
1577 # assumes the data is an array. Otherwise just a scalar
1578 my $subname = "${package}::${protected}$name";
1579 if (grep { /^a/i } @_
1580 or length($access) > length('readable_'))
1581 {
1582 no strict "refs";
1583 *$subname = sub {
1584 use strict "refs";
23e33b60 1585 Carp::carp_extra_args(\@_) if main::DEBUG && @_ > 1;
051df77b 1586 my $addr; { no overloading; $addr = pack 'J', $_[0]; }
99870f4d
KW
1587 if (ref $field->{$addr} ne 'ARRAY') {
1588 my $type = ref $field->{$addr};
1589 $type = 'scalar' unless $type;
1590 Carp::my_carp_bug("Trying to read $name as an array when it is a $type. Big problems.");
1591 return;
1592 }
1593 return scalar @{$field->{$addr}} unless wantarray;
1594
1595 # Make a copy; had problems with caller modifying the
1596 # original otherwise
1597 my @return = @{$field->{$addr}};
1598 return @return;
1599 }
1600 }
1601 else {
1602
1603 # Here not an array value, a simpler function.
1604 no strict "refs";
1605 *$subname = sub {
1606 use strict "refs";
23e33b60 1607 Carp::carp_extra_args(\@_) if main::DEBUG && @_ > 1;
f998e60c 1608 no overloading;
051df77b 1609 return $field->{pack 'J', $_[0]};
99870f4d
KW
1610 }
1611 }
1612 }
1613 elsif (substr('settable', 0, length $access) eq $access) {
1614 my $subname = "${package}::${protected}set_$name";
1615 no strict "refs";
1616 *$subname = sub {
1617 use strict "refs";
23e33b60
KW
1618 if (main::DEBUG) {
1619 return Carp::carp_too_few_args(\@_, 2) if @_ < 2;
1620 Carp::carp_extra_args(\@_) if @_ > 2;
1621 }
1622 # $self is $_[0]; $value is $_[1]
f998e60c 1623 no overloading;
051df77b 1624 $field->{pack 'J', $_[0]} = $_[1];
99870f4d
KW
1625 return;
1626 }
1627 }
1628 else {
1629 Carp::my_carp_bug("Unknown accessor type $access. No accessor set.");
1630 }
cf25bb62 1631 }
99870f4d 1632 return;
cf25bb62 1633 }
99870f4d
KW
1634}
1635
1636package Input_file;
1637
1638# All input files use this object, which stores various attributes about them,
1639# and provides for convenient, uniform handling. The run method wraps the
1640# processing. It handles all the bookkeeping of opening, reading, and closing
1641# the file, returning only significant input lines.
1642#
1643# Each object gets a handler which processes the body of the file, and is
1644# called by run(). Most should use the generic, default handler, which has
1645# code scrubbed to handle things you might not expect. A handler should
1646# basically be a while(next_line()) {...} loop.
1647#
1648# You can also set up handlers to
1649# 1) call before the first line is read for pre processing
1650# 2) call to adjust each line of the input before the main handler gets them
1651# 3) call upon EOF before the main handler exits its loop
1652# 4) call at the end for post processing
1653#
1654# $_ is used to store the input line, and is to be filtered by the
1655# each_line_handler()s. So, if the format of the line is not in the desired
1656# format for the main handler, these are used to do that adjusting. They can
1657# be stacked (by enclosing them in an [ anonymous array ] in the constructor,
1658# so the $_ output of one is used as the input to the next. None of the other
1659# handlers are stackable, but could easily be changed to be so.
1660#
1661# Most of the handlers can call insert_lines() or insert_adjusted_lines()
1662# which insert the parameters as lines to be processed before the next input
1663# file line is read. This allows the EOF handler to flush buffers, for
1664# example. The difference between the two routines is that the lines inserted
1665# by insert_lines() are subjected to the each_line_handler()s. (So if you
1666# called it from such a handler, you would get infinite recursion.) Lines
1667# inserted by insert_adjusted_lines() go directly to the main handler without
1668# any adjustments. If the post-processing handler calls any of these, there
1669# will be no effect. Some error checking for these conditions could be added,
1670# but it hasn't been done.
1671#
1672# carp_bad_line() should be called to warn of bad input lines, which clears $_
1673# to prevent further processing of the line. This routine will output the
1674# message as a warning once, and then keep a count of the lines that have the
1675# same message, and output that count at the end of the file's processing.
1676# This keeps the number of messages down to a manageable amount.
1677#
1678# get_missings() should be called to retrieve any @missing input lines.
1679# Messages will be raised if this isn't done if the options aren't to ignore
1680# missings.
1681
1682sub trace { return main::trace(@_); }
1683
99870f4d
KW
1684{ # Closure
1685 # Keep track of fields that are to be put into the constructor.
1686 my %constructor_fields;
1687
1688 main::setup_package(Constructor_Fields => \%constructor_fields);
1689
1690 my %file; # Input file name, required
1691 main::set_access('file', \%file, qw{ c r });
1692
1693 my %first_released; # Unicode version file was first released in, required
1694 main::set_access('first_released', \%first_released, qw{ c r });
1695
1696 my %handler; # Subroutine to process the input file, defaults to
1697 # 'process_generic_property_file'
1698 main::set_access('handler', \%handler, qw{ c });
1699
1700 my %property;
1701 # name of property this file is for. defaults to none, meaning not
1702 # applicable, or is otherwise determinable, for example, from each line.
1703 main::set_access('property', \%property, qw{ c });
1704
1705 my %optional;
1706 # If this is true, the file is optional. If not present, no warning is
1707 # output. If it is present, the string given by this parameter is
1708 # evaluated, and if false the file is not processed.
1709 main::set_access('optional', \%optional, 'c', 'r');
1710
1711 my %non_skip;
1712 # This is used for debugging, to skip processing of all but a few input
1713 # files. Add 'non_skip => 1' to the constructor for those files you want
1714 # processed when you set the $debug_skip global.
1715 main::set_access('non_skip', \%non_skip, 'c');
1716
37e2e78e
KW
1717 my %skip;
1718 # This is used to skip processing of this input file semi-permanently.
1719 # It is used for files that we aren't planning to process anytime soon,
1720 # but want to allow to be in the directory and not raise a message that we
1721 # are not handling. Mostly for test files. This is in contrast to the
1722 # non_skip element, which is supposed to be used very temporarily for
1723 # debugging. Sets 'optional' to 1
1724 main::set_access('skip', \%skip, 'c');
1725
99870f4d
KW
1726 my %each_line_handler;
1727 # list of subroutines to look at and filter each non-comment line in the
1728 # file. defaults to none. The subroutines are called in order, each is
1729 # to adjust $_ for the next one, and the final one adjusts it for
1730 # 'handler'
1731 main::set_access('each_line_handler', \%each_line_handler, 'c');
1732
1733 my %has_missings_defaults;
1734 # ? Are there lines in the file giving default values for code points
1735 # missing from it?. Defaults to NO_DEFAULTS. Otherwise NOT_IGNORED is
1736 # the norm, but IGNORED means it has such lines, but the handler doesn't
1737 # use them. Having these three states allows us to catch changes to the
1738 # UCD that this program should track
1739 main::set_access('has_missings_defaults',
1740 \%has_missings_defaults, qw{ c r });
1741
1742 my %pre_handler;
1743 # Subroutine to call before doing anything else in the file. If undef, no
1744 # such handler is called.
1745 main::set_access('pre_handler', \%pre_handler, qw{ c });
1746
1747 my %eof_handler;
1748 # Subroutine to call upon getting an EOF on the input file, but before
1749 # that is returned to the main handler. This is to allow buffers to be
1750 # flushed. The handler is expected to call insert_lines() or
1751 # insert_adjusted() with the buffered material
1752 main::set_access('eof_handler', \%eof_handler, qw{ c r });
1753
1754 my %post_handler;
1755 # Subroutine to call after all the lines of the file are read in and
1756 # processed. If undef, no such handler is called.
1757 main::set_access('post_handler', \%post_handler, qw{ c });
1758
1759 my %progress_message;
1760 # Message to print to display progress in lieu of the standard one
1761 main::set_access('progress_message', \%progress_message, qw{ c });
1762
1763 my %handle;
1764 # cache open file handle, internal. Is undef if file hasn't been
1765 # processed at all, empty if has;
1766 main::set_access('handle', \%handle);
1767
1768 my %added_lines;
1769 # cache of lines added virtually to the file, internal
1770 main::set_access('added_lines', \%added_lines);
1771
1772 my %errors;
1773 # cache of errors found, internal
1774 main::set_access('errors', \%errors);
1775
1776 my %missings;
1777 # storage of '@missing' defaults lines
1778 main::set_access('missings', \%missings);
1779
1780 sub new {
1781 my $class = shift;
1782
1783 my $self = bless \do{ my $anonymous_scalar }, $class;
051df77b 1784 my $addr; { no overloading; $addr = pack 'J', $self; }
99870f4d
KW
1785
1786 # Set defaults
1787 $handler{$addr} = \&main::process_generic_property_file;
1788 $non_skip{$addr} = 0;
37e2e78e 1789 $skip{$addr} = 0;
99870f4d
KW
1790 $has_missings_defaults{$addr} = $NO_DEFAULTS;
1791 $handle{$addr} = undef;
1792 $added_lines{$addr} = [ ];
1793 $each_line_handler{$addr} = [ ];
1794 $errors{$addr} = { };
1795 $missings{$addr} = [ ];
1796
1797 # Two positional parameters.
99f78760 1798 return Carp::carp_too_few_args(\@_, 2) if main::DEBUG && @_ < 2;
99870f4d
KW
1799 $file{$addr} = main::internal_file_to_platform(shift);
1800 $first_released{$addr} = shift;
1801
1802 # The rest of the arguments are key => value pairs
1803 # %constructor_fields has been set up earlier to list all possible
1804 # ones. Either set or push, depending on how the default has been set
1805 # up just above.
1806 my %args = @_;
1807 foreach my $key (keys %args) {
1808 my $argument = $args{$key};
1809
1810 # Note that the fields are the lower case of the constructor keys
1811 my $hash = $constructor_fields{lc $key};
1812 if (! defined $hash) {
1813 Carp::my_carp_bug("Unrecognized parameters '$key => $argument' to new() for $self. Skipped");
1814 next;
1815 }
1816 if (ref $hash->{$addr} eq 'ARRAY') {
1817 if (ref $argument eq 'ARRAY') {
1818 foreach my $argument (@{$argument}) {
1819 next if ! defined $argument;
1820 push @{$hash->{$addr}}, $argument;
1821 }
1822 }
1823 else {
1824 push @{$hash->{$addr}}, $argument if defined $argument;
1825 }
1826 }
1827 else {
1828 $hash->{$addr} = $argument;
1829 }
1830 delete $args{$key};
1831 };
1832
1833 # If the file has a property for it, it means that the property is not
1834 # listed in the file's entries. So add a handler to the list of line
1835 # handlers to insert the property name into the lines, to provide a
1836 # uniform interface to the final processing subroutine.
1837 # the final code doesn't have to worry about that.
1838 if ($property{$addr}) {
1839 push @{$each_line_handler{$addr}}, \&_insert_property_into_line;
1840 }
1841
1842 if ($non_skip{$addr} && ! $debug_skip && $verbosity) {
1843 print "Warning: " . __PACKAGE__ . " constructor for $file{$addr} has useless 'non_skip' in it\n";
a3a8c5f0 1844 }
99870f4d 1845
37e2e78e
KW
1846 $optional{$addr} = 1 if $skip{$addr};
1847
99870f4d 1848 return $self;
d73e5302
JH
1849 }
1850
cf25bb62 1851
99870f4d
KW
1852 use overload
1853 fallback => 0,
1854 qw("") => "_operator_stringify",
1855 "." => \&main::_operator_dot,
1856 ;
cf25bb62 1857
99870f4d
KW
1858 sub _operator_stringify {
1859 my $self = shift;
cf25bb62 1860
99870f4d 1861 return __PACKAGE__ . " object for " . $self->file;
d73e5302 1862 }
d73e5302 1863
99870f4d
KW
1864 # flag to make sure extracted files are processed early
1865 my $seen_non_extracted_non_age = 0;
d73e5302 1866
99870f4d
KW
1867 sub run {
1868 # Process the input object $self. This opens and closes the file and
1869 # calls all the handlers for it. Currently, this can only be called
1870 # once per file, as it destroy's the EOF handler
d73e5302 1871
99870f4d
KW
1872 my $self = shift;
1873 Carp::carp_extra_args(\@_) if main::DEBUG && @_;
b6922eda 1874
051df77b 1875 my $addr; { no overloading; $addr = pack 'J', $self; }
b6922eda 1876
99870f4d 1877 my $file = $file{$addr};
d73e5302 1878
99870f4d
KW
1879 # Don't process if not expecting this file (because released later
1880 # than this Unicode version), and isn't there. This means if someone
1881 # copies it into an earlier version's directory, we will go ahead and
1882 # process it.
