This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
mktables: Clarify comment in generated tables
[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:
ffe43484 12# my $addr = do { no overloading; 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
c1739a4a 314# would look like qr/\p{JSN=}/ currently without modifying the regex code.
678f13d5
KW
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.
99870f4d 416#
678f13d5
KW
417# HOW TO ADD A FILE TO BE PROCESSED
418#
419# A new file from Unicode needs to have an object constructed for it in
420# @input_file_objects, probably at the end or at the end of the extracted
421# ones. The program should warn you if its name will clash with others on
422# restrictive file systems, like DOS. If so, figure out a better name, and
423# add lines to the README.perl file giving that. If the file is a character
424# property, it should be in the format that Unicode has by default
425# standardized for such files for the more recently introduced ones.
426# If so, the Input_file constructor for @input_file_objects can just be the
427# file name and release it first appeared in. If not, then it should be
428# possible to construct an each_line_handler() to massage the line into the
429# standardized form.
430#
431# For non-character properties, more code will be needed. You can look at
432# the existing entries for clues.
433#
434# UNICODE VERSIONS NOTES
435#
436# The Unicode UCD has had a number of errors in it over the versions. And
437# these remain, by policy, in the standard for that version. Therefore it is
438# risky to correct them, because code may be expecting the error. So this
439# program doesn't generally make changes, unless the error breaks the Perl
440# core. As an example, some versions of 2.1.x Jamo.txt have the wrong value
441# for U+1105, which causes real problems for the algorithms for Jamo
442# calculations, so it is changed here.
443#
444# But it isn't so clear cut as to what to do about concepts that are
445# introduced in a later release; should they extend back to earlier releases
446# where the concept just didn't exist? It was easier to do this than to not,
447# so that's what was done. For example, the default value for code points not
448# in the files for various properties was probably undefined until changed by
449# some version. No_Block for blocks is such an example. This program will
450# assign No_Block even in Unicode versions that didn't have it. This has the
451# benefit that code being written doesn't have to special case earlier
452# versions; and the detriment that it doesn't match the Standard precisely for
453# the affected versions.
454#
455# Here are some observations about some of the issues in early versions:
456#
457# The number of code points in \p{alpha} halve in 2.1.9. It turns out that
458# the reason is that the CJK block starting at 4E00 was removed from PropList,
459# and was not put back in until 3.1.0
460#
461# Unicode introduced the synonym Space for White_Space in 4.1. Perl has
462# always had a \p{Space}. In release 3.2 only, they are not synonymous. The
463# reason is that 3.2 introduced U+205F=medium math space, which was not
464# classed as white space, but Perl figured out that it should have been. 4.0
465# reclassified it correctly.
466#
467# Another change between 3.2 and 4.0 is the CCC property value ATBL. In 3.2
468# this was erroneously a synonym for 202. In 4.0, ATB became 202, and ATBL
469# was left with no code points, as all the ones that mapped to 202 stayed
470# mapped to 202. Thus if your program used the numeric name for the class,
471# it would not have been affected, but if it used the mnemonic, it would have
472# been.
473#
474# \p{Script=Hrkt} (Katakana_Or_Hiragana) came in 4.0.1. Before that code
475# points which eventually came to have this script property value, instead
476# mapped to "Unknown". But in the next release all these code points were
477# moved to \p{sc=common} instead.
99870f4d
KW
478#
479# The default for missing code points for BidiClass is complicated. Starting
480# in 3.1.1, the derived file DBidiClass.txt handles this, but this program
481# tries to do the best it can for earlier releases. It is done in
482# process_PropertyAliases()
483#
484##############################################################################
485
486my $UNDEF = ':UNDEF:'; # String to print out for undefined values in tracing
487 # and errors
488my $MAX_LINE_WIDTH = 78;
489
490# Debugging aid to skip most files so as to not be distracted by them when
491# concentrating on the ones being debugged. Add
492# non_skip => 1,
493# to the constructor for those files you want processed when you set this.
494# Files with a first version number of 0 are special: they are always
495# processed regardless of the state of this flag.
496my $debug_skip = 0;
497
498# Set to 1 to enable tracing.
499our $to_trace = 0;
500
501{ # Closure for trace: debugging aid
502 my $print_caller = 1; # ? Include calling subroutine name
503 my $main_with_colon = 'main::';
504 my $main_colon_length = length($main_with_colon);
505
506 sub trace {
507 return unless $to_trace; # Do nothing if global flag not set
508
509 my @input = @_;
510
511 local $DB::trace = 0;
512 $DB::trace = 0; # Quiet 'used only once' message
513
514 my $line_number;
515
516 # Loop looking up the stack to get the first non-trace caller
517 my $caller_line;
518 my $caller_name;
519 my $i = 0;
520 do {
521 $line_number = $caller_line;
522 (my $pkg, my $file, $caller_line, my $caller) = caller $i++;
523 $caller = $main_with_colon unless defined $caller;
524
525 $caller_name = $caller;
526
527 # get rid of pkg
528 $caller_name =~ s/.*:://;
529 if (substr($caller_name, 0, $main_colon_length)
530 eq $main_with_colon)
531 {
532 $caller_name = substr($caller_name, $main_colon_length);
533 }
534
535 } until ($caller_name ne 'trace');
536
537 # If the stack was empty, we were called from the top level
538 $caller_name = 'main' if ($caller_name eq ""
539 || $caller_name eq 'trace');
540
541 my $output = "";
542 foreach my $string (@input) {
543 #print STDERR __LINE__, ": ", join ", ", @input, "\n";
544 if (ref $string eq 'ARRAY' || ref $string eq 'HASH') {
545 $output .= simple_dumper($string);
546 }
547 else {
548 $string = "$string" if ref $string;
549 $string = $UNDEF unless defined $string;
550 chomp $string;
551 $string = '""' if $string eq "";
552 $output .= " " if $output ne ""
553 && $string ne ""
554 && substr($output, -1, 1) ne " "
555 && substr($string, 0, 1) ne " ";
556 $output .= $string;
557 }
558 }
559
99f78760
KW
560 print STDERR sprintf "%4d: ", $line_number if defined $line_number;
561 print STDERR "$caller_name: " if $print_caller;
99870f4d
KW
562 print STDERR $output, "\n";
563 return;
564 }
565}
566
567# This is for a rarely used development feature that allows you to compare two
568# versions of the Unicode standard without having to deal with changes caused
569# by the code points introduced in the later verson. Change the 0 to a SINGLE
570# dotted Unicode release number (e.g. 2.1). Only code points introduced in
571# that release and earlier will be used; later ones are thrown away. You use
572# the version number of the earliest one you want to compare; then run this
573# program on directory structures containing each release, and compare the
574# outputs. These outputs will therefore include only the code points common
575# to both releases, and you can see the changes caused just by the underlying
576# release semantic changes. For versions earlier than 3.2, you must copy a
577# version of DAge.txt into the directory.
578my $string_compare_versions = DEBUG && 0; # e.g., v2.1;
579my $compare_versions = DEBUG
580 && $string_compare_versions
581 && pack "C*", split /\./, $string_compare_versions;
582
583sub uniques {
584 # Returns non-duplicated input values. From "Perl Best Practices:
585 # Encapsulated Cleverness". p. 455 in first edition.
586
587 my %seen;
0e407844
NC
588 # Arguably this breaks encapsulation, if the goal is to permit multiple
589 # distinct objects to stringify to the same value, and be interchangeable.
590 # However, for this program, no two objects stringify identically, and all
591 # lists passed to this function are either objects or strings. So this
592 # doesn't affect correctness, but it does give a couple of percent speedup.
593 no overloading;
99870f4d
KW
594 return grep { ! $seen{$_}++ } @_;
595}
596
597$0 = File::Spec->canonpath($0);
598
599my $make_test_script = 0; # ? Should we output a test script
600my $write_unchanged_files = 0; # ? Should we update the output files even if
601 # we don't think they have changed
602my $use_directory = ""; # ? Should we chdir somewhere.
603my $pod_directory; # input directory to store the pod file.
604my $pod_file = 'perluniprops';
605my $t_path; # Path to the .t test file
606my $file_list = 'mktables.lst'; # File to store input and output file names.
607 # This is used to speed up the build, by not
608 # executing the main body of the program if
609 # nothing on the list has changed since the
610 # previous build
611my $make_list = 1; # ? Should we write $file_list. Set to always
612 # make a list so that when the pumpking is
613 # preparing a release, s/he won't have to do
614 # special things
615my $glob_list = 0; # ? Should we try to include unknown .txt files
616 # in the input.
617my $output_range_counts = 1; # ? Should we include the number of code points
618 # in ranges in the output
9ef2b94f
KW
619my $output_names = 0; # ? Should character names be in the output
620my @viacode; # Contains the 1 million character names, if
621 # $output_names is true
622
99870f4d
KW
623# Verbosity levels; 0 is quiet
624my $NORMAL_VERBOSITY = 1;
625my $PROGRESS = 2;
626my $VERBOSE = 3;
627
628my $verbosity = $NORMAL_VERBOSITY;
629
630# Process arguments
631while (@ARGV) {
cf25bb62
JH
632 my $arg = shift @ARGV;
633 if ($arg eq '-v') {
99870f4d
KW
634 $verbosity = $VERBOSE;
635 }
636 elsif ($arg eq '-p') {
637 $verbosity = $PROGRESS;
638 $| = 1; # Flush buffers as we go.
639 }
640 elsif ($arg eq '-q') {
641 $verbosity = 0;
642 }
643 elsif ($arg eq '-w') {
644 $write_unchanged_files = 1; # update the files even if havent changed
645 }
646 elsif ($arg eq '-check') {
6ae7e459
YO
647 my $this = shift @ARGV;
648 my $ok = shift @ARGV;
649 if ($this ne $ok) {
650 print "Skipping as check params are not the same.\n";
651 exit(0);
652 }
00a8df5c 653 }
99870f4d
KW
654 elsif ($arg eq '-P' && defined ($pod_directory = shift)) {
655 -d $pod_directory or croak "Directory '$pod_directory' doesn't exist";
656 }
3df51b85
KW
657 elsif ($arg eq '-maketest' || ($arg eq '-T' && defined ($t_path = shift)))
658 {
99870f4d 659 $make_test_script = 1;
99870f4d
KW
660 }
661 elsif ($arg eq '-makelist') {
662 $make_list = 1;
663 }
664 elsif ($arg eq '-C' && defined ($use_directory = shift)) {
665 -d $use_directory or croak "Unknown directory '$use_directory'";
666 }
667 elsif ($arg eq '-L') {
668
669 # Existence not tested until have chdir'd
670 $file_list = shift;
671 }
672 elsif ($arg eq '-globlist') {
673 $glob_list = 1;
674 }
675 elsif ($arg eq '-c') {
676 $output_range_counts = ! $output_range_counts
677 }
9ef2b94f
KW
678 elsif ($arg eq '-output_names') {
679 $output_names = 1;
680 }
99870f4d
KW
681 else {
682 my $with_c = 'with';
683 $with_c .= 'out' if $output_range_counts; # Complements the state
684 croak <<END;
685usage: $0 [-c|-p|-q|-v|-w] [-C dir] [-L filelist] [ -P pod_dir ]
686 [ -T test_file_path ] [-globlist] [-makelist] [-maketest]
687 [-check A B ]
688 -c : Output comments $with_c number of code points in ranges
689 -q : Quiet Mode: Only output serious warnings.