1883 return if $first_released{$addr} gt $v_version && ! -e $file;
1884
1885 # If in debugging mode and this file doesn't have the non-skip
1886 # flag set, and isn't one of the critical files, skip it.
1887 if ($debug_skip
1888 && $first_released{$addr} ne v0
1889 && ! $non_skip{$addr})
1890 {
1891 print "Skipping $file in debugging\n" if $verbosity;
1892 return;
1893 }
1894
1895 # File could be optional
37e2e78e 1896 if ($optional{$addr}) {
99870f4d
KW
1897 return unless -e $file;
1898 my $result = eval $optional{$addr};
1899 if (! defined $result) {
1900 Carp::my_carp_bug("Got '$@' when tried to eval $optional{$addr}. $file Skipped.");
1901 return;
1902 }
1903 if (! $result) {
1904 if ($verbosity) {
1905 print STDERR "Skipping processing input file '$file' because '$optional{$addr}' is not true\n";
1906 }
1907 return;
1908 }
1909 }
1910
1911 if (! defined $file || ! -e $file) {
1912
1913 # If the file doesn't exist, see if have internal data for it
1914 # (based on first_released being 0).
1915 if ($first_released{$addr} eq v0) {
1916 $handle{$addr} = 'pretend_is_open';
1917 }
1918 else {
1919 if (! $optional{$addr} # File could be optional
1920 && $v_version ge $first_released{$addr})
1921 {
1922 print STDERR "Skipping processing input file '$file' because not found\n" if $v_version ge $first_released{$addr};
1923 }
1924 return;
1925 }
1926 }
1927 else {
1928
37e2e78e
KW
1929 # Here, the file exists. Some platforms may change the case of
1930 # its name
99870f4d 1931 if ($seen_non_extracted_non_age) {
517956bf 1932 if ($file =~ /$EXTRACTED/i) {
99870f4d 1933 Carp::my_carp_bug(join_lines(<<END
99f78760 1934$file should be processed just after the 'Prop...Alias' files, and before
99870f4d
KW
1935anything not in the $EXTRACTED_DIR directory. Proceeding, but the results may
1936have subtle problems
1937END
1938 ));
1939 }
1940 }
1941 elsif ($EXTRACTED_DIR
1942 && $first_released{$addr} ne v0
517956bf
CB
1943 && $file !~ /$EXTRACTED/i
1944 && lc($file) ne 'dage.txt')
99870f4d
KW
1945 {
1946 # We don't set this (by the 'if' above) if we have no
1947 # extracted directory, so if running on an early version,
1948 # this test won't work. Not worth worrying about.
1949 $seen_non_extracted_non_age = 1;
1950 }
1951
1952 # And mark the file as having being processed, and warn if it
1953 # isn't a file we are expecting. As we process the files,
1954 # they are deleted from the hash, so any that remain at the
1955 # end of the program are files that we didn't process.
517956bf
CB
1956 my $fkey = File::Spec->rel2abs($file);
1957 my $expecting = delete $potential_files{$fkey};
1958 $expecting = delete $potential_files{lc($fkey)} unless defined $expecting;
678f13d5
KW
1959 Carp::my_carp("Was not expecting '$file'.") if
1960 ! $expecting
99870f4d
KW
1961 && ! defined $handle{$addr};
1962
37e2e78e
KW
1963 # Having deleted from expected files, we can quit if not to do
1964 # anything. Don't print progress unless really want verbosity
1965 if ($skip{$addr}) {
1966 print "Skipping $file.\n" if $verbosity >= $VERBOSE;
1967 return;
1968 }
1969
99870f4d
KW
1970 # Open the file, converting the slashes used in this program
1971 # into the proper form for the OS
1972 my $file_handle;
1973 if (not open $file_handle, "<", $file) {
1974 Carp::my_carp("Can't open $file. Skipping: $!");
1975 return 0;
1976 }
1977 $handle{$addr} = $file_handle; # Cache the open file handle
1978 }
1979
1980 if ($verbosity >= $PROGRESS) {
1981 if ($progress_message{$addr}) {
1982 print "$progress_message{$addr}\n";
1983 }
1984 else {
1985 # If using a virtual file, say so.
1986 print "Processing ", (-e $file)
1987 ? $file
1988 : "substitute $file",
1989 "\n";
1990 }
1991 }
1992
1993
1994 # Call any special handler for before the file.
1995 &{$pre_handler{$addr}}($self) if $pre_handler{$addr};
1996
1997 # Then the main handler
1998 &{$handler{$addr}}($self);
1999
2000 # Then any special post-file handler.
2001 &{$post_handler{$addr}}($self) if $post_handler{$addr};
2002
2003 # If any errors have been accumulated, output the counts (as the first
2004 # error message in each class was output when it was encountered).
2005 if ($errors{$addr}) {
2006 my $total = 0;
2007 my $types = 0;
2008 foreach my $error (keys %{$errors{$addr}}) {
2009 $total += $errors{$addr}->{$error};
2010 delete $errors{$addr}->{$error};
2011 $types++;
2012 }
2013 if ($total > 1) {
2014 my $message
2015 = "A total of $total lines had errors in $file. ";
2016
2017 $message .= ($types == 1)
2018 ? '(Only the first one was displayed.)'
2019 : '(Only the first of each type was displayed.)';
2020 Carp::my_carp($message);
2021 }
2022 }
2023
2024 if (@{$missings{$addr}}) {
2025 Carp::my_carp_bug("Handler for $file didn't look at all the \@missing lines. Generated tables likely are wrong");
2026 }
2027
2028 # If a real file handle, close it.
2029 close $handle{$addr} or Carp::my_carp("Can't close $file: $!") if
2030 ref $handle{$addr};
2031 $handle{$addr} = ""; # Uses empty to indicate that has already seen
2032 # the file, as opposed to undef
2033 return;
2034 }
2035
2036 sub next_line {
2037 # Sets $_ to be the next logical input line, if any. Returns non-zero
2038 # if such a line exists. 'logical' means that any lines that have
2039 # been added via insert_lines() will be returned in $_ before the file
2040 # is read again.
2041
2042 my $self = shift;
2043 Carp::carp_extra_args(\@_) if main::DEBUG && @_;
2044
051df77b 2045 my $addr; { no overloading; $addr = pack 'J', $self; }
99870f4d
KW
2046
2047 # Here the file is open (or if the handle is not a ref, is an open
2048 # 'virtual' file). Get the next line; any inserted lines get priority
2049 # over the file itself.
2050 my $adjusted;
2051
2052 LINE:
2053 while (1) { # Loop until find non-comment, non-empty line
2054 #local $to_trace = 1 if main::DEBUG;
2055 my $inserted_ref = shift @{$added_lines{$addr}};
2056 if (defined $inserted_ref) {
2057 ($adjusted, $_) = @{$inserted_ref};
2058 trace $adjusted, $_ if main::DEBUG && $to_trace;
2059 return 1 if $adjusted;
2060 }
2061 else {
2062 last if ! ref $handle{$addr}; # Don't read unless is real file
2063 last if ! defined ($_ = readline $handle{$addr});
2064 }
2065 chomp;
2066 trace $_ if main::DEBUG && $to_trace;
2067
2068 # See if this line is the comment line that defines what property
2069 # value that code points that are not listed in the file should
2070 # have. The format or existence of these lines is not guaranteed
2071 # by Unicode since they are comments, but the documentation says
2072 # that this was added for machine-readability, so probably won't
2073 # change. This works starting in Unicode Version 5.0. They look
2074 # like:
2075 #
2076 # @missing: 0000..10FFFF; Not_Reordered
2077 # @missing: 0000..10FFFF; Decomposition_Mapping; <code point>
2078 # @missing: 0000..10FFFF; ; NaN
2079 #
2080 # Save the line for a later get_missings() call.
2081 if (/$missing_defaults_prefix/) {
2082 if ($has_missings_defaults{$addr} == $NO_DEFAULTS) {
2083 $self->carp_bad_line("Unexpected \@missing line. Assuming no missing entries");
2084 }
2085 elsif ($has_missings_defaults{$addr} == $NOT_IGNORED) {
2086 my @defaults = split /\s* ; \s*/x, $_;
2087
2088 # The first field is the @missing, which ends in a
2089 # semi-colon, so can safely shift.
2090 shift @defaults;
2091
2092 # Some of these lines may have empty field placeholders
2093 # which get in the way. An example is:
2094 # @missing: 0000..10FFFF; ; NaN
2095 # Remove them. Process starting from the top so the
2096 # splice doesn't affect things still to be looked at.
2097 for (my $i = @defaults - 1; $i >= 0; $i--) {
2098 next if $defaults[$i] ne "";
2099 splice @defaults, $i, 1;
2100 }
2101
2102 # What's left should be just the property (maybe) and the
2103 # default. Having only one element means it doesn't have
2104 # the property.
2105 my $default;
2106 my $property;
2107 if (@defaults >= 1) {
2108 if (@defaults == 1) {
2109 $default = $defaults[0];
2110 }
2111 else {
2112 $property = $defaults[0];
2113 $default = $defaults[1];
2114 }
2115 }
2116
2117 if (@defaults < 1
2118 || @defaults > 2
2119 || ($default =~ /^</
2120 && $default !~ /^<code *point>$/i
2121 && $default !~ /^<none>$/i))
2122 {
2123 $self->carp_bad_line("Unrecognized \@missing line: $_. Assuming no missing entries");
2124 }
2125 else {
2126
2127 # If the property is missing from the line, it should
2128 # be the one for the whole file
2129 $property = $property{$addr} if ! defined $property;
2130
2131 # Change <none> to the null string, which is what it
2132 # really means. If the default is the code point
2133 # itself, set it to <code point>, which is what
2134 # Unicode uses (but sometimes they've forgotten the
2135 # space)
2136 if ($default =~ /^<none>$/i) {
2137 $default = "";
2138 }
2139 elsif ($default =~ /^<code *point>$/i) {
2140 $default = $CODE_POINT;
2141 }
2142
2143 # Store them as a sub-arrays with both components.
2144 push @{$missings{$addr}}, [ $default, $property ];
2145 }
2146 }
2147
2148 # There is nothing for the caller to process on this comment
2149 # line.
2150 next;
2151 }
2152
2153 # Remove comments and trailing space, and skip this line if the
2154 # result is empty
2155 s/#.*//;
2156 s/\s+$//;
2157 next if /^$/;
2158
2159 # Call any handlers for this line, and skip further processing of
2160 # the line if the handler sets the line to null.
2161 foreach my $sub_ref (@{$each_line_handler{$addr}}) {
2162 &{$sub_ref}($self);
2163 next LINE if /^$/;
2164 }
2165
2166 # Here the line is ok. return success.
2167 return 1;
2168 } # End of looping through lines.
2169
2170 # If there is an EOF handler, call it (only once) and if it generates
2171 # more lines to process go back in the loop to handle them.
2172 if ($eof_handler{$addr}) {
2173 &{$eof_handler{$addr}}($self);
2174 $eof_handler{$addr} = ""; # Currently only get one shot at it.
2175 goto LINE if $added_lines{$addr};
2176 }
2177
2178 # Return failure -- no more lines.
2179 return 0;
2180
2181 }
2182
2183# Not currently used, not fully tested.
2184# sub peek {
2185# # Non-destructive look-ahead one non-adjusted, non-comment, non-blank
2186# # record. Not callable from an each_line_handler(), nor does it call
2187# # an each_line_handler() on the line.
2188#
2189# my $self = shift;
051df77b 2190# my $addr; { no overloading; $addr = pack 'J', $self; }
99870f4d
KW
2191#
2192# foreach my $inserted_ref (@{$added_lines{$addr}}) {
2193# my ($adjusted, $line) = @{$inserted_ref};
2194# next if $adjusted;
2195#
2196# # Remove comments and trailing space, and return a non-empty
2197# # resulting line
2198# $line =~ s/#.*//;
2199# $line =~ s/\s+$//;
2200# return $line if $line ne "";
2201# }
2202#
2203# return if ! ref $handle{$addr}; # Don't read unless is real file
2204# while (1) { # Loop until find non-comment, non-empty line
2205# local $to_trace = 1 if main::DEBUG;
2206# trace $_ if main::DEBUG && $to_trace;
2207# return if ! defined (my $line = readline $handle{$addr});
2208# chomp $line;
2209# push @{$added_lines{$addr}}, [ 0, $line ];
2210#
2211# $line =~ s/#.*//;
2212# $line =~ s/\s+$//;
2213# return $line if $line ne "";
2214# }
2215#
2216# return;
2217# }
2218
2219
2220 sub insert_lines {
2221 # Lines can be inserted so that it looks like they were in the input
2222 # file at the place it was when this routine is called. See also
2223 # insert_adjusted_lines(). Lines inserted via this routine go through
2224 # any each_line_handler()
2225
2226 my $self = shift;
2227
2228 # Each inserted line is an array, with the first element being 0 to
2229 # indicate that this line hasn't been adjusted, and needs to be
2230 # processed.
f998e60c 2231 no overloading;
051df77b 2232 push @{$added_lines{pack 'J', $self}}, map { [ 0, $_ ] } @_;
99870f4d
KW
2233 return;
2234 }
2235
2236 sub insert_adjusted_lines {
2237 # Lines can be inserted so that it looks like they were in the input
2238 # file at the place it was when this routine is called. See also
2239 # insert_lines(). Lines inserted via this routine are already fully
2240 # adjusted, ready to be processed; each_line_handler()s handlers will
2241 # not be called. This means this is not a completely general
2242 # facility, as only the last each_line_handler on the stack should
2243 # call this. It could be made more general, by passing to each of the
2244 # line_handlers their position on the stack, which they would pass on
2245 # to this routine, and that would replace the boolean first element in
2246 # the anonymous array pushed here, so that the next_line routine could
2247 # use that to call only those handlers whose index is after it on the
2248 # stack. But this is overkill for what is needed now.
2249
2250 my $self = shift;
2251 trace $_[0] if main::DEBUG && $to_trace;
2252
2253 # Each inserted line is an array, with the first element being 1 to
2254 # indicate that this line has been adjusted
f998e60c 2255 no overloading;
051df77b 2256 push @{$added_lines{pack 'J', $self}}, map { [ 1, $_ ] } @_;
99870f4d
KW
2257 return;
2258 }
2259
2260 sub get_missings {
2261 # Returns the stored up @missings lines' values, and clears the list.
2262 # The values are in an array, consisting of the default in the first
2263 # element, and the property in the 2nd. However, since these lines
2264 # can be stacked up, the return is an array of all these arrays.
2265
2266 my $self = shift;
2267 Carp::carp_extra_args(\@_) if main::DEBUG && @_;
2268
051df77b 2269 my $addr; { no overloading; $addr = pack 'J', $self; }
99870f4d
KW
2270
2271 # If not accepting a list return, just return the first one.
2272 return shift @{$missings{$addr}} unless wantarray;
2273
2274 my @return = @{$missings{$addr}};
2275 undef @{$missings{$addr}};
2276 return @return;
2277 }
2278
2279 sub _insert_property_into_line {
2280 # Add a property field to $_, if this file requires it.
2281
f998e60c 2282 my $self = shift;
051df77b 2283 my $addr; { no overloading; $addr = pack 'J', $self; }
f998e60c 2284 my $property = $property{$addr};
99870f4d
KW
2285 Carp::carp_extra_args(\@_) if main::DEBUG && @_;
2286
2287 $_ =~ s/(;|$)/; $property$1/;
2288 return;
2289 }
2290
2291 sub carp_bad_line {
2292 # Output consistent error messages, using either a generic one, or the
2293 # one given by the optional parameter. To avoid gazillions of the
2294 # same message in case the syntax of a file is way off, this routine
2295 # only outputs the first instance of each message, incrementing a
2296 # count so the totals can be output at the end of the file.
2297
2298 my $self = shift;
2299 my $message = shift;
2300 Carp::carp_extra_args(\@_) if main::DEBUG && @_;
2301
051df77b 2302 my $addr; { no overloading; $addr = pack 'J', $self; }
99870f4d
KW
2303
2304 $message = 'Unexpected line' unless $message;
2305
2306 # No trailing punctuation so as to fit with our addenda.
2307 $message =~ s/[.:;,]$//;
2308
2309 # If haven't seen this exact message before, output it now. Otherwise
2310 # increment the count of how many times it has occurred
2311 unless ($errors{$addr}->{$message}) {
2312 Carp::my_carp("$message in '$_' in "
f998e60c 2313 . $file{$addr}
99870f4d
KW
2314 . " at line $.. Skipping this line;");
2315 $errors{$addr}->{$message} = 1;
2316 }
2317 else {
2318 $errors{$addr}->{$message}++;
2319 }
2320
2321 # Clear the line to prevent any further (meaningful) processing of it.
2322 $_ = "";
2323
2324 return;
2325 }
2326} # End closure
2327
2328package Multi_Default;
2329
2330# Certain properties in early versions of Unicode had more than one possible
2331# default for code points missing from the files. In these cases, one
2332# default applies to everything left over after all the others are applied,
2333# and for each of the others, there is a description of which class of code
2334# points applies to it. This object helps implement this by storing the
2335# defaults, and for all but that final default, an eval string that generates
2336# the class that it applies to.
2337
2338
2339{ # Closure
2340
2341 main::setup_package();
2342
2343 my %class_defaults;
2344 # The defaults structure for the classes
2345 main::set_access('class_defaults', \%class_defaults);
2346
2347 my %other_default;
2348 # The default that applies to everything left over.
2349 main::set_access('other_default', \%other_default, 'r');
2350
2351
2352 sub new {
2353 # The constructor is called with default => eval pairs, terminated by
2354 # the left-over default. e.g.
2355 # Multi_Default->new(
2356 # 'T' => '$gc->table("Mn") + $gc->table("Cf") - 0x200C
2357 # - 0x200D',
2358 # 'R' => 'some other expression that evaluates to code points',
2359 # .
2360 # .
2361 # .
2362 # 'U'));
2363
2364 my $class = shift;
2365
2366 my $self = bless \do{my $anonymous_scalar}, $class;
051df77b 2367 my $addr; { no overloading; $addr = pack 'J', $self; }
99870f4d
KW
2368
2369 while (@_ > 1) {
2370 my $default = shift;
2371 my $eval = shift;
2372 $class_defaults{$addr}->{$default} = $eval;
2373 }
2374
2375 $other_default{$addr} = shift;
2376
2377 return $self;
2378 }
2379
2380 sub get_next_defaults {
2381 # Iterates and returns the next class of defaults.
2382 my $self = shift;
2383 Carp::carp_extra_args(\@_) if main::DEBUG && @_;
2384
051df77b 2385 my $addr; { no overloading; $addr = pack 'J', $self; }
99870f4d
KW
2386
2387 return each %{$class_defaults{$addr}};
2388 }
2389}
2390
2391package Alias;
2392
2393# An alias is one of the names that a table goes by. This class defines them
2394# including some attributes. Everything is currently setup in the
2395# constructor.
2396
2397
2398{ # Closure
2399
2400 main::setup_package();
2401
2402 my %name;
2403 main::set_access('name', \%name, 'r');
2404
2405 my %loose_match;
2406 # Determined by the constructor code if this name should match loosely or
2407 # not. The constructor parameters can override this, but it isn't fully
2408 # implemented, as should have ability to override Unicode one's via
2409 # something like a set_loose_match()
2410 main::set_access('loose_match', \%loose_match, 'r');
2411
2412 my %make_pod_entry;
2413 # Some aliases should not get their own entries because they are covered
2414 # by a wild-card, and some we want to discourage use of. Binary
2415 main::set_access('make_pod_entry', \%make_pod_entry, 'r');
2416
2417 my %status;
2418 # Aliases have a status, like deprecated, or even suppressed (which means
2419 # they don't appear in documentation). Enum
2420 main::set_access('status', \%status, 'r');
2421
2422 my %externally_ok;
2423 # Similarly, some aliases should not be considered as usable ones for
2424 # external use, such as file names, or we don't want documentation to
2425 # recommend them. Boolean
2426 main::set_access('externally_ok', \%externally_ok, 'r');
2427
2428 sub new {
2429 my $class = shift;
2430
2431 my $self = bless \do { my $anonymous_scalar }, $class;
051df77b 2432 my $addr; { no overloading; $addr = pack 'J', $self; }
99870f4d
KW
2433
2434 $name{$addr} = shift;
2435 $loose_match{$addr} = shift;
2436 $make_pod_entry{$addr} = shift;
2437 $externally_ok{$addr} = shift;
2438 $status{$addr} = shift;
2439
2440 Carp::carp_extra_args(\@_) if main::DEBUG && @_;
2441
2442 # Null names are never ok externally
2443 $externally_ok{$addr} = 0 if $name{$addr} eq "";
2444
2445 return $self;
2446 }
2447}
2448
2449package Range;
2450
2451# A range is the basic unit for storing code points, and is described in the
2452# comments at the beginning of the program. Each range has a starting code
2453# point; an ending code point (not less than the starting one); a value
2454# that applies to every code point in between the two end-points, inclusive;
2455# and an enum type that applies to the value. The type is for the user's
2456# convenience, and has no meaning here, except that a non-zero type is
2457# considered to not obey the normal Unicode rules for having standard forms.
2458#
2459# The same structure is used for both map and match tables, even though in the
2460# latter, the value (and hence type) is irrelevant and could be used as a
2461# comment. In map tables, the value is what all the code points in the range
2462# map to. Type 0 values have the standardized version of the value stored as
2463# well, so as to not have to recalculate it a lot.
2464
2465sub trace { return main::trace(@_); }
2466
2467{ # Closure
2468
2469 main::setup_package();
2470
2471 my %start;
2472 main::set_access('start', \%start, 'r', 's');
2473
2474 my %end;
2475 main::set_access('end', \%end, 'r', 's');
2476
2477 my %value;
2478 main::set_access('value', \%value, 'r');
2479
2480 my %type;
2481 main::set_access('type', \%type, 'r');
2482
2483 my %standard_form;
2484 # The value in internal standard form. Defined only if the type is 0.
2485 main::set_access('standard_form', \%standard_form);
2486
2487 # Note that if these fields change, the dump() method should as well
2488
2489 sub new {
2490 return Carp::carp_too_few_args(\@_, 3) if main::DEBUG && @_ < 3;
2491 my $class = shift;
2492
2493 my $self = bless \do { my $anonymous_scalar }, $class;
051df77b 2494 my $addr; { no overloading; $addr = pack 'J', $self; }
99870f4d
KW
2495
2496 $start{$addr} = shift;
2497 $end{$addr} = shift;
2498
2499 my %args = @_;
2500
2501 my $value = delete $args{'Value'}; # Can be 0
2502 $value = "" unless defined $value;
2503 $value{$addr} = $value;
2504
2505 $type{$addr} = delete $args{'Type'} || 0;
2506
2507 Carp::carp_extra_args(\%args) if main::DEBUG && %args;
2508
2509 if (! $type{$addr}) {
2510 $standard_form{$addr} = main::standardize($value);
2511 }
2512
2513 return $self;
2514 }
2515
2516 use overload
2517 fallback => 0,
2518 qw("") => "_operator_stringify",
2519 "." => \&main::_operator_dot,
2520 ;
2521
2522 sub _operator_stringify {
2523 my $self = shift;
051df77b 2524 my $addr; { no overloading; $addr = pack 'J', $self; }
99870f4d
KW
2525
2526 # Output it like '0041..0065 (value)'
2527 my $return = sprintf("%04X", $start{$addr})
2528 . '..'