690 -p : Set verbosity level to normal plus show progress.
691 -v : Set Verbosity level high: Show progress and non-serious
692 warnings
693 -w : Write files regardless
694 -C dir : Change to this directory before proceeding. All relative paths
695 except those specified by the -P and -T options will be done
696 with respect to this directory.
697 -P dir : Output $pod_file file to directory 'dir'.
3df51b85 698 -T path : Create a test script as 'path'; overrides -maketest
99870f4d
KW
699 -L filelist : Use alternate 'filelist' instead of standard one
700 -globlist : Take as input all non-Test *.txt files in current and sub
701 directories
3df51b85
KW
702 -maketest : Make test script 'TestProp.pl' in current (or -C directory),
703 overrides -T
99870f4d 704 -makelist : Rewrite the file list $file_list based on current setup
9ef2b94f
KW
705 -output_names : Output each character's name in the table files; useful for
706 doing what-ifs, looking at diffs; is slow, memory intensive,
707 resulting tables are usable but very large.
99870f4d
KW
708 -check A B : Executes $0 only if A and B are the same
709END
710 }
711}
712
713# Stores the most-recently changed file. If none have changed, can skip the
714# build
715my $youngest = -M $0; # Do this before the chdir!
716
717# Change directories now, because need to read 'version' early.
718if ($use_directory) {
3df51b85 719 if ($pod_directory && ! File::Spec->file_name_is_absolute($pod_directory)) {
99870f4d
KW
720 $pod_directory = File::Spec->rel2abs($pod_directory);
721 }
3df51b85 722 if ($t_path && ! File::Spec->file_name_is_absolute($t_path)) {
99870f4d 723 $t_path = File::Spec->rel2abs($t_path);
00a8df5c 724 }
99870f4d 725 chdir $use_directory or croak "Failed to chdir to '$use_directory':$!";
3df51b85 726 if ($pod_directory && File::Spec->file_name_is_absolute($pod_directory)) {
99870f4d 727 $pod_directory = File::Spec->abs2rel($pod_directory);
02b1aeec 728 }
3df51b85 729 if ($t_path && File::Spec->file_name_is_absolute($t_path)) {
99870f4d 730 $t_path = File::Spec->abs2rel($t_path);
02b1aeec 731 }
00a8df5c
YO
732}
733
99870f4d
KW
734# Get Unicode version into regular and v-string. This is done now because
735# various tables below get populated based on it. These tables are populated
736# here to be near the top of the file, and so easily seeable by those needing
737# to modify things.
738open my $VERSION, "<", "version"
739 or croak "$0: can't open required file 'version': $!\n";
740my $string_version = <$VERSION>;
741close $VERSION;
742chomp $string_version;
743my $v_version = pack "C*", split /\./, $string_version; # v string
744
745# The following are the complete names of properties with property values that
746# are known to not match any code points in some versions of Unicode, but that
747# may change in the future so they should be matchable, hence an empty file is
748# generated for them.
749my @tables_that_may_be_empty = (
750 'Joining_Type=Left_Joining',
751 );
752push @tables_that_may_be_empty, 'Script=Common' if $v_version le v4.0.1;
753push @tables_that_may_be_empty, 'Title' if $v_version lt v2.0.0;
754push @tables_that_may_be_empty, 'Script=Katakana_Or_Hiragana'
755 if $v_version ge v4.1.0;
756
757# The lists below are hashes, so the key is the item in the list, and the
758# value is the reason why it is in the list. This makes generation of
759# documentation easier.
760
761my %why_suppressed; # No file generated for these.
762
763# Files aren't generated for empty extraneous properties. This is arguable.
764# Extraneous properties generally come about because a property is no longer
765# used in a newer version of Unicode. If we generated a file without code
766# points, programs that used to work on that property will still execute
767# without errors. It just won't ever match (or will always match, with \P{}).
768# This means that the logic is now likely wrong. I (khw) think its better to
769# find this out by getting an error message. Just move them to the table
770# above to change this behavior
771my %why_suppress_if_empty_warn_if_not = (
772
773 # It is the only property that has ever officially been removed from the
774 # Standard. The database never contained any code points for it.
775 'Special_Case_Condition' => 'Obsolete',
776
777 # Apparently never official, but there were code points in some versions of
778 # old-style PropList.txt
779 'Non_Break' => 'Obsolete',
780);
781
782# These would normally go in the warn table just above, but they were changed
783# a long time before this program was written, so warnings about them are
784# moot.
785if ($v_version gt v3.2.0) {
786 push @tables_that_may_be_empty,
787 'Canonical_Combining_Class=Attached_Below_Left'
788}
789
790# These are listed in the Property aliases file in 5.2, but Unihan is ignored
791# unless explicitly added.
792if ($v_version ge v5.2.0) {
793 my $unihan = 'Unihan; remove from list if using Unihan';
ea25a9b2 794 foreach my $table (qw (
99870f4d
KW
795 kAccountingNumeric
796 kOtherNumeric
797 kPrimaryNumeric
798 kCompatibilityVariant
799 kIICore
800 kIRG_GSource
801 kIRG_HSource
802 kIRG_JSource
803 kIRG_KPSource
804 kIRG_MSource
805 kIRG_KSource
806 kIRG_TSource
807 kIRG_USource
808 kIRG_VSource
809 kRSUnicode
ea25a9b2 810 ))
99870f4d
KW
811 {
812 $why_suppress_if_empty_warn_if_not{$table} = $unihan;
813 }
ca12659b
NC
814}
815
99870f4d
KW
816# Properties that this program ignores.
817my @unimplemented_properties = (
818'Unicode_Radical_Stroke' # Remove if changing to handle this one.
819);
d73e5302 820
99870f4d
KW
821# There are several types of obsolete properties defined by Unicode. These
822# must be hand-edited for every new Unicode release.
823my %why_deprecated; # Generates a deprecated warning message if used.
824my %why_stabilized; # Documentation only
825my %why_obsolete; # Documentation only
826
827{ # Closure
828 my $simple = 'Perl uses the more complete version of this property';
829 my $unihan = 'Unihan properties are by default not enabled in the Perl core. Instead use CPAN: Unicode::Unihan';
830
831 my $other_properties = 'other properties';
832 my $contributory = "Used by Unicode internally for generating $other_properties and not intended to be used stand-alone";
833 my $why_no_expand = "Easily computed, and yet doesn't cover the common encoding forms (UTF-16/8)",
834
835 %why_deprecated = (
836 'Grapheme_Link' => 'Deprecated by Unicode. Use ccc=vr (Canonical_Combining_Class=Virama) instead',
837 'Jamo_Short_Name' => $contributory,
838 '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',
839 'Other_Alphabetic' => $contributory,
840 'Other_Default_Ignorable_Code_Point' => $contributory,
841 'Other_Grapheme_Extend' => $contributory,
842 'Other_ID_Continue' => $contributory,
843 'Other_ID_Start' => $contributory,
844 'Other_Lowercase' => $contributory,
845 'Other_Math' => $contributory,
846 'Other_Uppercase' => $contributory,
847 );
848
849 %why_suppressed = (
850 # There is a lib/unicore/Decomposition.pl (used by normalize.pm) which
851 # contains the same information, but without the algorithmically
852 # determinable Hangul syllables'. This file is not published, so it's
853 # existence is not noted in the comment.
854 'Decomposition_Mapping' => 'Accessible via Unicode::Normalize',
855
856 'ISO_Comment' => 'Apparently no demand for it, but can access it through Unicode::UCD::charinfo. Obsoleted, and code points for it removed in Unicode 5.2',
857 'Unicode_1_Name' => "$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",
858
859 'Simple_Case_Folding' => "$simple. Can access this through Unicode::UCD::casefold",
860 'Simple_Lowercase_Mapping' => "$simple. Can access this through Unicode::UCD::charinfo",
861 'Simple_Titlecase_Mapping' => "$simple. Can access this through Unicode::UCD::charinfo",
862 'Simple_Uppercase_Mapping' => "$simple. Can access this through Unicode::UCD::charinfo",
863
864 'Name' => "Accessible via 'use charnames;'",
865 'Name_Alias' => "Accessible via 'use charnames;'",
866
867 # These are sort of jumping the gun; deprecation is proposed for
868 # Unicode version 6.0, but they have never been exposed by Perl, and
869 # likely are soon to be deprecated, so best not to expose them.
870 FC_NFKC_Closure => 'Use NFKC_Casefold instead',
871 Expands_On_NFC => $why_no_expand,
872 Expands_On_NFD => $why_no_expand,
873 Expands_On_NFKC => $why_no_expand,
874 Expands_On_NFKD => $why_no_expand,
875 );
876
877 # The following are suppressed because they were made contributory or
878 # deprecated by Unicode before Perl ever thought about supporting them.
879 foreach my $property ('Jamo_Short_Name', 'Grapheme_Link') {
880 $why_suppressed{$property} = $why_deprecated{$property};
881 }
cf25bb62 882
99870f4d
KW
883 # Customize the message for all the 'Other_' properties
884 foreach my $property (keys %why_deprecated) {
885 next if (my $main_property = $property) !~ s/^Other_//;
886 $why_deprecated{$property} =~ s/$other_properties/the $main_property property (which should be used instead)/;
887 }
888}
889
890if ($v_version ge 4.0.0) {
891 $why_stabilized{'Hyphen'} = 'Use the Line_Break property instead; see www.unicode.org/reports/tr14';
892}
893if ($v_version ge 5.2.0) {
894 $why_obsolete{'ISO_Comment'} = 'Code points for it have been removed';
895}
896
897# Probably obsolete forever
898if ($v_version ge v4.1.0) {
899 $why_suppressed{'Script=Katakana_Or_Hiragana'} = 'Obsolete. All code points previously matched by this have been moved to "Script=Common"';
900}
901
902# This program can create files for enumerated-like properties, such as
903# 'Numeric_Type'. This file would be the same format as for a string
904# property, with a mapping from code point to its value, so you could look up,
905# for example, the script a code point is in. But no one so far wants this
906# mapping, or they have found another way to get it since this is a new
907# feature. So no file is generated except if it is in this list.
908my @output_mapped_properties = split "\n", <<END;
909END
910
911# If you are using the Unihan database, you need to add the properties that
912# you want to extract from it to this table. For your convenience, the
913# properties in the 5.2 PropertyAliases.txt file are listed, commented out
914my @cjk_properties = split "\n", <<'END';
915#cjkAccountingNumeric; kAccountingNumeric
916#cjkOtherNumeric; kOtherNumeric
917#cjkPrimaryNumeric; kPrimaryNumeric
918#cjkCompatibilityVariant; kCompatibilityVariant
919#cjkIICore ; kIICore
920#cjkIRG_GSource; kIRG_GSource
921#cjkIRG_HSource; kIRG_HSource
922#cjkIRG_JSource; kIRG_JSource
923#cjkIRG_KPSource; kIRG_KPSource
924#cjkIRG_KSource; kIRG_KSource
925#cjkIRG_TSource; kIRG_TSource
926#cjkIRG_USource; kIRG_USource
927#cjkIRG_VSource; kIRG_VSource
928#cjkRSUnicode; kRSUnicode ; Unicode_Radical_Stroke; URS
929END
930
931# Similarly for the property values. For your convenience, the lines in the
932# 5.2 PropertyAliases.txt file are listed. Just remove the first BUT NOT both
933# '#' marks
934my @cjk_property_values = split "\n", <<'END';
935## @missing: 0000..10FFFF; cjkAccountingNumeric; NaN
936## @missing: 0000..10FFFF; cjkCompatibilityVariant; <code point>
937## @missing: 0000..10FFFF; cjkIICore; <none>
938## @missing: 0000..10FFFF; cjkIRG_GSource; <none>
939## @missing: 0000..10FFFF; cjkIRG_HSource; <none>
940## @missing: 0000..10FFFF; cjkIRG_JSource; <none>
941## @missing: 0000..10FFFF; cjkIRG_KPSource; <none>
942## @missing: 0000..10FFFF; cjkIRG_KSource; <none>
943## @missing: 0000..10FFFF; cjkIRG_TSource; <none>
944## @missing: 0000..10FFFF; cjkIRG_USource; <none>
945## @missing: 0000..10FFFF; cjkIRG_VSource; <none>
946## @missing: 0000..10FFFF; cjkOtherNumeric; NaN
947## @missing: 0000..10FFFF; cjkPrimaryNumeric; NaN
948## @missing: 0000..10FFFF; cjkRSUnicode; <none>
949END
950
951# The input files don't list every code point. Those not listed are to be
952# defaulted to some value. Below are hard-coded what those values are for
953# non-binary properties as of 5.1. Starting in 5.0, there are
954# machine-parsable comment lines in the files the give the defaults; so this
955# list shouldn't have to be extended. The claim is that all missing entries
956# for binary properties will default to 'N'. Unicode tried to change that in
957# 5.2, but the beta period produced enough protest that they backed off.
958#
959# The defaults for the fields that appear in UnicodeData.txt in this hash must
960# be in the form that it expects. The others may be synonyms.
961my $CODE_POINT = '<code point>';
962my %default_mapping = (
963 Age => "Unassigned",
964 # Bidi_Class => Complicated; set in code
965 Bidi_Mirroring_Glyph => "",
966 Block => 'No_Block',
967 Canonical_Combining_Class => 0,
968 Case_Folding => $CODE_POINT,
969 Decomposition_Mapping => $CODE_POINT,
970 Decomposition_Type => 'None',
971 East_Asian_Width => "Neutral",
972 FC_NFKC_Closure => $CODE_POINT,
973 General_Category => 'Cn',
974 Grapheme_Cluster_Break => 'Other',
975 Hangul_Syllable_Type => 'NA',
976 ISO_Comment => "",
977 Jamo_Short_Name => "",
978 Joining_Group => "No_Joining_Group",
979 # Joining_Type => Complicated; set in code
980 kIICore => 'N', # Is converted to binary
981 #Line_Break => Complicated; set in code
982 Lowercase_Mapping => $CODE_POINT,
983 Name => "",
984 Name_Alias => "",
985 NFC_QC => 'Yes',
986 NFD_QC => 'Yes',
987 NFKC_QC => 'Yes',
988 NFKD_QC => 'Yes',
989 Numeric_Type => 'None',
990 Numeric_Value => 'NaN',
991 Script => ($v_version le 4.1.0) ? 'Common' : 'Unknown',
992 Sentence_Break => 'Other',
993 Simple_Case_Folding => $CODE_POINT,
994 Simple_Lowercase_Mapping => $CODE_POINT,
995 Simple_Titlecase_Mapping => $CODE_POINT,
996 Simple_Uppercase_Mapping => $CODE_POINT,
997 Titlecase_Mapping => $CODE_POINT,
998 Unicode_1_Name => "",
999 Unicode_Radical_Stroke => "",
1000 Uppercase_Mapping => $CODE_POINT,
1001 Word_Break => 'Other',
1002);
1003
1004# Below are files that Unicode furnishes, but this program ignores, and why
1005my %ignored_files = (
1006 'CJKRadicals.txt' => 'Unihan data',
1007 'Index.txt' => 'An index, not actual data',
1008 'NamedSqProv.txt' => 'Not officially part of the Unicode standard; Append it to NamedSequences.txt if you want to process the contents.',
1009 'NamesList.txt' => 'Just adds commentary',
1010 'NormalizationCorrections.txt' => 'Data is already in other files.',
1011 'Props.txt' => 'Adds nothing to PropList.txt; only in very early releases',
1012 'ReadMe.txt' => 'Just comments',
1013 'README.TXT' => 'Just comments',
1014 'StandardizedVariants.txt' => 'Only for glyph changes, not a Unicode character property. Does not fit into current scheme where one code point is mapped',
1015);
1016
678f13d5 1017### End of externally interesting definitions, except for @input_file_objects
99870f4d
KW
1018
1019my $HEADER=<<"EOF";
1020# !!!!!!! DO NOT EDIT THIS FILE !!!!!!!
3df51b85
KW
1021# This file is machine-generated by $0 from the Unicode
1022# database, Version $string_version. Any changes made here will be lost!
cf25bb62
JH
1023EOF
1024
b6922eda 1025my $INTERNAL_ONLY=<<"EOF";
99870f4d
KW
1026
1027# !!!!!!! INTERNAL PERL USE ONLY !!!!!!!
b6922eda 1028# This file is for internal use by the Perl program only. The format and even
99870f4d
KW
1029# the name or existence of this file are subject to change without notice.
1030# Don't use it directly.
1031EOF
1032
1033my $DEVELOPMENT_ONLY=<<"EOF";
1034# !!!!!!! DEVELOPMENT USE ONLY !!!!!!!
1035# This file contains information artificially constrained to code points
1036# present in Unicode release $string_compare_versions.
1037# IT CANNOT BE RELIED ON. It is for use during development only and should
23e33b60 1038# not be used for production.
b6922eda
KW
1039
1040EOF
1041
99870f4d
KW
1042my $LAST_UNICODE_CODEPOINT_STRING = "10FFFF";
1043my $LAST_UNICODE_CODEPOINT = hex $LAST_UNICODE_CODEPOINT_STRING;
1044my $MAX_UNICODE_CODEPOINTS = $LAST_UNICODE_CODEPOINT + 1;
1045
1046# Matches legal code point. 4-6 hex numbers, If there are 6, the first
1047# two must be 10; if there are 5, the first must not be a 0. Written this way
1048# to decrease backtracking
1049my $code_point_re =
1050 qr/ \b (?: 10[0-9A-F]{4} | [1-9A-F][0-9A-F]{4} | [0-9A-F]{4} ) \b/x;
1051
1052# This matches the beginning of the line in the Unicode db files that give the
1053# defaults for code points not listed (i.e., missing) in the file. The code
1054# depends on this ending with a semi-colon, so it can assume it is a valid
1055# field when the line is split() by semi-colons
1056my $missing_defaults_prefix =
1057 qr/^#\s+\@missing:\s+0000\.\.$LAST_UNICODE_CODEPOINT_STRING\s*;/;
1058
1059# Property types. Unicode has more types, but these are sufficient for our
1060# purposes.
1061my $UNKNOWN = -1; # initialized to illegal value
1062my $NON_STRING = 1; # Either binary or enum
1063my $BINARY = 2;
1064my $ENUM = 3; # Include catalog
1065my $STRING = 4; # Anything else: string or misc
1066
1067# Some input files have lines that give default values for code points not
1068# contained in the file. Sometimes these should be ignored.
1069my $NO_DEFAULTS = 0; # Must evaluate to false
1070my $NOT_IGNORED = 1;
1071my $IGNORED = 2;
1072
1073# Range types. Each range has a type. Most ranges are type 0, for normal,
1074# and will appear in the main body of the tables in the output files, but
1075# there are other types of ranges as well, listed below, that are specially
1076# handled. There are pseudo-types as well that will never be stored as a
1077# type, but will affect the calculation of the type.
1078
1079# 0 is for normal, non-specials
1080my $MULTI_CP = 1; # Sequence of more than code point
1081my $HANGUL_SYLLABLE = 2;
1082my $CP_IN_NAME = 3; # The NAME contains the code point appended to it.
1083my $NULL = 4; # The map is to the null string; utf8.c can't
1084 # handle these, nor is there an accepted syntax
1085 # for them in \p{} constructs
f86864ac 1086my $COMPUTE_NO_MULTI_CP = 5; # Pseudo-type; means that ranges that would
99870f4d
KW
1087 # otherwise be $MULTI_CP type are instead type 0
1088
1089# process_generic_property_file() can accept certain overrides in its input.
1090# Each of these must begin AND end with $CMD_DELIM.
1091my $CMD_DELIM = "\a";
1092my $REPLACE_CMD = 'replace'; # Override the Replace
1093my $MAP_TYPE_CMD = 'map_type'; # Override the Type
1094
1095my $NO = 0;
1096my $YES = 1;
1097
1098# Values for the Replace argument to add_range.
1099# $NO # Don't replace; add only the code points not
1100 # already present.
1101my $IF_NOT_EQUIVALENT = 1; # Replace only under certain conditions; details in
1102 # the comments at the subroutine definition.
1103my $UNCONDITIONALLY = 2; # Replace without conditions.
1104my $MULTIPLE = 4; # Don't replace, but add a duplicate record if
1105 # already there
1106
1107# Flags to give property statuses. The phrases are to remind maintainers that
1108# if the flag is changed, the indefinite article referring to it in the
1109# documentation may need to be as well.