2529 . sprintf("%04X", $end{$addr});
2530 my $value = $value{$addr};
2531 my $type = $type{$addr};
2532 $return .= ' (';
2533 $return .= "$value";
2534 $return .= ", Type=$type" if $type != 0;
2535 $return .= ')';
2536
2537 return $return;
2538 }
2539
2540 sub standard_form {
2541 # The standard form is the value itself if the standard form is
2542 # undefined (that is if the value is special)
2543
2544 my $self = shift;
2545 Carp::carp_extra_args(\@_) if main::DEBUG && @_;
2546
051df77b 2547 my $addr; { no overloading; $addr = pack 'J', $self; }
99870f4d
KW
2548
2549 return $standard_form{$addr} if defined $standard_form{$addr};
2550 return $value{$addr};
2551 }
2552
2553 sub dump {
2554 # Human, not machine readable. For machine readable, comment out this
2555 # entire routine and let the standard one take effect.
2556 my $self = shift;
2557 my $indent = shift;
2558 Carp::carp_extra_args(\@_) if main::DEBUG && @_;
2559
051df77b 2560 my $addr; { no overloading; $addr = pack 'J', $self; }
99870f4d
KW
2561
2562 my $return = $indent
2563 . sprintf("%04X", $start{$addr})
2564 . '..'
2565 . sprintf("%04X", $end{$addr})
2566 . " '$value{$addr}';";
2567 if (! defined $standard_form{$addr}) {
2568 $return .= "(type=$type{$addr})";
2569 }
2570 elsif ($standard_form{$addr} ne $value{$addr}) {
2571 $return .= "(standard '$standard_form{$addr}')";
2572 }
2573 return $return;
2574 }
2575} # End closure
2576
2577package _Range_List_Base;
2578
2579# Base class for range lists. A range list is simply an ordered list of
2580# ranges, so that the ranges with the lowest starting numbers are first in it.
2581#
2582# When a new range is added that is adjacent to an existing range that has the
2583# same value and type, it merges with it to form a larger range.
2584#
2585# Ranges generally do not overlap, except that there can be multiple entries
2586# of single code point ranges. This is because of NameAliases.txt.
2587#
2588# In this program, there is a standard value such that if two different
2589# values, have the same standard value, they are considered equivalent. This
2590# value was chosen so that it gives correct results on Unicode data
2591
2592# There are a number of methods to manipulate range lists, and some operators
2593# are overloaded to handle them.
2594
99870f4d
KW
2595sub trace { return main::trace(@_); }
2596
2597{ # Closure
2598
2599 our $addr;
2600
2601 main::setup_package();
2602
2603 my %ranges;
2604 # The list of ranges
2605 main::set_access('ranges', \%ranges, 'readable_array');
2606
2607 my %max;
2608 # The highest code point in the list. This was originally a method, but
2609 # actual measurements said it was used a lot.
2610 main::set_access('max', \%max, 'r');
2611
2612 my %each_range_iterator;
2613 # Iterator position for each_range()
2614 main::set_access('each_range_iterator', \%each_range_iterator);
2615
2616 my %owner_name_of;
2617 # Name of parent this is attached to, if any. Solely for better error
2618 # messages.
2619 main::set_access('owner_name_of', \%owner_name_of, 'p_r');
2620
2621 my %_search_ranges_cache;
2622 # A cache of the previous result from _search_ranges(), for better
2623 # performance
2624 main::set_access('_search_ranges_cache', \%_search_ranges_cache);
2625
2626 sub new {
2627 my $class = shift;
2628 my %args = @_;
2629
2630 # Optional initialization data for the range list.
2631 my $initialize = delete $args{'Initialize'};
2632
2633 my $self;
2634
2635 # Use _union() to initialize. _union() returns an object of this
2636 # class, which means that it will call this constructor recursively.
2637 # But it won't have this $initialize parameter so that it won't
2638 # infinitely loop on this.
2639 return _union($class, $initialize, %args) if defined $initialize;
2640
2641 $self = bless \do { my $anonymous_scalar }, $class;
051df77b 2642 my $addr; { no overloading; $addr = pack 'J', $self; }
99870f4d
KW
2643
2644 # Optional parent object, only for debug info.
2645 $owner_name_of{$addr} = delete $args{'Owner'};
2646 $owner_name_of{$addr} = "" if ! defined $owner_name_of{$addr};
2647
2648 # Stringify, in case it is an object.
2649 $owner_name_of{$addr} = "$owner_name_of{$addr}";
2650
2651 # This is used only for error messages, and so a colon is added
2652 $owner_name_of{$addr} .= ": " if $owner_name_of{$addr} ne "";
2653
2654 Carp::carp_extra_args(\%args) if main::DEBUG && %args;
2655
2656 # Max is initialized to a negative value that isn't adjacent to 0,
2657 # for simpler tests
2658 $max{$addr} = -2;
2659
2660 $_search_ranges_cache{$addr} = 0;
2661 $ranges{$addr} = [];
2662
2663 return $self;
2664 }
2665
2666 use overload
2667 fallback => 0,
2668 qw("") => "_operator_stringify",
2669 "." => \&main::_operator_dot,
2670 ;
2671
2672 sub _operator_stringify {
2673 my $self = shift;
051df77b 2674 my $addr; { no overloading; $addr = pack 'J', $self; }
99870f4d
KW
2675
2676 return "Range_List attached to '$owner_name_of{$addr}'"
2677 if $owner_name_of{$addr};
2678 return "anonymous Range_List " . \$self;
2679 }
2680
2681 sub _union {
2682 # Returns the union of the input code points. It can be called as
2683 # either a constructor or a method. If called as a method, the result
2684 # will be a new() instance of the calling object, containing the union
2685 # of that object with the other parameter's code points; if called as
2686 # a constructor, the first parameter gives the class the new object
2687 # should be, and the second parameter gives the code points to go into
2688 # it.
2689 # In either case, there are two parameters looked at by this routine;
2690 # any additional parameters are passed to the new() constructor.
2691 #
2692 # The code points can come in the form of some object that contains
2693 # ranges, and has a conventionally named method to access them; or
2694 # they can be an array of individual code points (as integers); or
2695 # just a single code point.
2696 #
2697 # If they are ranges, this routine doesn't make any effort to preserve
2698 # the range values of one input over the other. Therefore this base
2699 # class should not allow _union to be called from other than
2700 # initialization code, so as to prevent two tables from being added
2701 # together where the range values matter. The general form of this
2702 # routine therefore belongs in a derived class, but it was moved here
2703 # to avoid duplication of code. The failure to overload this in this
2704 # class keeps it safe.
2705 #
2706
2707 my $self;
2708 my @args; # Arguments to pass to the constructor
2709
2710 my $class = shift;
2711
2712 # If a method call, will start the union with the object itself, and
2713 # the class of the new object will be the same as self.
2714 if (ref $class) {
2715 $self = $class;
2716 $class = ref $self;
2717 push @args, $self;
2718 }
2719
2720 # Add the other required parameter.
2721 push @args, shift;
2722 # Rest of parameters are passed on to the constructor
2723
2724 # Accumulate all records from both lists.
2725 my @records;
2726 for my $arg (@args) {
2727 #local $to_trace = 0 if main::DEBUG;
2728 trace "argument = $arg" if main::DEBUG && $to_trace;
2729 if (! defined $arg) {
2730 my $message = "";
2731 if (defined $self) {
f998e60c 2732 no overloading;
051df77b 2733 $message .= $owner_name_of{pack 'J', $self};
99870f4d
KW
2734 }
2735 Carp::my_carp_bug($message .= "Undefined argument to _union. No union done.");
2736 return;
2737 }
2738 $arg = [ $arg ] if ! ref $arg;
2739 my $type = ref $arg;
2740 if ($type eq 'ARRAY') {
2741 foreach my $element (@$arg) {
2742 push @records, Range->new($element, $element);
2743 }
2744 }
2745 elsif ($arg->isa('Range')) {
2746 push @records, $arg;
2747 }
2748 elsif ($arg->can('ranges')) {
2749 push @records, $arg->ranges;
2750 }
2751 else {
2752 my $message = "";
2753 if (defined $self) {
f998e60c 2754 no overloading;
051df77b 2755 $message .= $owner_name_of{pack 'J', $self};
99870f4d
KW
2756 }
2757 Carp::my_carp_bug($message . "Cannot take the union of a $type. No union done.");
2758 return;
2759 }
2760 }
2761
2762 # Sort with the range containing the lowest ordinal first, but if
2763 # two ranges start at the same code point, sort with the bigger range
2764 # of the two first, because it takes fewer cycles.
2765 @records = sort { ($a->start <=> $b->start)
2766 or
2767 # if b is shorter than a, b->end will be
2768 # less than a->end, and we want to select
2769 # a, so want to return -1
2770 ($b->end <=> $a->end)
2771 } @records;
2772
2773 my $new = $class->new(@_);
2774
2775 # Fold in records so long as they add new information.
2776 for my $set (@records) {
2777 my $start = $set->start;
2778 my $end = $set->end;
2779 my $value = $set->value;
2780 if ($start > $new->max) {
2781 $new->_add_delete('+', $start, $end, $value);
2782 }
2783 elsif ($end > $new->max) {
2784 $new->_add_delete('+', $new->max +1, $end, $value);
2785 }
2786 }
2787
2788 return $new;
2789 }
2790
2791 sub range_count { # Return the number of ranges in the range list
2792 my $self = shift;
2793 Carp::carp_extra_args(\@_) if main::DEBUG && @_;
2794
f998e60c 2795 no overloading;
051df77b 2796 return scalar @{$ranges{pack 'J', $self}};
99870f4d
KW
2797 }
2798
2799 sub min {
2800 # Returns the minimum code point currently in the range list, or if
2801 # the range list is empty, 2 beyond the max possible. This is a
2802 # method because used so rarely, that not worth saving between calls,
2803 # and having to worry about changing it as ranges are added and
2804 # deleted.
2805
2806 my $self = shift;
2807 Carp::carp_extra_args(\@_) if main::DEBUG && @_;
2808
051df77b 2809 my $addr; { no overloading; $addr = pack 'J', $self; }
99870f4d
KW
2810
2811 # If the range list is empty, return a large value that isn't adjacent
2812 # to any that could be in the range list, for simpler tests
2813 return $LAST_UNICODE_CODEPOINT + 2 unless scalar @{$ranges{$addr}};
2814 return $ranges{$addr}->[0]->start;
2815 }
2816
2817 sub contains {
2818 # Boolean: Is argument in the range list? If so returns $i such that:
2819 # range[$i]->end < $codepoint <= range[$i+1]->end
2820 # which is one beyond what you want; this is so that the 0th range
2821 # doesn't return false
2822 my $self = shift;
2823 my $codepoint = shift;
2824 Carp::carp_extra_args(\@_) if main::DEBUG && @_;
2825
99870f4d
KW
2826 my $i = $self->_search_ranges($codepoint);
2827 return 0 unless defined $i;
2828
2829 # The search returns $i, such that
2830 # range[$i-1]->end < $codepoint <= range[$i]->end
2831 # So is in the table if and only iff it is at least the start position
2832 # of range $i.
f998e60c 2833 no overloading;
051df77b 2834 return 0 if $ranges{pack 'J', $self}->[$i]->start > $codepoint;
99870f4d
KW
2835 return $i + 1;
2836 }
2837
2838 sub value_of {
2839 # Returns the value associated with the code point, undef if none
2840
2841 my $self = shift;
2842 my $codepoint = shift;
2843 Carp::carp_extra_args(\@_) if main::DEBUG && @_;
2844
99870f4d
KW
2845 my $i = $self->contains($codepoint);
2846 return unless $i;
2847
2848 # contains() returns 1 beyond where we should look
f998e60c 2849 no overloading;
051df77b 2850 return $ranges{pack 'J', $self}->[$i-1]->value;
99870f4d
KW
2851 }
2852
2853 sub _search_ranges {
2854 # Find the range in the list which contains a code point, or where it
2855 # should go if were to add it. That is, it returns $i, such that:
2856 # range[$i-1]->end < $codepoint <= range[$i]->end
2857 # Returns undef if no such $i is possible (e.g. at end of table), or
2858 # if there is an error.
2859
2860 my $self = shift;
2861 my $code_point = shift;
2862 Carp::carp_extra_args(\@_) if main::DEBUG && @_;
2863
051df77b 2864 my $addr; { no overloading; $addr = pack 'J', $self; }
99870f4d
KW
2865
2866 return if $code_point > $max{$addr};
2867 my $r = $ranges{$addr}; # The current list of ranges
2868 my $range_list_size = scalar @$r;
2869 my $i;
2870
2871 use integer; # want integer division
2872
2873 # Use the cached result as the starting guess for this one, because,
2874 # an experiment on 5.1 showed that 90% of the time the cache was the
2875 # same as the result on the next call (and 7% it was one less).
2876 $i = $_search_ranges_cache{$addr};
2877 $i = 0 if $i >= $range_list_size; # Reset if no longer valid (prob.
2878 # from an intervening deletion
2879 #local $to_trace = 1 if main::DEBUG;
2880 trace "previous \$i is still valid: $i" if main::DEBUG && $to_trace && $code_point <= $r->[$i]->end && ($i == 0 || $r->[$i-1]->end < $code_point);
2881 return $i if $code_point <= $r->[$i]->end
2882 && ($i == 0 || $r->[$i-1]->end < $code_point);
2883
2884 # Here the cache doesn't yield the correct $i. Try adding 1.
2885 if ($i < $range_list_size - 1
2886 && $r->[$i]->end < $code_point &&
2887 $code_point <= $r->[$i+1]->end)
2888 {
2889 $i++;
2890 trace "next \$i is correct: $i" if main::DEBUG && $to_trace;
2891 $_search_ranges_cache{$addr} = $i;
2892 return $i;
2893 }
2894
2895 # Here, adding 1 also didn't work. We do a binary search to
2896 # find the correct position, starting with current $i
2897 my $lower = 0;
2898 my $upper = $range_list_size - 1;
2899 while (1) {
2900 trace "top of loop i=$i:", sprintf("%04X", $r->[$lower]->start), "[$lower] .. ", sprintf("%04X", $r->[$i]->start), "[$i] .. ", sprintf("%04X", $r->[$upper]->start), "[$upper]" if main::DEBUG && $to_trace;
2901
2902 if ($code_point <= $r->[$i]->end) {
2903
2904 # Here we have met the upper constraint. We can quit if we
2905 # also meet the lower one.
2906 last if $i == 0 || $r->[$i-1]->end < $code_point;
2907
2908 $upper = $i; # Still too high.
2909
2910 }
2911 else {
2912
2913 # Here, $r[$i]->end < $code_point, so look higher up.
2914 $lower = $i;
2915 }
2916
2917 # Split search domain in half to try again.
2918 my $temp = ($upper + $lower) / 2;
2919
2920 # No point in continuing unless $i changes for next time
2921 # in the loop.
2922 if ($temp == $i) {
2923
2924 # We can't reach the highest element because of the averaging.
2925 # So if one below the upper edge, force it there and try one
2926 # more time.
2927 if ($i == $range_list_size - 2) {
2928
2929 trace "Forcing to upper edge" if main::DEBUG && $to_trace;
2930 $i = $range_list_size - 1;
2931
2932 # Change $lower as well so if fails next time through,
2933 # taking the average will yield the same $i, and we will
2934 # quit with the error message just below.
2935 $lower = $i;
2936 next;
2937 }
2938 Carp::my_carp_bug("$owner_name_of{$addr}Can't find where the range ought to go. No action taken.");
2939 return;
2940 }
2941 $i = $temp;
2942 } # End of while loop
2943
2944 if (main::DEBUG && $to_trace) {
2945 trace 'i-1=[', $i-1, ']', $r->[$i-1] if $i;
2946 trace "i= [ $i ]", $r->[$i];
2947 trace 'i+1=[', $i+1, ']', $r->[$i+1] if $i < $range_list_size - 1;
2948 }
2949
2950 # Here we have found the offset. Cache it as a starting point for the
2951 # next call.
2952 $_search_ranges_cache{$addr} = $i;
2953 return $i;
2954 }
2955
2956 sub _add_delete {
2957 # Add, replace or delete ranges to or from a list. The $type
2958 # parameter gives which:
2959 # '+' => insert or replace a range, returning a list of any changed
2960 # ranges.
2961 # '-' => delete a range, returning a list of any deleted ranges.
2962 #
2963 # The next three parameters give respectively the start, end, and
2964 # value associated with the range. 'value' should be null unless the
2965 # operation is '+';
2966 #
2967 # The range list is kept sorted so that the range with the lowest
2968 # starting position is first in the list, and generally, adjacent
2969 # ranges with the same values are merged into single larger one (see
2970 # exceptions below).
2971 #
2972 # There are more parameters, all are key => value pairs:
2973 # Type gives the type of the value. It is only valid for '+'.
2974 # All ranges have types; if this parameter is omitted, 0 is
2975 # assumed. Ranges with type 0 are assumed to obey the
2976 # Unicode rules for casing, etc; ranges with other types are
2977 # not. Otherwise, the type is arbitrary, for the caller's
2978 # convenience, and looked at only by this routine to keep
2979 # adjacent ranges of different types from being merged into
2980 # a single larger range, and when Replace =>
2981 # $IF_NOT_EQUIVALENT is specified (see just below).
2982 # Replace determines what to do if the range list already contains
2983 # ranges which coincide with all or portions of the input
2984 # range. It is only valid for '+':
2985 # => $NO means that the new value is not to replace
2986 # any existing ones, but any empty gaps of the
2987 # range list coinciding with the input range
2988 # will be filled in with the new value.
2989 # => $UNCONDITIONALLY means to replace the existing values with
2990 # this one unconditionally. However, if the
2991 # new and old values are identical, the
2992 # replacement is skipped to save cycles
2993 # => $IF_NOT_EQUIVALENT means to replace the existing values
2994 # with this one if they are not equivalent.
2995 # Ranges are equivalent if their types are the
2996 # same, and they are the same string, or if
2997 # both are type 0 ranges, if their Unicode
2998 # standard forms are identical. In this last
2999 # case, the routine chooses the more "modern"
3000 # one to use. This is because some of the
3001 # older files are formatted with values that
3002 # are, for example, ALL CAPs, whereas the
3003 # derived files have a more modern style,
3004 # which looks better. By looking for this
3005 # style when the pre-existing and replacement
3006 # standard forms are the same, we can move to
3007 # the modern style
3008 # => $MULTIPLE means that if this range duplicates an
3009 # existing one, but has a different value,
3010 # don't replace the existing one, but insert
3011 # this, one so that the same range can occur
3012 # multiple times.
3013 # => anything else is the same as => $IF_NOT_EQUIVALENT
3014 #
3015 # "same value" means identical for type-0 ranges, and it means having
3016 # the same standard forms for non-type-0 ranges.
3017
3018 return Carp::carp_too_few_args(\@_, 5) if main::DEBUG && @_ < 5;
3019
3020 my $self = shift;
3021 my $operation = shift; # '+' for add/replace; '-' for delete;
3022 my $start = shift;
3023 my $end = shift;
3024 my $value = shift;
3025
3026 my %args = @_;
3027
3028 $value = "" if not defined $value; # warning: $value can be "0"
3029
3030 my $replace = delete $args{'Replace'};
3031 $replace = $IF_NOT_EQUIVALENT unless defined $replace;
3032
3033 my $type = delete $args{'Type'};
3034 $type = 0 unless defined $type;
3035
3036 Carp::carp_extra_args(\%args) if main::DEBUG && %args;
3037
051df77b 3038 my $addr; { no overloading; $addr = pack 'J', $self; }
99870f4d
KW
3039
3040 if ($operation ne '+' && $operation ne '-') {
3041 Carp::my_carp_bug("$owner_name_of{$addr}First parameter to _add_delete must be '+' or '-'. No action taken.");
3042 return;
3043 }
3044 unless (defined $start && defined $end) {
3045 Carp::my_carp_bug("$owner_name_of{$addr}Undefined start and/or end to _add_delete. No action taken.");
3046 return;
3047 }
3048 unless ($end >= $start) {
3049 Carp::my_carp_bug("$owner_name_of{$addr}End of range (" . sprintf("%04X", $end) . ") must not be before start (" . sprintf("%04X", $start) . "). No action taken.");
3050 return;
3051 }
3052 #local $to_trace = 1 if main::DEBUG;
3053
3054 if ($operation eq '-') {
3055 if ($replace != $IF_NOT_EQUIVALENT) {
3056 Carp::my_carp_bug("$owner_name_of{$addr}Replace => \$IF_NOT_EQUIVALENT is required when deleting a range from a range list. Assuming Replace => \$IF_NOT_EQUIVALENT.");
3057 $replace = $IF_NOT_EQUIVALENT;
3058 }
3059 if ($type) {
3060 Carp::my_carp_bug("$owner_name_of{$addr}Type => 0 is required when deleting a range from a range list. Assuming Type => 0.");
3061 $type = 0;
3062 }
3063 if ($value ne "") {
3064 Carp::my_carp_bug("$owner_name_of{$addr}Value => \"\" is required when deleting a range from a range list. Assuming Value => \"\".");
3065 $value = "";
3066 }
3067 }
3068
3069 my $r = $ranges{$addr}; # The current list of ranges
3070 my $range_list_size = scalar @$r; # And its size
3071 my $max = $max{$addr}; # The current high code point in
3072 # the list of ranges
3073
3074 # Do a special case requiring fewer machine cycles when the new range
3075 # starts after the current highest point. The Unicode input data is
3076 # structured so this is common.
3077 if ($start > $max) {
3078
3079 trace "$owner_name_of{$addr} $operation", sprintf("%04X", $start) . '..' . sprintf("%04X", $end) . " ($value) type=$type" if main::DEBUG && $to_trace;
3080 return if $operation eq '-'; # Deleting a non-existing range is a
3081 # no-op
3082
3083 # If the new range doesn't logically extend the current final one
3084 # in the range list, create a new range at the end of the range
3085 # list. (max cleverly is initialized to a negative number not
3086 # adjacent to 0 if the range list is empty, so even adding a range
3087 # to an empty range list starting at 0 will have this 'if'
3088 # succeed.)
3089 if ($start > $max + 1 # non-adjacent means can't extend.
3090 || @{$r}[-1]->value ne $value # values differ, can't extend.
3091 || @{$r}[-1]->type != $type # types differ, can't extend.
3092 ) {
3093 push @$r, Range->new($start, $end,
3094 Value => $value,
3095 Type => $type);
3096 }
3097 else {
3098
3099 # Here, the new range starts just after the current highest in
3100 # the range list, and they have the same type and value.
3101 # Extend the current range to incorporate the new one.
3102 @{$r}[-1]->set_end($end);
3103 }
3104
3105 # This becomes the new maximum.
3106 $max{$addr} = $end;
3107
3108 return;
3109 }
3110 #local $to_trace = 0 if main::DEBUG;
3111
3112 trace "$owner_name_of{$addr} $operation", sprintf("%04X", $start) . '..' . sprintf("%04X", $end) . " ($value) replace=$replace" if main::DEBUG && $to_trace;
3113
3114 # Here, the input range isn't after the whole rest of the range list.
3115 # Most likely 'splice' will be needed. The rest of the routine finds
3116 # the needed splice parameters, and if necessary, does the splice.
3117 # First, find the offset parameter needed by the splice function for
3118 # the input range. Note that the input range may span multiple
3119 # existing ones, but we'll worry about that later. For now, just find
3120 # the beginning. If the input range is to be inserted starting in a
3121 # position not currently in the range list, it must (obviously) come
3122 # just after the range below it, and just before the range above it.