1110my $NORMAL = "";
1111my $SUPPRESSED = 'z'; # The character should never actually be seen, since
1112 # it is suppressed
37e2e78e 1113my $PLACEHOLDER = 'P'; # Implies no pod entry generated
99870f4d
KW
1114my $DEPRECATED = 'D';
1115my $a_bold_deprecated = "a 'B<$DEPRECATED>'";
1116my $A_bold_deprecated = "A 'B<$DEPRECATED>'";
1117my $DISCOURAGED = 'X';
1118my $a_bold_discouraged = "an 'B<$DISCOURAGED>'";
1119my $A_bold_discouraged = "An 'B<$DISCOURAGED>'";
1120my $STRICTER = 'T';
1121my $a_bold_stricter = "a 'B<$STRICTER>'";
1122my $A_bold_stricter = "A 'B<$STRICTER>'";
1123my $STABILIZED = 'S';
1124my $a_bold_stabilized = "an 'B<$STABILIZED>'";
1125my $A_bold_stabilized = "An 'B<$STABILIZED>'";
1126my $OBSOLETE = 'O';
1127my $a_bold_obsolete = "an 'B<$OBSOLETE>'";
1128my $A_bold_obsolete = "An 'B<$OBSOLETE>'";
1129
1130my %status_past_participles = (
1131 $DISCOURAGED => 'discouraged',
1132 $SUPPRESSED => 'should never be generated',
1133 $STABILIZED => 'stabilized',
1134 $OBSOLETE => 'obsolete',
37e2e78e 1135 $DEPRECATED => 'deprecated',
99870f4d
KW
1136);
1137
1138# The format of the values of the map tables:
1139my $BINARY_FORMAT = 'b';
1140my $DECIMAL_FORMAT = 'd';
1141my $FLOAT_FORMAT = 'f';
1142my $INTEGER_FORMAT = 'i';
1143my $HEX_FORMAT = 'x';
1144my $RATIONAL_FORMAT = 'r';
1145my $STRING_FORMAT = 's';
1146
1147my %map_table_formats = (
1148 $BINARY_FORMAT => 'binary',
1149 $DECIMAL_FORMAT => 'single decimal digit',
1150 $FLOAT_FORMAT => 'floating point number',
1151 $INTEGER_FORMAT => 'integer',
1152 $HEX_FORMAT => 'positive hex whole number; a code point',
1153 $RATIONAL_FORMAT => 'rational: an integer or a fraction',
1a9d544b 1154 $STRING_FORMAT => 'string',
99870f4d
KW
1155);
1156
1157# Unicode didn't put such derived files in a separate directory at first.
1158my $EXTRACTED_DIR = (-d 'extracted') ? 'extracted' : "";
1159my $EXTRACTED = ($EXTRACTED_DIR) ? "$EXTRACTED_DIR/" : "";
1160my $AUXILIARY = 'auxiliary';
1161
1162# Hashes that will eventually go into Heavy.pl for the use of utf8_heavy.pl
1163my %loose_to_file_of; # loosely maps table names to their respective
1164 # files
1165my %stricter_to_file_of; # same; but for stricter mapping.
1166my %nv_floating_to_rational; # maps numeric values floating point numbers to
1167 # their rational equivalent
1168my %loose_property_name_of; # Loosely maps property names to standard form
1169
1170# These constants names and values were taken from the Unicode standard,
1171# version 5.1, section 3.12. They are used in conjunction with Hangul
6e5a209b
KW
1172# syllables. The '_string' versions are so generated tables can retain the
1173# hex format, which is the more familiar value
1174my $SBase_string = "0xAC00";
1175my $SBase = CORE::hex $SBase_string;
1176my $LBase_string = "0x1100";
1177my $LBase = CORE::hex $LBase_string;
1178my $VBase_string = "0x1161";
1179my $VBase = CORE::hex $VBase_string;
1180my $TBase_string = "0x11A7";
1181my $TBase = CORE::hex $TBase_string;
99870f4d
KW
1182my $SCount = 11172;
1183my $LCount = 19;
1184my $VCount = 21;
1185my $TCount = 28;
1186my $NCount = $VCount * $TCount;
1187
1188# For Hangul syllables; These store the numbers from Jamo.txt in conjunction
1189# with the above published constants.
1190my %Jamo;
1191my %Jamo_L; # Leading consonants
1192my %Jamo_V; # Vowels
1193my %Jamo_T; # Trailing consonants
1194
37e2e78e 1195my @backslash_X_tests; # List of tests read in for testing \X
99870f4d
KW
1196my @unhandled_properties; # Will contain a list of properties found in
1197 # the input that we didn't process.
f86864ac 1198my @match_properties; # Properties that have match tables, to be
99870f4d
KW
1199 # listed in the pod
1200my @map_properties; # Properties that get map files written
1201my @named_sequences; # NamedSequences.txt contents.
1202my %potential_files; # Generated list of all .txt files in the directory
1203 # structure so we can warn if something is being
1204 # ignored.
1205my @files_actually_output; # List of files we generated.
1206my @more_Names; # Some code point names are compound; this is used
1207 # to store the extra components of them.
1208my $MIN_FRACTION_LENGTH = 3; # How many digits of a floating point number at
1209 # the minimum before we consider it equivalent to a
1210 # candidate rational
1211my $MAX_FLOATING_SLOP = 10 ** - $MIN_FRACTION_LENGTH; # And in floating terms
1212
1213# These store references to certain commonly used property objects
1214my $gc;
1215my $perl;
1216my $block;
3e20195b
KW
1217my $perl_charname;
1218my $print;
99870f4d
KW
1219
1220# Are there conflicting names because of beginning with 'In_', or 'Is_'
1221my $has_In_conflicts = 0;
1222my $has_Is_conflicts = 0;
1223
1224sub internal_file_to_platform ($) {
1225 # Convert our file paths which have '/' separators to those of the
1226 # platform.
1227
1228 my $file = shift;
1229 return undef unless defined $file;
1230
1231 return File::Spec->join(split '/', $file);
d07a55ed 1232}
5beb625e 1233
99870f4d
KW
1234sub file_exists ($) { # platform independent '-e'. This program internally
1235 # uses slash as a path separator.
1236 my $file = shift;
1237 return 0 if ! defined $file;
1238 return -e internal_file_to_platform($file);
1239}
5beb625e 1240
99870f4d 1241sub objaddr($) {
23e33b60
KW
1242 # Returns the address of the blessed input object.
1243 # It doesn't check for blessedness because that would do a string eval
1244 # every call, and the program is structured so that this is never called
1245 # for a non-blessed object.
99870f4d 1246
23e33b60 1247 no overloading; # If overloaded, numifying below won't work.
99870f4d
KW
1248
1249 # Numifying a ref gives its address.
051df77b 1250 return pack 'J', $_[0];
99870f4d
KW
1251}
1252
23e33b60
KW
1253# Commented code below should work on Perl 5.8.
1254## This 'require' doesn't necessarily work in miniperl, and even if it does,
1255## the native perl version of it (which is what would operate under miniperl)
1256## is extremely slow, as it does a string eval every call.
1257#my $has_fast_scalar_util = $\18 !~ /miniperl/
1258# && defined eval "require Scalar::Util";
1259#
1260#sub objaddr($) {
1261# # Returns the address of the blessed input object. Uses the XS version if
1262# # available. It doesn't check for blessedness because that would do a
1263# # string eval every call, and the program is structured so that this is
1264# # never called for a non-blessed object.
1265#
1266# return Scalar::Util::refaddr($_[0]) if $has_fast_scalar_util;
1267#
1268# # Check at least that is a ref.
1269# my $pkg = ref($_[0]) or return undef;
1270#
1271# # Change to a fake package to defeat any overloaded stringify
1272# bless $_[0], 'main::Fake';
1273#
1274# # Numifying a ref gives its address.
051df77b 1275# my $addr = pack 'J', $_[0];
23e33b60
KW
1276#
1277# # Return to original class
1278# bless $_[0], $pkg;
1279# return $addr;
1280#}
1281
99870f4d
KW
1282sub max ($$) {
1283 my $a = shift;
1284 my $b = shift;
1285 return $a if $a >= $b;
1286 return $b;
1287}
1288
1289sub min ($$) {
1290 my $a = shift;
1291 my $b = shift;
1292 return $a if $a <= $b;
1293 return $b;
1294}
1295
1296sub clarify_number ($) {
1297 # This returns the input number with underscores inserted every 3 digits
1298 # in large (5 digits or more) numbers. Input must be entirely digits, not
1299 # checked.
1300
1301 my $number = shift;
1302 my $pos = length($number) - 3;
1303 return $number if $pos <= 1;
1304 while ($pos > 0) {
1305 substr($number, $pos, 0) = '_';
1306 $pos -= 3;
5beb625e 1307 }
99870f4d 1308 return $number;
99598c8c
JH
1309}
1310
12ac2576 1311
99870f4d 1312package Carp;
7ebf06b3 1313
99870f4d
KW
1314# These routines give a uniform treatment of messages in this program. They
1315# are placed in the Carp package to cause the stack trace to not include them,
1316# although an alternative would be to use another package and set @CARP_NOT
1317# for it.
12ac2576 1318
99870f4d 1319our $Verbose = 1 if main::DEBUG; # Useful info when debugging
12ac2576 1320
99f78760
KW
1321# This is a work-around suggested by Nicholas Clark to fix a problem with Carp
1322# and overload trying to load Scalar:Util under miniperl. See
1323# http://www.xray.mpe.mpg.de/mailing-lists/perl5-porters/2009-11/msg01057.html
1324undef $overload::VERSION;
1325
99870f4d
KW
1326sub my_carp {
1327 my $message = shift || "";
1328 my $nofold = shift || 0;
7ebf06b3 1329
99870f4d
KW
1330 if ($message) {
1331 $message = main::join_lines($message);
1332 $message =~ s/^$0: *//; # Remove initial program name
1333 $message =~ s/[.;,]+$//; # Remove certain ending punctuation
1334 $message = "\n$0: $message;";
12ac2576 1335
99870f4d
KW
1336 # Fold the message with program name, semi-colon end punctuation
1337 # (which looks good with the message that carp appends to it), and a
1338 # hanging indent for continuation lines.
1339 $message = main::simple_fold($message, "", 4) unless $nofold;
1340 $message =~ s/\n$//; # Remove the trailing nl so what carp
1341 # appends is to the same line
1342 }
12ac2576 1343
99870f4d 1344 return $message if defined wantarray; # If a caller just wants the msg
12ac2576 1345
99870f4d
KW
1346 carp $message;
1347 return;
1348}
7ebf06b3 1349
99870f4d
KW
1350sub my_carp_bug {
1351 # This is called when it is clear that the problem is caused by a bug in
1352 # this program.