3123 # Slightly less obviously, it will occupy the position currently
3124 # occupied by the range that is to come after it. More formally, we
3125 # are looking for the position, $i, in the array of ranges, such that:
3126 #
3127 # r[$i-1]->start <= r[$i-1]->end < $start < r[$i]->start <= r[$i]->end
3128 #
3129 # (The ordered relationships within existing ranges are also shown in
3130 # the equation above). However, if the start of the input range is
3131 # within an existing range, the splice offset should point to that
3132 # existing range's position in the list; that is $i satisfies a
3133 # somewhat different equation, namely:
3134 #
3135 #r[$i-1]->start <= r[$i-1]->end < r[$i]->start <= $start <= r[$i]->end
3136 #
3137 # More briefly, $start can come before or after r[$i]->start, and at
3138 # this point, we don't know which it will be. However, these
3139 # two equations share these constraints:
3140 #
3141 # r[$i-1]->end < $start <= r[$i]->end
3142 #
3143 # And that is good enough to find $i.
3144
3145 my $i = $self->_search_ranges($start);
3146 if (! defined $i) {
3147 Carp::my_carp_bug("Searching $self for range beginning with $start unexpectedly returned undefined. Operation '$operation' not performed");
3148 return;
3149 }
3150
3151 # The search function returns $i such that:
3152 #
3153 # r[$i-1]->end < $start <= r[$i]->end
3154 #
3155 # That means that $i points to the first range in the range list
3156 # that could possibly be affected by this operation. We still don't
3157 # know if the start of the input range is within r[$i], or if it
3158 # points to empty space between r[$i-1] and r[$i].
3159 trace "[$i] is the beginning splice point. Existing range there is ", $r->[$i] if main::DEBUG && $to_trace;
3160
3161 # Special case the insertion of data that is not to replace any
3162 # existing data.
3163 if ($replace == $NO) { # If $NO, has to be operation '+'
3164 #local $to_trace = 1 if main::DEBUG;
3165 trace "Doesn't replace" if main::DEBUG && $to_trace;
3166
3167 # Here, the new range is to take effect only on those code points
3168 # that aren't already in an existing range. This can be done by
3169 # looking through the existing range list and finding the gaps in
3170 # the ranges that this new range affects, and then calling this
3171 # function recursively on each of those gaps, leaving untouched
3172 # anything already in the list. Gather up a list of the changed
3173 # gaps first so that changes to the internal state as new ranges
3174 # are added won't be a problem.
3175 my @gap_list;
3176
3177 # First, if the starting point of the input range is outside an
3178 # existing one, there is a gap from there to the beginning of the
3179 # existing range -- add a span to fill the part that this new
3180 # range occupies
3181 if ($start < $r->[$i]->start) {
3182 push @gap_list, Range->new($start,
3183 main::min($end,
3184 $r->[$i]->start - 1),
3185 Type => $type);
3186 trace "gap before $r->[$i] [$i], will add", $gap_list[-1] if main::DEBUG && $to_trace;
3187 }
3188
3189 # Then look through the range list for other gaps until we reach
3190 # the highest range affected by the input one.
3191 my $j;
3192 for ($j = $i+1; $j < $range_list_size; $j++) {
3193 trace "j=[$j]", $r->[$j] if main::DEBUG && $to_trace;
3194 last if $end < $r->[$j]->start;
3195
3196 # If there is a gap between when this range starts and the
3197 # previous one ends, add a span to fill it. Note that just
3198 # because there are two ranges doesn't mean there is a
3199 # non-zero gap between them. It could be that they have
3200 # different values or types
3201 if ($r->[$j-1]->end + 1 != $r->[$j]->start) {
3202 push @gap_list,
3203 Range->new($r->[$j-1]->end + 1,
3204 $r->[$j]->start - 1,
3205 Type => $type);
3206 trace "gap between $r->[$j-1] and $r->[$j] [$j], will add: $gap_list[-1]" if main::DEBUG && $to_trace;
3207 }
3208 }
3209
3210 # Here, we have either found an existing range in the range list,
3211 # beyond the area affected by the input one, or we fell off the
3212 # end of the loop because the input range affects the whole rest
3213 # of the range list. In either case, $j is 1 higher than the
3214 # highest affected range. If $j == $i, it means that there are no
3215 # affected ranges, that the entire insertion is in the gap between
3216 # r[$i-1], and r[$i], which we already have taken care of before
3217 # the loop.
3218 # On the other hand, if there are affected ranges, it might be
3219 # that there is a gap that needs filling after the final such
3220 # range to the end of the input range
3221 if ($r->[$j-1]->end < $end) {
3222 push @gap_list, Range->new(main::max($start,
3223 $r->[$j-1]->end + 1),
3224 $end,
3225 Type => $type);
3226 trace "gap after $r->[$j-1], will add $gap_list[-1]" if main::DEBUG && $to_trace;
3227 }
3228
3229 # Call recursively to fill in all the gaps.
3230 foreach my $gap (@gap_list) {
3231 $self->_add_delete($operation,
3232 $gap->start,
3233 $gap->end,
3234 $value,
3235 Type => $type);
3236 }
3237
3238 return;
3239 }
3240
3241 # Here, we have taken care of the case where $replace is $NO, which
3242 # means that whatever action we now take is done unconditionally. It
3243 # still could be that this call will result in a no-op, if duplicates
3244 # aren't allowed, and we are inserting a range that merely duplicates
3245 # data already in the range list; or also if deleting a non-existent
3246 # range.
3247 # $i still points to the first potential affected range. Now find the
3248 # highest range affected, which will determine the length parameter to
3249 # splice. (The input range can span multiple existing ones.) While
3250 # we are looking through the range list, see also if this is an
3251 # insertion that will change the values of at least one of the
3252 # affected ranges. We don't need to do this check unless this is an
3253 # insertion of non-multiples, and also since this is a boolean, we
3254 # don't need to do it if have already determined that it will make a
3255 # change; just unconditionally change them. $cdm is created to be 1
3256 # if either of these is true. (The 'c' in the name comes from below)
3257 my $cdm = ($operation eq '-' || $replace == $MULTIPLE);
3258 my $j; # This will point to the highest affected range
3259
3260 # For non-zero types, the standard form is the value itself;
3261 my $standard_form = ($type) ? $value : main::standardize($value);
3262
3263 for ($j = $i; $j < $range_list_size; $j++) {
3264 trace "Looking for highest affected range; the one at $j is ", $r->[$j] if main::DEBUG && $to_trace;
3265
3266 # If find a range that it doesn't overlap into, we can stop
3267 # searching
3268 last if $end < $r->[$j]->start;
3269
3270 # Here, overlaps the range at $j. If the value's don't match,
3271 # and this is supposedly an insertion, it becomes a change
3272 # instead. This is what the 'c' stands for in $cdm.
3273 if (! $cdm) {
3274 if ($r->[$j]->standard_form ne $standard_form) {
3275 $cdm = 1;
3276 }
3277 else {
3278
3279 # Here, the two values are essentially the same. If the
3280 # two are actually identical, replacing wouldn't change
3281 # anything so skip it.
3282 my $pre_existing = $r->[$j]->value;
3283 if ($pre_existing ne $value) {
3284
3285 # Here the new and old standardized values are the
3286 # same, but the non-standardized values aren't. If
3287 # replacing unconditionally, then replace
3288 if( $replace == $UNCONDITIONALLY) {
3289 $cdm = 1;
3290 }
3291 else {
3292
3293 # Here, are replacing conditionally. Decide to
3294 # replace or not based on which appears to look
3295 # the "nicest". If one is mixed case and the
3296 # other isn't, choose the mixed case one.
3297 my $new_mixed = $value =~ /[A-Z]/
3298 && $value =~ /[a-z]/;
3299 my $old_mixed = $pre_existing =~ /[A-Z]/
3300 && $pre_existing =~ /[a-z]/;
3301
3302 if ($old_mixed != $new_mixed) {
3303 $cdm = 1 if $new_mixed;
3304 if (main::DEBUG && $to_trace) {
3305 if ($cdm) {
3306 trace "Replacing $pre_existing with $value";
3307 }
3308 else {
3309 trace "Retaining $pre_existing over $value";
3310 }
3311 }
3312 }
3313 else {
3314
3315 # Here casing wasn't different between the two.
3316 # If one has hyphens or underscores and the
3317 # other doesn't, choose the one with the
3318 # punctuation.
3319 my $new_punct = $value =~ /[-_]/;
3320 my $old_punct = $pre_existing =~ /[-_]/;
3321
3322 if ($old_punct != $new_punct) {
3323 $cdm = 1 if $new_punct;
3324 if (main::DEBUG && $to_trace) {
3325 if ($cdm) {
3326 trace "Replacing $pre_existing with $value";
3327 }
3328 else {
3329 trace "Retaining $pre_existing over $value";
3330 }
3331 }
3332 } # else existing one is just as "good";
3333 # retain it to save cycles.
3334 }
3335 }
3336 }
3337 }
3338 }
3339 } # End of loop looking for highest affected range.
3340
3341 # Here, $j points to one beyond the highest range that this insertion
3342 # affects (hence to beyond the range list if that range is the final
3343 # one in the range list).
3344
3345 # The splice length is all the affected ranges. Get it before
3346 # subtracting, for efficiency, so we don't have to later add 1.
3347 my $length = $j - $i;
3348
3349 $j--; # $j now points to the highest affected range.
3350 trace "Final affected range is $j: $r->[$j]" if main::DEBUG && $to_trace;
3351
3352 # If inserting a multiple record, this is where it goes, after all the
3353 # existing ones for this range. This implies an insertion, and no
3354 # change to any existing ranges. Note that $j can be -1 if this new
3355 # range doesn't actually duplicate any existing, and comes at the
3356 # beginning of the list, in which case we can handle it like any other
3357 # insertion, and is easier to do so.
3358 if ($replace == $MULTIPLE && $j >= 0) {
3359
3360 # This restriction could be remedied with a little extra work, but
3361 # it won't hopefully ever be necessary
3362 if ($r->[$j]->start != $r->[$j]->end) {
3363 Carp::my_carp_bug("$owner_name_of{$addr}Can't cope with adding a multiple when the other range ($r->[$j]) contains more than one code point. No action taken.");
3364 return;
3365 }
3366
3367 # Don't add an exact duplicate, as it isn't really a multiple
3368 return if $value eq $r->[$j]->value && $type eq $r->[$j]->type;
3369
3370 trace "Adding multiple record at $j+1 with $start..$end, $value" if main::DEBUG && $to_trace;
3371 my @return = splice @$r,
3372 $j+1,
3373 0,
3374 Range->new($start,
3375 $end,
3376 Value => $value,
3377 Type => $type);
3378 if (main::DEBUG && $to_trace) {
3379 trace "After splice:";
3380 trace 'j-2=[', $j-2, ']', $r->[$j-2] if $j >= 2;
3381 trace 'j-1=[', $j-1, ']', $r->[$j-1] if $j >= 1;
3382 trace "j =[", $j, "]", $r->[$j] if $j >= 0;
3383 trace 'j+1=[', $j+1, ']', $r->[$j+1] if $j < @$r - 1;
3384 trace 'j+2=[', $j+2, ']', $r->[$j+2] if $j < @$r - 2;
3385 trace 'j+3=[', $j+3, ']', $r->[$j+3] if $j < @$r - 3;
3386 }
3387 return @return;
3388 }
3389
3390 # Here, have taken care of $NO and $MULTIPLE replaces.
3391 # $j points to the highest affected range. But it can be < $i or even
3392 # -1. These happen only if the insertion is entirely in the gap
3393 # between r[$i-1] and r[$i]. Here's why: j < i means that the j loop
3394 # above exited first time through with $end < $r->[$i]->start. (And
3395 # then we subtracted one from j) This implies also that $start <
3396 # $r->[$i]->start, but we know from above that $r->[$i-1]->end <
3397 # $start, so the entire input range is in the gap.
3398 if ($j < $i) {
3399
3400 # Here the entire input range is in the gap before $i.
3401
3402 if (main::DEBUG && $to_trace) {
3403 if ($i) {
3404 trace "Entire range is between $r->[$i-1] and $r->[$i]";
3405 }
3406 else {
3407 trace "Entire range is before $r->[$i]";
3408 }
3409 }
3410 return if $operation ne '+'; # Deletion of a non-existent range is
3411 # a no-op
3412 }
3413 else {
3414
3415 # Here the entire input range is not in the gap before $i. There
3416 # is an affected one, and $j points to the highest such one.
3417
3418 # At this point, here is the situation:
3419 # This is not an insertion of a multiple, nor of tentative ($NO)
3420 # data.
3421 # $i points to the first element in the current range list that
3422 # may be affected by this operation. In fact, we know
3423 # that the range at $i is affected because we are in
3424 # the else branch of this 'if'
3425 # $j points to the highest affected range.
3426 # In other words,
3427 # r[$i-1]->end < $start <= r[$i]->end
3428 # And:
3429 # r[$i-1]->end < $start <= $end <= r[$j]->end
3430 #
3431 # Also:
3432 # $cdm is a boolean which is set true if and only if this is a
3433 # change or deletion (multiple was handled above). In
3434 # other words, it could be renamed to be just $cd.
3435
3436 # We now have enough information to decide if this call is a no-op
3437 # or not. It is a no-op if it is a deletion of a non-existent
3438 # range, or an insertion of already existing data.
3439
3440 if (main::DEBUG && $to_trace && ! $cdm
3441 && $i == $j
3442 && $start >= $r->[$i]->start)
3443 {
3444 trace "no-op";
3445 }
3446 return if ! $cdm # change or delete => not no-op
3447 && $i == $j # more than one affected range => not no-op
3448
3449 # Here, r[$i-1]->end < $start <= $end <= r[$i]->end
3450 # Further, $start and/or $end is >= r[$i]->start
3451 # The test below hence guarantees that
3452 # r[$i]->start < $start <= $end <= r[$i]->end
3453 # This means the input range is contained entirely in
3454 # the one at $i, so is a no-op
3455 && $start >= $r->[$i]->start;
3456 }
3457
3458 # Here, we know that some action will have to be taken. We have
3459 # calculated the offset and length (though adjustments may be needed)
3460 # for the splice. Now start constructing the replacement list.
3461 my @replacement;
3462 my $splice_start = $i;
3463
3464 my $extends_below;
3465 my $extends_above;
3466
3467 # See if should extend any adjacent ranges.
3468 if ($operation eq '-') { # Don't extend deletions
3469 $extends_below = $extends_above = 0;
3470 }
3471 else { # Here, should extend any adjacent ranges. See if there are
3472 # any.
3473 $extends_below = ($i > 0
3474 # can't extend unless adjacent
3475 && $r->[$i-1]->end == $start -1
3476 # can't extend unless are same standard value
3477 && $r->[$i-1]->standard_form eq $standard_form
3478 # can't extend unless share type
3479 && $r->[$i-1]->type == $type);
3480 $extends_above = ($j+1 < $range_list_size
3481 && $r->[$j+1]->start == $end +1
3482 && $r->[$j+1]->standard_form eq $standard_form
3483 && $r->[$j-1]->type == $type);
3484 }
3485 if ($extends_below && $extends_above) { # Adds to both
3486 $splice_start--; # start replace at element below
3487 $length += 2; # will replace on both sides
3488 trace "Extends both below and above ranges" if main::DEBUG && $to_trace;
3489
3490 # The result will fill in any gap, replacing both sides, and
3491 # create one large range.
3492 @replacement = Range->new($r->[$i-1]->start,
3493 $r->[$j+1]->end,
3494 Value => $value,
3495 Type => $type);
3496 }
3497 else {
3498
3499 # Here we know that the result won't just be the conglomeration of
3500 # a new range with both its adjacent neighbors. But it could
3501 # extend one of them.
3502
3503 if ($extends_below) {
3504
3505 # Here the new element adds to the one below, but not to the
3506 # one above. If inserting, and only to that one range, can
3507 # just change its ending to include the new one.
3508 if ($length == 0 && ! $cdm) {
3509 $r->[$i-1]->set_end($end);
3510 trace "inserted range extends range to below so it is now $r->[$i-1]" if main::DEBUG && $to_trace;
3511 return;
3512 }
3513 else {
3514 trace "Changing inserted range to start at ", sprintf("%04X", $r->[$i-1]->start), " instead of ", sprintf("%04X", $start) if main::DEBUG && $to_trace;
3515 $splice_start--; # start replace at element below
3516 $length++; # will replace the element below
3517 $start = $r->[$i-1]->start;
3518 }
3519 }
3520 elsif ($extends_above) {
3521
3522 # Here the new element adds to the one above, but not below.
3523 # Mirror the code above
3524 if ($length == 0 && ! $cdm) {
3525 $r->[$j+1]->set_start($start);
3526 trace "inserted range extends range to above so it is now $r->[$j+1]" if main::DEBUG && $to_trace;
3527 return;
3528 }
3529 else {
3530 trace "Changing inserted range to end at ", sprintf("%04X", $r->[$j+1]->end), " instead of ", sprintf("%04X", $end) if main::DEBUG && $to_trace;
3531 $length++; # will replace the element above
3532 $end = $r->[$j+1]->end;
3533 }
3534 }
3535
3536 trace "Range at $i is $r->[$i]" if main::DEBUG && $to_trace;
3537
3538 # Finally, here we know there will have to be a splice.
3539 # If the change or delete affects only the highest portion of the
3540 # first affected range, the range will have to be split. The
3541 # splice will remove the whole range, but will replace it by a new
3542 # range containing just the unaffected part. So, in this case,
3543 # add to the replacement list just this unaffected portion.
3544 if (! $extends_below
3545 && $start > $r->[$i]->start && $start <= $r->[$i]->end)
3546 {
3547 push @replacement,
3548 Range->new($r->[$i]->start,
3549 $start - 1,
3550 Value => $r->[$i]->value,
3551 Type => $r->[$i]->type);
3552 }
3553
3554 # In the case of an insert or change, but not a delete, we have to
3555 # put in the new stuff; this comes next.
3556 if ($operation eq '+') {
3557 push @replacement, Range->new($start,
3558 $end,
3559 Value => $value,
3560 Type => $type);
3561 }
3562
3563 trace "Range at $j is $r->[$j]" if main::DEBUG && $to_trace && $j != $i;
3564 #trace "$end >=", $r->[$j]->start, " && $end <", $r->[$j]->end if main::DEBUG && $to_trace;
3565
3566 # And finally, if we're changing or deleting only a portion of the
3567 # highest affected range, it must be split, as the lowest one was.
3568 if (! $extends_above
3569 && $j >= 0 # Remember that j can be -1 if before first
3570 # current element
3571 && $end >= $r->[$j]->start
3572 && $end < $r->[$j]->end)
3573 {
3574 push @replacement,
3575 Range->new($end + 1,
3576 $r->[$j]->end,
3577 Value => $r->[$j]->value,
3578 Type => $r->[$j]->type);
3579 }
3580 }
3581
3582 # And do the splice, as calculated above
3583 if (main::DEBUG && $to_trace) {
3584 trace "replacing $length element(s) at $i with ";
3585 foreach my $replacement (@replacement) {
3586 trace " $replacement";
3587 }
3588 trace "Before splice:";
3589 trace 'i-2=[', $i-2, ']', $r->[$i-2] if $i >= 2;
3590 trace 'i-1=[', $i-1, ']', $r->[$i-1] if $i >= 1;
3591 trace "i =[", $i, "]", $r->[$i];
3592 trace 'i+1=[', $i+1, ']', $r->[$i+1] if $i < @$r - 1;
3593 trace 'i+2=[', $i+2, ']', $r->[$i+2] if $i < @$r - 2;
3594 }
3595
3596 my @return = splice @$r, $splice_start, $length, @replacement;
3597
3598 if (main::DEBUG && $to_trace) {
3599 trace "After splice:";
3600 trace 'i-2=[', $i-2, ']', $r->[$i-2] if $i >= 2;
3601 trace 'i-1=[', $i-1, ']', $r->[$i-1] if $i >= 1;
3602 trace "i =[", $i, "]", $r->[$i];
3603 trace 'i+1=[', $i+1, ']', $r->[$i+1] if $i < @$r - 1;
3604 trace 'i+2=[', $i+2, ']', $r->[$i+2] if $i < @$r - 2;
3605 trace "removed @return";
3606 }
3607
3608 # An actual deletion could have changed the maximum in the list.
3609 # There was no deletion if the splice didn't return something, but
3610 # otherwise recalculate it. This is done too rarely to worry about
3611 # performance.
3612 if ($operation eq '-' && @return) {
3613 $max{$addr} = $r->[-1]->end;
3614 }
3615 return @return;
3616 }
3617
3618 sub reset_each_range { # reset the iterator for each_range();
3619 my $self = shift;
3620 Carp::carp_extra_args(\@_) if main::DEBUG && @_;
3621
f998e60c 3622 no overloading;
051df77b 3623 undef $each_range_iterator{pack 'J', $self};
99870f4d
KW
3624 return;
3625 }
3626
3627 sub each_range {
3628 # Iterate over each range in a range list. Results are undefined if
3629 # the range list is changed during the iteration.
3630
3631 my $self = shift;
3632 Carp::carp_extra_args(\@_) if main::DEBUG && @_;
3633
051df77b 3634 my $addr; { no overloading; $addr = pack 'J', $self; }
99870f4d
KW
3635
3636 return if $self->is_empty;
3637
3638 $each_range_iterator{$addr} = -1
3639 if ! defined $each_range_iterator{$addr};
3640 $each_range_iterator{$addr}++;
3641 return $ranges{$addr}->[$each_range_iterator{$addr}]
3642 if $each_range_iterator{$addr} < @{$ranges{$addr}};
3643 undef $each_range_iterator{$addr};
3644 return;
3645 }
3646
3647 sub count { # Returns count of code points in range list
3648 my $self = shift;
3649 Carp::carp_extra_args(\@_) if main::DEBUG && @_;
3650
051df77b 3651 my $addr; { no overloading; $addr = pack 'J', $self; }
99870f4d
KW
3652
3653 my $count = 0;
3654 foreach my $range (@{$ranges{$addr}}) {
3655 $count += $range->end - $range->start + 1;
3656 }
3657 return $count;
3658 }
3659
3660 sub delete_range { # Delete a range
3661 my $self = shift;
3662 my $start = shift;
3663 my $end = shift;
3664
3665 Carp::carp_extra_args(\@_) if main::DEBUG && @_;
3666
3667 return $self->_add_delete('-', $start, $end, "");
3668 }
3669
3670 sub is_empty { # Returns boolean as to if a range list is empty
3671 my $self = shift;
3672 Carp::carp_extra_args(\@_) if main::DEBUG && @_;
3673
f998e60c 3674 no overloading;
051df77b 3675 return scalar @{$ranges{pack 'J', $self}} == 0;
99870f4d
KW
3676 }
3677
3678 sub hash {
3679 # Quickly returns a scalar suitable for separating tables into
3680 # buckets, i.e. it is a hash function of the contents of a table, so
3681 # there are relatively few conflicts.
3682
3683 my $self = shift;
3684 Carp::carp_extra_args(\@_) if main::DEBUG && @_;
3685
051df77b 3686 my $addr; { no overloading; $addr = pack 'J', $self; }
99870f4d
KW
3687
3688 # These are quickly computable. Return looks like 'min..max;count'
3689 return $self->min . "..$max{$addr};" . scalar @{$ranges{$addr}};
3690 }
3691} # End closure for _Range_List_Base
3692
3693package Range_List;
3694use base '_Range_List_Base';
3695
3696# A Range_List is a range list for match tables; i.e. the range values are
3697# not significant. Thus a number of operations can be safely added to it,
3698# such as inversion, intersection. Note that union is also an unsafe
3699# operation when range values are cared about, and that method is in the base
3700# class, not here. But things are set up so that that method is callable only
3701# during initialization. Only in this derived class, is there an operation
3702# that combines two tables. A Range_Map can thus be used to initialize a
3703# Range_List, and its mappings will be in the list, but are not significant to
3704# this class.