7ebf06b3 1353
99870f4d
KW
1354 my $message = shift;
1355 $message =~ s/^$0: *//;
1356 $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");
1357 carp $message;
1358 return;
1359}
7ebf06b3 1360
99870f4d
KW
1361sub carp_too_few_args {
1362 if (@_ != 2) {
1363 my_carp_bug("Wrong number of arguments: to 'carp_too_few_arguments'. No action taken.");
1364 return;
12ac2576 1365 }
7ebf06b3 1366
99870f4d
KW
1367 my $args_ref = shift;
1368 my $count = shift;
7ebf06b3 1369
99870f4d
KW
1370 my_carp_bug("Need at least $count arguments to "
1371 . (caller 1)[3]
1372 . ". Instead got: '"
1373 . join ', ', @$args_ref
1374 . "'. No action taken.");
1375 return;
12ac2576
JP
1376}
1377
99870f4d
KW
1378sub carp_extra_args {
1379 my $args_ref = shift;
1380 my_carp_bug("Too many arguments to 'carp_extra_args': (" . join(', ', @_) . "); Extras ignored.") if @_;
12ac2576 1381
99870f4d
KW
1382 unless (ref $args_ref) {
1383 my_carp_bug("Argument to 'carp_extra_args' ($args_ref) must be a ref. Not checking arguments.");
1384 return;
1385 }
1386 my ($package, $file, $line) = caller;
1387 my $subroutine = (caller 1)[3];
cf25bb62 1388
99870f4d
KW
1389 my $list;
1390 if (ref $args_ref eq 'HASH') {
1391 foreach my $key (keys %$args_ref) {
1392 $args_ref->{$key} = $UNDEF unless defined $args_ref->{$key};
cf25bb62 1393 }
99870f4d 1394 $list = join ', ', each %{$args_ref};
cf25bb62 1395 }
99870f4d
KW
1396 elsif (ref $args_ref eq 'ARRAY') {
1397 foreach my $arg (@$args_ref) {
1398 $arg = $UNDEF unless defined $arg;
1399 }
1400 $list = join ', ', @$args_ref;
1401 }
1402 else {
1403 my_carp_bug("Can't cope with ref "
1404 . ref($args_ref)
1405 . " . argument to 'carp_extra_args'. Not checking arguments.");
1406 return;
1407 }
1408
1409 my_carp_bug("Unrecognized parameters in options: '$list' to $subroutine. Skipped.");
1410 return;
d73e5302
JH
1411}
1412
99870f4d
KW
1413package main;
1414
1415{ # Closure
1416
1417 # This program uses the inside-out method for objects, as recommended in
1418 # "Perl Best Practices". This closure aids in generating those. There
1419 # are two routines. setup_package() is called once per package to set
1420 # things up, and then set_access() is called for each hash representing a
1421 # field in the object. These routines arrange for the object to be
1422 # properly destroyed when no longer used, and for standard accessor
1423 # functions to be generated. If you need more complex accessors, just
1424 # write your own and leave those accesses out of the call to set_access().
1425 # More details below.
1426
1427 my %constructor_fields; # fields that are to be used in constructors; see
1428 # below
1429
1430 # The values of this hash will be the package names as keys to other
1431 # hashes containing the name of each field in the package as keys, and
1432 # references to their respective hashes as values.
1433 my %package_fields;
1434
1435 sub setup_package {
1436 # Sets up the package, creating standard DESTROY and dump methods
1437 # (unless already defined). The dump method is used in debugging by
1438 # simple_dumper().
1439 # The optional parameters are:
1440 # a) a reference to a hash, that gets populated by later
1441 # set_access() calls with one of the accesses being
1442 # 'constructor'. The caller can then refer to this, but it is
1443 # not otherwise used by these two routines.
1444 # b) a reference to a callback routine to call during destruction
1445 # of the object, before any fields are actually destroyed
1446
1447 my %args = @_;
1448 my $constructor_ref = delete $args{'Constructor_Fields'};
1449 my $destroy_callback = delete $args{'Destroy_Callback'};
1450 Carp::carp_extra_args(\@_) if main::DEBUG && %args;
1451
1452 my %fields;
1453 my $package = (caller)[0];
1454
1455 $package_fields{$package} = \%fields;
1456 $constructor_fields{$package} = $constructor_ref;
1457
1458 unless ($package->can('DESTROY')) {
1459 my $destroy_name = "${package}::DESTROY";
1460 no strict "refs";
1461
1462 # Use typeglob to give the anonymous subroutine the name we want
1463 *$destroy_name = sub {
1464 my $self = shift;
ffe43484 1465 my $addr = do { no overloading; pack 'J', $self; };
99870f4d
KW
1466
1467 $self->$destroy_callback if $destroy_callback;
1468 foreach my $field (keys %{$package_fields{$package}}) {
1469 #print STDERR __LINE__, ": Destroying ", ref $self, " ", sprintf("%04X", $addr), ": ", $field, "\n";
1470 delete $package_fields{$package}{$field}{$addr};
1471 }
1472 return;
1473 }
1474 }
1475
1476 unless ($package->can('dump')) {
1477 my $dump_name = "${package}::dump";
1478 no strict "refs";
1479 *$dump_name = sub {
1480 my $self = shift;
1481 return dump_inside_out($self, $package_fields{$package}, @_);
1482 }
1483 }
1484 return;
1485 }
1486
1487 sub set_access {
1488 # Arrange for the input field to be garbage collected when no longer
1489 # needed. Also, creates standard accessor functions for the field
1490 # based on the optional parameters-- none if none of these parameters:
1491 # 'addable' creates an 'add_NAME()' accessor function.
1492 # 'readable' or 'readable_array' creates a 'NAME()' accessor
1493 # function.
1494 # 'settable' creates a 'set_NAME()' accessor function.
1495 # 'constructor' doesn't create an accessor function, but adds the
1496 # field to the hash that was previously passed to
1497 # setup_package();
1498 # Any of the accesses can be abbreviated down, so that 'a', 'ad',
1499 # 'add' etc. all mean 'addable'.
1500 # The read accessor function will work on both array and scalar
1501 # values. If another accessor in the parameter list is 'a', the read
1502 # access assumes an array. You can also force it to be array access
1503 # by specifying 'readable_array' instead of 'readable'
1504 #
1505 # A sort-of 'protected' access can be set-up by preceding the addable,
1506 # readable or settable with some initial portion of 'protected_' (but,
1507 # the underscore is required), like 'p_a', 'pro_set', etc. The
1508 # "protection" is only by convention. All that happens is that the
1509 # accessor functions' names begin with an underscore. So instead of
1510 # calling set_foo, the call is _set_foo. (Real protection could be
c1739a4a 1511 # accomplished by having a new subroutine, end_package, called at the
99870f4d
KW
1512 # end of each package, and then storing the __LINE__ ranges and
1513 # checking them on every accessor. But that is way overkill.)
1514
1515 # We create anonymous subroutines as the accessors and then use
1516 # typeglobs to assign them to the proper package and name
1517
1518 my $name = shift; # Name of the field
1519 my $field = shift; # Reference to the inside-out hash containing the
1520 # field
1521
1522 my $package = (caller)[0];
1523
1524 if (! exists $package_fields{$package}) {
1525 croak "$0: Must call 'setup_package' before 'set_access'";
1526 }
d73e5302 1527
99870f4d
KW
1528 # Stash the field so DESTROY can get it.
1529 $package_fields{$package}{$name} = $field;
cf25bb62 1530
99870f4d
KW
1531 # Remaining arguments are the accessors. For each...
1532 foreach my $access (@_) {
1533 my $access = lc $access;
cf25bb62 1534
99870f4d 1535 my $protected = "";
cf25bb62 1536
99870f4d
KW
1537 # Match the input as far as it goes.
1538 if ($access =~ /^(p[^_]*)_/) {
1539 $protected = $1;
1540 if (substr('protected_', 0, length $protected)
1541 eq $protected)
1542 {
1543
1544 # Add 1 for the underscore not included in $protected
1545 $access = substr($access, length($protected) + 1);
1546 $protected = '_';
1547 }
1548 else {
1549 $protected = "";
1550 }
1551 }
1552
1553 if (substr('addable', 0, length $access) eq $access) {
1554 my $subname = "${package}::${protected}add_$name";
1555 no strict "refs";
1556
1557 # add_ accessor. Don't add if already there, which we
1558 # determine using 'eq' for scalars and '==' otherwise.
1559 *$subname = sub {
1560 use strict "refs";
1561 return Carp::carp_too_few_args(\@_, 2) if main::DEBUG && @_ < 2;
1562 my $self = shift;
1563 my $value = shift;
ffe43484 1564 my $addr = do { no overloading; pack 'J', $self; };
99870f4d
KW
1565 Carp::carp_extra_args(\@_) if main::DEBUG && @_;
1566 if (ref $value) {
f998e60c 1567 return if grep { $value == $_ } @{$field->{$addr}};
99870f4d
KW
1568 }
1569 else {
f998e60c 1570 return if grep { $value eq $_ } @{$field->{$addr}};
99870f4d 1571 }
f998e60c 1572 push @{$field->{$addr}}, $value;
99870f4d
KW
1573 return;
1574 }
1575 }
1576 elsif (substr('constructor', 0, length $access) eq $access) {
1577 if ($protected) {
1578 Carp::my_carp_bug("Can't set-up 'protected' constructors")
1579 }
1580 else {
1581 $constructor_fields{$package}{$name} = $field;
1582 }
1583 }
1584 elsif (substr('readable_array', 0, length $access) eq $access) {
1585
1586 # Here has read access. If one of the other parameters for
1587 # access is array, or this one specifies array (by being more
1588 # than just 'readable_'), then create a subroutine that
1589 # assumes the data is an array. Otherwise just a scalar
1590 my $subname = "${package}::${protected}$name";
1591 if (grep { /^a/i } @_
1592 or length($access) > length('readable_'))
1593 {
1594 no strict "refs";
1595 *$subname = sub {
1596 use strict "refs";
23e33b60 1597 Carp::carp_extra_args(\@_) if main::DEBUG && @_ > 1;
ffe43484 1598 my $addr = do { no overloading; pack 'J', $_[0]; };
99870f4d
KW
1599 if (ref $field->{$addr} ne 'ARRAY') {
1600 my $type = ref $field->{$addr};
1601 $type = 'scalar' unless $type;
1602 Carp::my_carp_bug("Trying to read $name as an array when it is a $type. Big problems.");
1603 return;
1604 }
1605 return scalar @{$field->{$addr}} unless wantarray;
1606
1607 # Make a copy; had problems with caller modifying the
1608 # original otherwise
1609 my @return = @{$field->{$addr}};
1610 return @return;
1611 }
1612 }
1613 else {
1614
1615 # Here not an array value, a simpler function.
1616 no strict "refs";
1617 *$subname = sub {
1618 use strict "refs";
23e33b60 1619 Carp::carp_extra_args(\@_) if main::DEBUG && @_ > 1;
f998e60c 1620 no overloading;
051df77b 1621 return $field->{pack 'J', $_[0]};
99870f4d
KW
1622 }
1623 }
1624 }
1625 elsif (substr('settable', 0, length $access) eq $access) {
1626 my $subname = "${package}::${protected}set_$name";
1627 no strict "refs";
1628 *$subname = sub {
1629 use strict "refs";
23e33b60
KW
1630 if (main::DEBUG) {
1631 return Carp::carp_too_few_args(\@_, 2) if @_ < 2;
1632 Carp::carp_extra_args(\@_) if @_ > 2;
1633 }
1634 # $self is $_[0]; $value is $_[1]
f998e60c 1635 no overloading;
051df77b 1636 $field->{pack 'J', $_[0]} = $_[1];
99870f4d
KW
1637 return;
1638 }
1639 }
1640 else {
1641 Carp::my_carp_bug("Unknown accessor type $access. No accessor set.");
1642 }
cf25bb62 1643 }
99870f4d 1644 return;
cf25bb62 1645 }
99870f4d
KW
1646}
1647
1648package Input_file;
1649
1650# All input files use this object, which stores various attributes about them,
1651# and provides for convenient, uniform handling. The run method wraps the
1652# processing. It handles all the bookkeeping of opening, reading, and closing
1653# the file, returning only significant input lines.
1654#
1655# Each object gets a handler which processes the body of the file, and is
1656# called by run(). Most should use the generic, default handler, which has
1657# code scrubbed to handle things you might not expect. A handler should
1658# basically be a while(next_line()) {...} loop.