3705
3706sub trace { return main::trace(@_); }
3707
3708{ # Closure
3709
3710 use overload
3711 fallback => 0,
3712 '+' => sub { my $self = shift;
3713 my $other = shift;
3714
3715 return $self->_union($other)
3716 },
3717 '&' => sub { my $self = shift;
3718 my $other = shift;
3719
3720 return $self->_intersect($other, 0);
3721 },
3722 '~' => "_invert",
3723 '-' => "_subtract",
3724 ;
3725
3726 sub _invert {
3727 # Returns a new Range_List that gives all code points not in $self.
3728
3729 my $self = shift;
3730
3731 my $new = Range_List->new;
3732
3733 # Go through each range in the table, finding the gaps between them
3734 my $max = -1; # Set so no gap before range beginning at 0
3735 for my $range ($self->ranges) {
3736 my $start = $range->start;
3737 my $end = $range->end;
3738
3739 # If there is a gap before this range, the inverse will contain
3740 # that gap.
3741 if ($start > $max + 1) {
3742 $new->add_range($max + 1, $start - 1);
3743 }
3744 $max = $end;
3745 }
3746
3747 # And finally, add the gap from the end of the table to the max
3748 # possible code point
3749 if ($max < $LAST_UNICODE_CODEPOINT) {
3750 $new->add_range($max + 1, $LAST_UNICODE_CODEPOINT);
3751 }
3752 return $new;
3753 }
3754
3755 sub _subtract {
3756 # Returns a new Range_List with the argument deleted from it. The
3757 # argument can be a single code point, a range, or something that has
3758 # a range, with the _range_list() method on it returning them
3759
3760 my $self = shift;
3761 my $other = shift;
3762 my $reversed = shift;
3763 Carp::carp_extra_args(\@_) if main::DEBUG && @_;
3764
3765 if ($reversed) {
3766 Carp::my_carp_bug("Can't cope with a "
3767 . __PACKAGE__
3768 . " being the second parameter in a '-'. Subtraction ignored.");
3769 return $self;
3770 }
3771
3772 my $new = Range_List->new(Initialize => $self);
3773
3774 if (! ref $other) { # Single code point
3775 $new->delete_range($other, $other);
3776 }
3777 elsif ($other->isa('Range')) {
3778 $new->delete_range($other->start, $other->end);
3779 }
3780 elsif ($other->can('_range_list')) {
3781 foreach my $range ($other->_range_list->ranges) {
3782 $new->delete_range($range->start, $range->end);
3783 }
3784 }
3785 else {
3786 Carp::my_carp_bug("Can't cope with a "
3787 . ref($other)
3788 . " argument to '-'. Subtraction ignored."
3789 );
3790 return $self;
3791 }
3792
3793 return $new;
3794 }
3795
3796 sub _intersect {
3797 # Returns either a boolean giving whether the two inputs' range lists
3798 # intersect (overlap), or a new Range_List containing the intersection
3799 # of the two lists. The optional final parameter being true indicates
3800 # to do the check instead of the intersection.
3801
3802 my $a_object = shift;
3803 my $b_object = shift;
3804 my $check_if_overlapping = shift;
3805 $check_if_overlapping = 0 unless defined $check_if_overlapping;
3806 Carp::carp_extra_args(\@_) if main::DEBUG && @_;
3807
3808 if (! defined $b_object) {
3809 my $message = "";
3810 $message .= $a_object->_owner_name_of if defined $a_object;
3811 Carp::my_carp_bug($message .= "Called with undefined value. Intersection not done.");
3812 return;
3813 }
3814
3815 # a & b = !(!a | !b), or in our terminology = ~ ( ~a + -b )
3816 # Thus the intersection could be much more simply be written:
3817 # return ~(~$a_object + ~$b_object);
3818 # But, this is slower, and when taking the inverse of a large
3819 # range_size_1 table, back when such tables were always stored that
3820 # way, it became prohibitively slow, hence the code was changed to the
3821 # below
3822
3823 if ($b_object->isa('Range')) {
3824 $b_object = Range_List->new(Initialize => $b_object,
3825 Owner => $a_object->_owner_name_of);
3826 }
3827 $b_object = $b_object->_range_list if $b_object->can('_range_list');
3828
3829 my @a_ranges = $a_object->ranges;
3830 my @b_ranges = $b_object->ranges;
3831
3832 #local $to_trace = 1 if main::DEBUG;
3833 trace "intersecting $a_object with ", scalar @a_ranges, "ranges and $b_object with", scalar @b_ranges, " ranges" if main::DEBUG && $to_trace;
3834
3835 # Start with the first range in each list
3836 my $a_i = 0;
3837 my $range_a = $a_ranges[$a_i];
3838 my $b_i = 0;
3839 my $range_b = $b_ranges[$b_i];
3840
3841 my $new = __PACKAGE__->new(Owner => $a_object->_owner_name_of)
3842 if ! $check_if_overlapping;
3843
3844 # If either list is empty, there is no intersection and no overlap
3845 if (! defined $range_a || ! defined $range_b) {
3846 return $check_if_overlapping ? 0 : $new;
3847 }
3848 trace "range_a[$a_i]=$range_a; range_b[$b_i]=$range_b" if main::DEBUG && $to_trace;
3849
3850 # Otherwise, must calculate the intersection/overlap. Start with the
3851 # very first code point in each list
3852 my $a = $range_a->start;
3853 my $b = $range_b->start;
3854
3855 # Loop through all the ranges of each list; in each iteration, $a and
3856 # $b are the current code points in their respective lists
3857 while (1) {
3858
3859 # If $a and $b are the same code point, ...
3860 if ($a == $b) {
3861
3862 # it means the lists overlap. If just checking for overlap
3863 # know the answer now,
3864 return 1 if $check_if_overlapping;
3865
3866 # The intersection includes this code point plus anything else
3867 # common to both current ranges.
3868 my $start = $a;
3869 my $end = main::min($range_a->end, $range_b->end);
3870 if (! $check_if_overlapping) {
3871 trace "adding intersection range ", sprintf("%04X", $start) . ".." . sprintf("%04X", $end) if main::DEBUG && $to_trace;
3872 $new->add_range($start, $end);
3873 }
3874
3875 # Skip ahead to the end of the current intersect
3876 $a = $b = $end;
3877
3878 # If the current intersect ends at the end of either range (as
3879 # it must for at least one of them), the next possible one
3880 # will be the beginning code point in it's list's next range.
3881 if ($a == $range_a->end) {
3882 $range_a = $a_ranges[++$a_i];
3883 last unless defined $range_a;
3884 $a = $range_a->start;
3885 }
3886 if ($b == $range_b->end) {
3887 $range_b = $b_ranges[++$b_i];
3888 last unless defined $range_b;
3889 $b = $range_b->start;
3890 }
3891
3892 trace "range_a[$a_i]=$range_a; range_b[$b_i]=$range_b" if main::DEBUG && $to_trace;
3893 }
3894 elsif ($a < $b) {
3895
3896 # Not equal, but if the range containing $a encompasses $b,
3897 # change $a to be the middle of the range where it does equal
3898 # $b, so the next iteration will get the intersection
3899 if ($range_a->end >= $b) {
3900 $a = $b;
3901 }
3902 else {
3903
3904 # Here, the current range containing $a is entirely below
3905 # $b. Go try to find a range that could contain $b.
3906 $a_i = $a_object->_search_ranges($b);
3907
3908 # If no range found, quit.
3909 last unless defined $a_i;
3910
3911 # The search returns $a_i, such that
3912 # range_a[$a_i-1]->end < $b <= range_a[$a_i]->end
3913 # Set $a to the beginning of this new range, and repeat.
3914 $range_a = $a_ranges[$a_i];
3915 $a = $range_a->start;
3916 }
3917 }
3918 else { # Here, $b < $a.
3919
3920 # Mirror image code to the leg just above
3921 if ($range_b->end >= $a) {
3922 $b = $a;
3923 }
3924 else {
3925 $b_i = $b_object->_search_ranges($a);
3926 last unless defined $b_i;
3927 $range_b = $b_ranges[$b_i];
3928 $b = $range_b->start;
3929 }
3930 }
3931 } # End of looping through ranges.
3932
3933 # Intersection fully computed, or now know that there is no overlap
3934 return $check_if_overlapping ? 0 : $new;
3935 }
3936
3937 sub overlaps {
3938 # Returns boolean giving whether the two arguments overlap somewhere
3939
3940 my $self = shift;
3941 my $other = shift;
3942 Carp::carp_extra_args(\@_) if main::DEBUG && @_;
3943
3944 return $self->_intersect($other, 1);
3945 }
3946
3947 sub add_range {
3948 # Add a range to the list.
3949
3950 my $self = shift;
3951 my $start = shift;
3952 my $end = shift;
3953 Carp::carp_extra_args(\@_) if main::DEBUG && @_;
3954
3955 return $self->_add_delete('+', $start, $end, "");
3956 }
3957
99870f4d
KW
3958 sub is_code_point_usable {
3959 # This used only for making the test script. See if the input
3960 # proposed trial code point is one that Perl will handle. If second
3961 # parameter is 0, it won't select some code points for various
3962 # reasons, noted below.
3963
3964 my $code = shift;
3965 my $try_hard = shift;
3966 Carp::carp_extra_args(\@_) if main::DEBUG && @_;
3967
3968 return 0 if $code < 0; # Never use a negative
3969
99870f4d
KW
3970 # shun null. I'm (khw) not sure why this was done, but NULL would be
3971 # the character very frequently used.
3972 return $try_hard if $code == 0x0000;
3973
3974 return 0 if $try_hard; # XXX Temporary until fix utf8.c
3975
3976 # shun non-character code points.
3977 return $try_hard if $code >= 0xFDD0 && $code <= 0xFDEF;
3978 return $try_hard if ($code & 0xFFFE) == 0xFFFE; # includes FFFF
3979
3980 return $try_hard if $code > $LAST_UNICODE_CODEPOINT; # keep in range
3981 return $try_hard if $code >= 0xD800 && $code <= 0xDFFF; # no surrogate
3982
3983 return 1;
3984 }
3985
3986 sub get_valid_code_point {
3987 # Return a code point that's part of the range list. Returns nothing
3988 # if the table is empty or we can't find a suitable code point. This
3989 # used only for making the test script.
3990
3991 my $self = shift;
3992 Carp::carp_extra_args(\@_) if main::DEBUG && @_;
3993
051df77b 3994 my $addr; { no overloading; $addr = pack 'J', $self; }
99870f4d
KW
3995
3996 # On first pass, don't choose less desirable code points; if no good
3997 # one is found, repeat, allowing a less desirable one to be selected.
3998 for my $try_hard (0, 1) {
3999
4000 # Look through all the ranges for a usable code point.
4001 for my $set ($self->ranges) {
4002
4003 # Try the edge cases first, starting with the end point of the
4004 # range.
4005 my $end = $set->end;
4006 return $end if is_code_point_usable($end, $try_hard);
4007
4008 # End point didn't, work. Start at the beginning and try
4009 # every one until find one that does work.
4010 for my $trial ($set->start .. $end - 1) {
4011 return $trial if is_code_point_usable($trial, $try_hard);
4012 }
4013 }
4014 }
4015 return (); # If none found, give up.
4016 }
4017
4018 sub get_invalid_code_point {
4019 # Return a code point that's not part of the table. Returns nothing
4020 # if the table covers all code points or a suitable code point can't
4021 # be found. This used only for making the test script.
4022
4023 my $self = shift;
4024 Carp::carp_extra_args(\@_) if main::DEBUG && @_;
4025
4026 # Just find a valid code point of the inverse, if any.
4027 return Range_List->new(Initialize => ~ $self)->get_valid_code_point;
4028 }
4029} # end closure for Range_List
4030
4031package Range_Map;
4032use base '_Range_List_Base';
4033
4034# A Range_Map is a range list in which the range values (called maps) are
4035# significant, and hence shouldn't be manipulated by our other code, which
4036# could be ambiguous or lose things. For example, in taking the union of two
4037# lists, which share code points, but which have differing values, which one
4038# has precedence in the union?
4039# It turns out that these operations aren't really necessary for map tables,
4040# and so this class was created to make sure they aren't accidentally
4041# applied to them.
4042
4043{ # Closure
4044
4045 sub add_map {
4046 # Add a range containing a mapping value to the list
4047