1659#
1660# You can also set up handlers to
1661# 1) call before the first line is read for pre processing
1662# 2) call to adjust each line of the input before the main handler gets them
1663# 3) call upon EOF before the main handler exits its loop
1664# 4) call at the end for post processing
1665#
1666# $_ is used to store the input line, and is to be filtered by the
1667# each_line_handler()s. So, if the format of the line is not in the desired
1668# format for the main handler, these are used to do that adjusting. They can
1669# be stacked (by enclosing them in an [ anonymous array ] in the constructor,
1670# so the $_ output of one is used as the input to the next. None of the other
1671# handlers are stackable, but could easily be changed to be so.
1672#
1673# Most of the handlers can call insert_lines() or insert_adjusted_lines()
1674# which insert the parameters as lines to be processed before the next input
1675# file line is read. This allows the EOF handler to flush buffers, for
1676# example. The difference between the two routines is that the lines inserted
1677# by insert_lines() are subjected to the each_line_handler()s. (So if you
1678# called it from such a handler, you would get infinite recursion.) Lines
1679# inserted by insert_adjusted_lines() go directly to the main handler without
1680# any adjustments. If the post-processing handler calls any of these, there
1681# will be no effect. Some error checking for these conditions could be added,
1682# but it hasn't been done.
1683#
1684# carp_bad_line() should be called to warn of bad input lines, which clears $_
1685# to prevent further processing of the line. This routine will output the
1686# message as a warning once, and then keep a count of the lines that have the
1687# same message, and output that count at the end of the file's processing.
1688# This keeps the number of messages down to a manageable amount.
1689#
1690# get_missings() should be called to retrieve any @missing input lines.
1691# Messages will be raised if this isn't done if the options aren't to ignore
1692# missings.
1693
1694sub trace { return main::trace(@_); }
1695
99870f4d
KW
1696{ # Closure
1697 # Keep track of fields that are to be put into the constructor.
1698 my %constructor_fields;
1699
1700 main::setup_package(Constructor_Fields => \%constructor_fields);
1701
1702 my %file; # Input file name, required
1703 main::set_access('file', \%file, qw{ c r });
1704
1705 my %first_released; # Unicode version file was first released in, required
1706 main::set_access('first_released', \%first_released, qw{ c r });
1707
1708 my %handler; # Subroutine to process the input file, defaults to
1709 # 'process_generic_property_file'
1710 main::set_access('handler', \%handler, qw{ c });
1711
1712 my %property;
1713 # name of property this file is for. defaults to none, meaning not
1714 # applicable, or is otherwise determinable, for example, from each line.
1715 main::set_access('property', \%property, qw{ c });
1716
1717 my %optional;
1718 # If this is true, the file is optional. If not present, no warning is
1719 # output. If it is present, the string given by this parameter is
1720 # evaluated, and if false the file is not processed.
1721 main::set_access('optional', \%optional, 'c', 'r');
1722
1723 my %non_skip;
1724 # This is used for debugging, to skip processing of all but a few input
1725 # files. Add 'non_skip => 1' to the constructor for those files you want
1726 # processed when you set the $debug_skip global.
1727 main::set_access('non_skip', \%non_skip, 'c');
1728
37e2e78e
KW
1729 my %skip;
1730 # This is used to skip processing of this input file semi-permanently.
1731 # It is used for files that we aren't planning to process anytime soon,
1732 # but want to allow to be in the directory and not raise a message that we
1733 # are not handling. Mostly for test files. This is in contrast to the
1734 # non_skip element, which is supposed to be used very temporarily for
1735 # debugging. Sets 'optional' to 1
1736 main::set_access('skip', \%skip, 'c');
1737
99870f4d
KW
1738 my %each_line_handler;
1739 # list of subroutines to look at and filter each non-comment line in the
1740 # file. defaults to none. The subroutines are called in order, each is
1741 # to adjust $_ for the next one, and the final one adjusts it for
1742 # 'handler'
1743 main::set_access('each_line_handler', \%each_line_handler, 'c');
1744
1745 my %has_missings_defaults;
1746 # ? Are there lines in the file giving default values for code points
1747 # missing from it?. Defaults to NO_DEFAULTS. Otherwise NOT_IGNORED is
1748 # the norm, but IGNORED means it has such lines, but the handler doesn't
1749 # use them. Having these three states allows us to catch changes to the
1750 # UCD that this program should track
1751 main::set_access('has_missings_defaults',
1752 \%has_missings_defaults, qw{ c r });
1753
1754 my %pre_handler;
1755 # Subroutine to call before doing anything else in the file. If undef, no
1756 # such handler is called.
1757 main::set_access('pre_handler', \%pre_handler, qw{ c });
1758
1759 my %eof_handler;
1760 # Subroutine to call upon getting an EOF on the input file, but before
1761 # that is returned to the main handler. This is to allow buffers to be
1762 # flushed. The handler is expected to call insert_lines() or
1763 # insert_adjusted() with the buffered material
1764 main::set_access('eof_handler', \%eof_handler, qw{ c r });
1765
1766 my %post_handler;
1767 # Subroutine to call after all the lines of the file are read in and
1768 # processed. If undef, no such handler is called.
1769 main::set_access('post_handler', \%post_handler, qw{ c });
1770
1771 my %progress_message;
1772 # Message to print to display progress in lieu of the standard one
1773 main::set_access('progress_message', \%progress_message, qw{ c });
1774
1775 my %handle;
1776 # cache open file handle, internal. Is undef if file hasn't been
1777 # processed at all, empty if has;
1778 main::set_access('handle', \%handle);
1779
1780 my %added_lines;
1781 # cache of lines added virtually to the file, internal
1782 main::set_access('added_lines', \%added_lines);
1783
1784 my %errors;
1785 # cache of errors found, internal
1786 main::set_access('errors', \%errors);
1787
1788 my %missings;
1789 # storage of '@missing' defaults lines
1790 main::set_access('missings', \%missings);
1791
1792 sub new {
1793 my $class = shift;
1794
1795 my $self = bless \do{ my $anonymous_scalar }, $class;
ffe43484 1796 my $addr = do { no overloading; pack 'J', $self; };
99870f4d
KW
1797
1798 # Set defaults
1799 $handler{$addr} = \&main::process_generic_property_file;
1800 $non_skip{$addr} = 0;
37e2e78e 1801 $skip{$addr} = 0;
99870f4d
KW
1802 $has_missings_defaults{$addr} = $NO_DEFAULTS;
1803 $handle{$addr} = undef;
1804 $added_lines{$addr} = [ ];
1805 $each_line_handler{$addr} = [ ];
1806 $errors{$addr} = { };
1807 $missings{$addr} = [ ];
1808
1809 # Two positional parameters.
99f78760 1810 return Carp::carp_too_few_args(\@_, 2) if main::DEBUG && @_ < 2;
99870f4d
KW
1811 $file{$addr} = main::internal_file_to_platform(shift);
1812 $first_released{$addr} = shift;
1813
1814 # The rest of the arguments are key => value pairs
1815 # %constructor_fields has been set up earlier to list all possible
1816 # ones. Either set or push, depending on how the default has been set
1817 # up just above.
1818 my %args = @_;
1819 foreach my $key (keys %args) {
1820 my $argument = $args{$key};
1821
1822 # Note that the fields are the lower case of the constructor keys
1823 my $hash = $constructor_fields{lc $key};
1824 if (! defined $hash) {
1825 Carp::my_carp_bug("Unrecognized parameters '$key => $argument' to new() for $self. Skipped");
1826 next;
1827 }
1828 if (ref $hash->{$addr} eq 'ARRAY') {
1829 if (ref $argument eq 'ARRAY') {
1830 foreach my $argument (@{$argument}) {
1831 next if ! defined $argument;
1832 push @{$hash->{$addr}}, $argument;
1833 }
1834 }
1835 else {
1836 push @{$hash->{$addr}}, $argument if defined $argument;
1837 }
1838 }
1839 else {
1840 $hash->{$addr} = $argument;
1841 }
1842 delete $args{$key};
1843 };
1844
1845 # If the file has a property for it, it means that the property is not
1846 # listed in the file's entries. So add a handler to the list of line
1847 # handlers to insert the property name into the lines, to provide a
1848 # uniform interface to the final processing subroutine.
1849 # the final code doesn't have to worry about that.
1850 if ($property{$addr}) {
1851 push @{$each_line_handler{$addr}}, \&_insert_property_into_line;
1852 }
1853
1854 if ($non_skip{$addr} && ! $debug_skip && $verbosity) {
1855 print "Warning: " . __PACKAGE__ . " constructor for $file{$addr} has useless 'non_skip' in it\n";
a3a8c5f0 1856 }
99870f4d 1857
37e2e78e
KW
1858 $optional{$addr} = 1 if $skip{$addr};
1859
99870f4d 1860 return $self;
d73e5302
JH
1861 }
1862
cf25bb62 1863
99870f4d
KW
1864 use overload
1865 fallback => 0,
1866 qw("") => "_operator_stringify",
1867 "." => \&main::_operator_dot,
1868 ;
cf25bb62 1869
99870f4d
KW
1870 sub _operator_stringify {
1871 my $self = shift;
cf25bb62 1872
99870f4d 1873 return __PACKAGE__ . " object for " . $self->file;
d73e5302 1874 }
d73e5302 1875
99870f4d
KW
1876 # flag to make sure extracted files are processed early
1877 my $seen_non_extracted_non_age = 0;
d73e5302 1878
99870f4d
KW
1879 sub run {
1880 # Process the input object $self. This opens and closes the file and
1881 # calls all the handlers for it. Currently, this can only be called
1882 # once per file, as it destroy's the EOF handler
d73e5302 1883
99870f4d
KW
1884 my $self = shift;
1885 Carp::carp_extra_args(\@_) if main::DEBUG && @_;
b6922eda 1886
ffe43484 1887 my $addr = do { no overloading; pack 'J', $self; };
b6922eda 1888
99870f4d 1889 my $file = $file{$addr};
d73e5302 1890
99870f4d
KW
1891 # Don't process if not expecting this file (because released later
1892 # than this Unicode version), and isn't there. This means if someone
1893 # copies it into an earlier version's directory, we will go ahead and
1894 # process it.
1895 return if $first_released{$addr} gt $v_version && ! -e $file;
1896
1897 # If in debugging mode and this file doesn't have the non-skip
1898 # flag set, and isn't one of the critical files, skip it.
1899 if ($debug_skip
1900 && $first_released{$addr} ne v0
1901 && ! $non_skip{$addr})
1902 {
1903 print "Skipping $file in debugging\n" if $verbosity;
1904 return;
1905 }
1906
1907 # File could be optional
37e2e78e 1908 if ($optional{$addr}) {
99870f4d
KW
1909 return unless -e $file;
1910 my $result = eval $optional{$addr};
1911 if (! defined $result) {
1912 Carp::my_carp_bug("Got '$@' when tried to eval $optional{$addr}. $file Skipped.");
1913 return;
1914 }
1915 if (! $result) {
1916 if ($verbosity) {
1917 print STDERR "Skipping processing input file '$file' because '$optional{$addr}' is not true\n";
1918 }
1919 return;
1920 }
1921 }
1922
1923 if (! defined $file || ! -e $file) {
1924
1925 # If the file doesn't exist, see if have internal data for it
1926 # (based on first_released being 0).
1927 if ($first_released{$addr} eq v0) {
1928 $handle{$addr} = 'pretend_is_open';
1929 }
1930 else {
1931 if (! $optional{$addr} # File could be optional
1932 && $v_version ge $first_released{$addr})
1933 {
1934 print STDERR "Skipping processing input file '$file' because not found\n" if $v_version ge $first_released{$addr};
1935 }
1936 return;
1937 }
1938 }
1939 else {
1940
37e2e78e
KW
1941 # Here, the file exists. Some platforms may change the case of
1942 # its name
99870f4d 1943 if ($seen_non_extracted_non_age) {
517956bf 1944 if ($file =~ /$EXTRACTED/i) {
99870f4d 1945 Carp::my_carp_bug(join_lines(<<END
99f78760 1946$file should be processed just after the 'Prop...Alias' files, and before
99870f4d
KW
1947anything not in the $EXTRACTED_DIR directory. Proceeding, but the results may
1948have subtle problems
1949END
1950 ));
1951 }
1952 }
1953 elsif ($EXTRACTED_DIR
1954 && $first_released{$addr} ne v0
517956bf
CB
1955 && $file !~ /$EXTRACTED/i
1956 && lc($file) ne 'dage.txt')
99870f4d
KW
1957 {
1958 # We don't set this (by the 'if' above) if we have no
1959 # extracted directory, so if running on an early version,
1960 # this test won't work. Not worth worrying about.
1961 $seen_non_extracted_non_age = 1;
1962 }
1963
1964 # And mark the file as having being processed, and warn if it
1965 # isn't a file we are expecting. As we process the files,
1966 # they are deleted from the hash, so any that remain at the
1967 # end of the program are files that we didn't process.
517956bf
CB
1968 my $fkey = File::Spec->rel2abs($file);
1969 my $expecting = delete $potential_files{$fkey};
1970 $expecting = delete $potential_files{lc($fkey)} unless defined $expecting;
678f13d5
KW
1971 Carp::my_carp("Was not expecting '$file'.") if
1972 ! $expecting
99870f4d
KW
1973 && ! defined $handle{$addr};
1974
37e2e78e
KW
1975 # Having deleted from expected files, we can quit if not to do
1976 # anything. Don't print progress unless really want verbosity
1977 if ($skip{$addr}) {
1978 print "Skipping $file.\n" if $verbosity >= $VERBOSE;
1979 return;
1980 }
1981
99870f4d
KW
1982 # Open the file, converting the slashes used in this program
1983 # into the proper form for the OS
1984 my $file_handle;
1985 if (not open $file_handle, "<", $file) {
1986 Carp::my_carp("Can't open $file. Skipping: $!");
1987 return 0;
1988 }
1989 $handle{$addr} = $file_handle; # Cache the open file handle
1990 }
1991
1992 if ($verbosity >= $PROGRESS) {
1993 if ($progress_message{$addr}) {
1994 print "$progress_message{$addr}\n";
1995 }
1996 else {
1997 # If using a virtual file, say so.
1998 print "Processing ", (-e $file)
1999 ? $file
2000 : "substitute $file",
2001 "\n";
2002 }
2003 }
2004
2005
2006 # Call any special handler for before the file.
2007 &{$pre_handler{$addr}}($self) if $pre_handler{$addr};
2008
2009 # Then the main handler
2010 &{$handler{$addr}}($self);
2011
2012 # Then any special post-file handler.
2013 &{$post_handler{$addr}}($self) if $post_handler{$addr};
2014
2015 # If any errors have been accumulated, output the counts (as the first
2016 # error message in each class was output when it was encountered).
2017 if ($errors{$addr}) {
2018 my $total = 0;
2019 my $types = 0;
2020 foreach my $error (keys %{$errors{$addr}}) {
2021 $total += $errors{$addr}->{$error};
2022 delete $errors{$addr}->{$error};
2023 $types++;
2024 }
2025 if ($total > 1) {
2026 my $message
2027 = "A total of $total lines had errors in $file. ";
2028
2029 $message .= ($types == 1)
2030 ? '(Only the first one was displayed.)'
2031 : '(Only the first of each type was displayed.)';
2032 Carp::my_carp($message);
2033 }
2034 }
2035
2036 if (@{$missings{$addr}}) {
2037 Carp::my_carp_bug("Handler for $file didn't look at all the \@missing lines. Generated tables likely are wrong");
2038 }
2039
2040 # If a real file handle, close it.
2041 close $handle{$addr} or Carp::my_carp("Can't close $file: $!") if
2042 ref $handle{$addr};
2043 $handle{$addr} = ""; # Uses empty to indicate that has already seen
2044 # the file, as opposed to undef
2045 return;
2046 }
2047
2048 sub next_line {
2049 # Sets $_ to be the next logical input line, if any. Returns non-zero
2050 # if such a line exists. 'logical' means that any lines that have
2051 # been added via insert_lines() will be returned in $_ before the file
2052 # is read again.
2053
2054 my $self = shift;
2055 Carp::carp_extra_args(\@_) if main::DEBUG && @_;
2056
ffe43484 2057 my $addr = do { no overloading; pack 'J', $self; };
99870f4d
KW
2058
2059 # Here the file is open (or if the handle is not a ref, is an open
2060 # 'virtual' file). Get the next line; any inserted lines get priority
2061 # over the file itself.
2062 my $adjusted;
2063
2064 LINE:
2065 while (1) { # Loop until find non-comment, non-empty line
2066 #local $to_trace = 1 if main::DEBUG;
2067 my $inserted_ref = shift @{$added_lines{$addr}};
2068 if (defined $inserted_ref) {
2069 ($adjusted, $_) = @{$inserted_ref};
2070 trace $adjusted, $_ if main::DEBUG && $to_trace;
2071 return 1 if $adjusted;
2072 }
2073 else {
2074 last if ! ref $handle{$addr}; # Don't read unless is real file
2075 last if ! defined ($_ = readline $handle{$addr});
2076 }
2077 chomp;
2078 trace $_ if main::DEBUG && $to_trace;
2079
2080 # See if this line is the comment line that defines what property
2081 # value that code points that are not listed in the file should
2082 # have. The format or existence of these lines is not guaranteed
2083 # by Unicode since they are comments, but the documentation says
2084 # that this was added for machine-readability, so probably won't
2085 # change. This works starting in Unicode Version 5.0. They look
2086 # like:
2087 #
2088 # @missing: 0000..10FFFF; Not_Reordered
2089 # @missing: 0000..10FFFF; Decomposition_Mapping; <code point>
2090 # @missing: 0000..10FFFF; ; NaN
2091 #
2092 # Save the line for a later get_missings() call.
2093 if (/$missing_defaults_prefix/) {
2094 if ($has_missings_defaults{$addr} == $NO_DEFAULTS) {
2095 $self->carp_bad_line("Unexpected \@missing line. Assuming no missing entries");
2096 }
2097 elsif ($has_missings_defaults{$addr} == $NOT_IGNORED) {
2098 my @defaults = split /\s* ; \s*/x, $_;
2099
2100 # The first field is the @missing, which ends in a
2101 # semi-colon, so can safely shift.
2102 shift @defaults;
2103
2104 # Some of these lines may have empty field placeholders
2105 # which get in the way. An example is:
2106 # @missing: 0000..10FFFF; ; NaN
2107 # Remove them. Process starting from the top so the
2108 # splice doesn't affect things still to be looked at.
2109 for (my $i = @defaults - 1; $i >= 0; $i--) {
2110 next if $defaults[$i] ne "";
2111 splice @defaults, $i, 1;
2112 }
2113
2114 # What's left should be just the property (maybe) and the
2115 # default. Having only one element means it doesn't have
2116 # the property.
2117 my $default;
2118 my $property;
2119 if (@defaults >= 1) {
2120 if (@defaults == 1) {
2121 $default = $defaults[0];
2122 }
2123 else {
2124 $property = $defaults[0];
2125 $default = $defaults[1];
2126 }
2127 }
2128
2129 if (@defaults < 1
2130 || @defaults > 2
2131 || ($default =~ /^</
2132 && $default !~ /^<code *point>$/i
2133 && $default !~ /^<none>$/i))
2134 {
2135 $self->carp_bad_line("Unrecognized \@missing line: $_. Assuming no missing entries");
2136 }
2137 else {
2138
2139 # If the property is missing from the line, it should
2140 # be the one for the whole file
2141 $property = $property{$addr} if ! defined $property;
2142
2143 # Change <none> to the null string, which is what it
2144 # really means. If the default is the code point
2145 # itself, set it to <code point>, which is what
2146 # Unicode uses (but sometimes they've forgotten the
2147 # space)
2148 if ($default =~ /^<none>$/i) {
2149 $default = "";
2150 }
2151 elsif ($default =~ /^<code *point>$/i) {
2152 $default = $CODE_POINT;
2153 }
2154
2155 # Store them as a sub-arrays with both components.
2156 push @{$missings{$addr}}, [ $default, $property ];
2157 }
2158 }
2159
2160 # There is nothing for the caller to process on this comment
2161 # line.
2162 next;
2163 }
2164
2165 # Remove comments and trailing space, and skip this line if the
2166 # result is empty
2167 s/#.*//;
2168 s/\s+$//;
2169 next if /^$/;
2170
2171 # Call any handlers for this line, and skip further processing of
2172 # the line if the handler sets the line to null.
2173 foreach my $sub_ref (@{$each_line_handler{$addr}}) {
2174 &{$sub_ref}($self);
2175 next LINE if /^$/;
2176 }
2177
2178 # Here the line is ok. return success.
2179 return 1;
2180 } # End of looping through lines.
2181
2182 # If there is an EOF handler, call it (only once) and if it generates
2183 # more lines to process go back in the loop to handle them.
2184 if ($eof_handler{$addr}) {
2185 &{$eof_handler{$addr}}($self);
2186 $eof_handler{$addr} = ""; # Currently only get one shot at it.
2187 goto LINE if $added_lines{$addr};
2188 }
2189
2190 # Return failure -- no more lines.
2191 return 0;
2192
2193 }
2194
2195# Not currently used, not fully tested.
2196# sub peek {
2197# # Non-destructive look-ahead one non-adjusted, non-comment, non-blank
2198# # record. Not callable from an each_line_handler(), nor does it call
2199# # an each_line_handler() on the line.
2200#
2201# my $self = shift;
ffe43484 2202# my $addr = do { no overloading; pack 'J', $self; };
99870f4d
KW
2203#
2204# foreach my $inserted_ref (@{$added_lines{$addr}}) {
2205# my ($adjusted, $line) = @{$inserted_ref};
2206# next if $adjusted;
2207#
2208# # Remove comments and trailing space, and return a non-empty
2209# # resulting line
2210# $line =~ s/#.*//;
2211# $line =~ s/\s+$//;
2212# return $line if $line ne "";
2213# }
2214#
2215# return if ! ref $handle{$addr}; # Don't read unless is real file
2216# while (1) { # Loop until find non-comment, non-empty line
2217# local $to_trace = 1 if main::DEBUG;
2218# trace $_ if main::DEBUG && $to_trace;
2219# return if ! defined (my $line = readline $handle{$addr});
2220# chomp $line;
2221# push @{$added_lines{$addr}}, [ 0, $line ];
2222#
2223# $line =~ s/#.*//;
2224# $line =~ s/\s+$//;
2225# return $line if $line ne "";
2226# }
2227#
2228# return;
2229# }
2230
2231
2232 sub insert_lines {
2233 # Lines can be inserted so that it looks like they were in the input
2234 # file at the place it was when this routine is called. See also
2235 # insert_adjusted_lines(). Lines inserted via this routine go through
2236 # any each_line_handler()
2237
2238 my $self = shift;
2239
2240 # Each inserted line is an array, with the first element being 0 to
2241 # indicate that this line hasn't been adjusted, and needs to be
2242 # processed.
f998e60c 2243 no overloading;
051df77b 2244 push @{$added_lines{pack 'J', $self}}, map { [ 0, $_ ] } @_;
99870f4d
KW
2245 return;
2246 }
2247
2248 sub insert_adjusted_lines {
2249 # Lines can be inserted so that it looks like they were in the input
2250 # file at the place it was when this routine is called. See also
2251 # insert_lines(). Lines inserted via this routine are already fully
2252 # adjusted, ready to be processed; each_line_handler()s handlers will
2253 # not be called. This means this is not a completely general
2254 # facility, as only the last each_line_handler on the stack should
2255 # call this. It could be made more general, by passing to each of the
2256 # line_handlers their position on the stack, which they would pass on
2257 # to this routine, and that would replace the boolean first element in
2258 # the anonymous array pushed here, so that the next_line routine could
2259 # use that to call only those handlers whose index is after it on the
2260 # stack. But this is overkill for what is needed now.
2261
2262 my $self = shift;
2263 trace $_[0] if main::DEBUG && $to_trace;
2264
2265 # Each inserted line is an array, with the first element being 1 to
2266 # indicate that this line has been adjusted
f998e60c 2267 no overloading;
051df77b 2268 push @{$added_lines{pack 'J', $self}}, map { [ 1, $_ ] } @_;
99870f4d
KW
2269 return;
2270 }
2271
2272 sub get_missings {
2273 # Returns the stored up @missings lines' values, and clears the list.
2274 # The values are in an array, consisting of the default in the first
2275 # element, and the property in the 2nd. However, since these lines
2276 # can be stacked up, the return is an array of all these arrays.
2277
2278 my $self = shift;
2279 Carp::carp_extra_args(\@_) if main::DEBUG && @_;
2280
ffe43484 2281 my $addr = do { no overloading; pack 'J', $self; };
99870f4d
KW
2282
2283 # If not accepting a list return, just return the first one.
2284 return shift @{$missings{$addr}} unless wantarray;
2285
2286 my @return = @{$missings{$addr}};
2287 undef @{$missings{$addr}};
2288 return @return;
2289 }
2290
2291 sub _insert_property_into_line {
2292 # Add a property field to $_, if this file requires it.
2293
f998e60c 2294 my $self = shift;
ffe43484 2295 my $addr = do { no overloading; pack 'J', $self; };
f998e60c 2296 my $property = $property{$addr};
99870f4d
KW
2297 Carp::carp_extra_args(\@_) if main::DEBUG && @_;
2298
2299 $_ =~ s/(;|$)/; $property$1/;
2300 return;
2301 }
2302
2303 sub carp_bad_line {
2304 # Output consistent error messages, using either a generic one, or the
2305 # one given by the optional parameter. To avoid gazillions of the
2306 # same message in case the syntax of a file is way off, this routine
2307 # only outputs the first instance of each message, incrementing a
2308 # count so the totals can be output at the end of the file.
2309
2310 my $self = shift;
2311 my $message = shift;
2312 Carp::carp_extra_args(\@_) if main::DEBUG && @_;
2313
ffe43484 2314 my $addr = do { no overloading; pack 'J', $self; };
99870f4d
KW
2315
2316 $message = 'Unexpected line' unless $message;
2317
2318 # No trailing punctuation so as to fit with our addenda.
2319 $message =~ s/[.:;,]$//;
2320
2321 # If haven't seen this exact message before, output it now. Otherwise
2322 # increment the count of how many times it has occurred
2323 unless ($errors{$addr}->{$message}) {
2324 Carp::my_carp("$message in '$_' in "
f998e60c 2325 . $file{$addr}
99870f4d
KW
2326 . " at line $.. Skipping this line;");
2327 $errors{$addr}->{$message} = 1;
2328 }
2329 else {
2330 $errors{$addr}->{$message}++;
2331 }
2332
2333 # Clear the line to prevent any further (meaningful) processing of it.
2334 $_ = "";
2335
2336 return;
2337 }
2338} # End closure
2339
2340package Multi_Default;
2341
2342# Certain properties in early versions of Unicode had more than one possible
2343# default for code points missing from the files. In these cases, one
2344# default applies to everything left over after all the others are applied,
2345# and for each of the others, there is a description of which class of code
2346# points applies to it. This object helps implement this by storing the
2347# defaults, and for all but that final default, an eval string that generates
2348# the class that it applies to.
2349
2350
2351{ # Closure
2352
2353 main::setup_package();
2354
2355 my %class_defaults;
2356 # The defaults structure for the classes
2357 main::set_access('class_defaults', \%class_defaults);
2358
2359 my %other_default;
2360 # The default that applies to everything left over.
2361 main::set_access('other_default', \%other_default, 'r');
2362
2363
2364 sub new {
2365 # The constructor is called with default => eval pairs, terminated by
2366 # the left-over default. e.g.
2367 # Multi_Default->new(
2368 # 'T' => '$gc->table("Mn") + $gc->table("Cf") - 0x200C
2369 # - 0x200D',
2370 # 'R' => 'some other expression that evaluates to code points',
2371 # .
2372 # .
2373 # .
2374 # 'U'));
2375
2376 my $class = shift;
2377
2378 my $self = bless \do{my $anonymous_scalar}, $class;
ffe43484 2379 my $addr = do { no overloading; pack 'J', $self; };
99870f4d
KW
2380
2381 while (@_ > 1) {
2382 my $default = shift;
2383 my $eval = shift;
2384 $class_defaults{$addr}->{$default} = $eval;
2385 }
2386
2387 $other_default{$addr} = shift;
2388
2389 return $self;
2390 }
2391
2392 sub get_next_defaults {
2393 # Iterates and returns the next class of defaults.
2394 my $self = shift;
2395 Carp::carp_extra_args(\@_) if main::DEBUG && @_;
2396
ffe43484 2397 my $addr = do { no overloading; pack 'J', $self; };
99870f4d
KW
2398
2399 return each %{$class_defaults{$addr}};
2400 }
2401}
2402
2403package Alias;
2404
2405# An alias is one of the names that a table goes by. This class defines them
2406# including some attributes. Everything is currently setup in the
2407# constructor.
2408
2409
2410{ # Closure
2411
2412 main::setup_package();
2413
2414 my %name;
2415 main::set_access('name', \%name, 'r');
2416
2417 my %loose_match;
2418 # Determined by the constructor code if this name should match loosely or
2419 # not. The constructor parameters can override this, but it isn't fully
2420 # implemented, as should have ability to override Unicode one's via
2421 # something like a set_loose_match()
2422 main::set_access('loose_match', \%loose_match, 'r');
2423
2424 my %make_pod_entry;
2425 # Some aliases should not get their own entries because they are covered
2426 # by a wild-card, and some we want to discourage use of. Binary
2427 main::set_access('make_pod_entry', \%make_pod_entry, 'r');
2428
2429 my %status;
2430 # Aliases have a status, like deprecated, or even suppressed (which means
2431 # they don't appear in documentation). Enum
2432 main::set_access('status', \%status, 'r');
2433
2434 my %externally_ok;
2435 # Similarly, some aliases should not be considered as usable ones for
2436 # external use, such as file names, or we don't want documentation to
2437 # recommend them. Boolean
2438 main::set_access('externally_ok', \%externally_ok, 'r');
2439
2440 sub new {
2441 my $class = shift;
2442
2443 my $self = bless \do { my $anonymous_scalar }, $class;
ffe43484 2444 my $addr = do { no overloading; pack 'J', $self; };
99870f4d
KW
2445
2446 $name{$addr} = shift;
2447 $loose_match{$addr} = shift;
2448 $make_pod_entry{$addr} = shift;
2449 $externally_ok{$addr} = shift;
2450 $status{$addr} = shift;
2451
2452 Carp::carp_extra_args(\@_) if main::DEBUG && @_;
2453
2454 # Null names are never ok externally
2455 $externally_ok{$addr} = 0 if $name{$addr} eq "";
2456
2457 return $self;
2458 }
2459}
2460
2461package Range;
2462
2463# A range is the basic unit for storing code points, and is described in the
2464# comments at the beginning of the program. Each range has a starting code
2465# point; an ending code point (not less than the starting one); a value
2466# that applies to every code point in between the two end-points, inclusive;
2467# and an enum type that applies to the value. The type is for the user's
2468# convenience, and has no meaning here, except that a non-zero type is
2469# considered to not obey the normal Unicode rules for having standard forms.
2470#
2471# The same structure is used for both map and match tables, even though in the
2472# latter, the value (and hence type) is irrelevant and could be used as a
2473# comment. In map tables, the value is what all the code points in the range
2474# map to. Type 0 values have the standardized version of the value stored as
2475# well, so as to not have to recalculate it a lot.
2476
2477sub trace { return main::trace(@_); }
2478
2479{ # Closure
2480
2481 main::setup_package();
2482
2483 my %start;
2484 main::set_access('start', \%start, 'r', 's');
2485
2486 my %end;
2487 main::set_access('end', \%end, 'r', 's');
2488
2489 my %value;
2490 main::set_access('value', \%value, 'r');
2491
2492 my %type;
2493 main::set_access('type', \%type, 'r');
2494
2495 my %standard_form;
2496 # The value in internal standard form. Defined only if the type is 0.
2497 main::set_access('standard_form', \%standard_form);
2498
2499 # Note that if these fields change, the dump() method should as well
2500
2501 sub new {
2502 return Carp::carp_too_few_args(\@_, 3) if main::DEBUG && @_ < 3;
2503 my $class = shift;
2504
2505 my $self = bless \do { my $anonymous_scalar }, $class;
ffe43484 2506 my $addr = do { no overloading; pack 'J', $self; };
99870f4d
KW
2507
2508 $start{$addr} = shift;
2509 $end{$addr} = shift;
2510
2511 my %args = @_;
2512
2513 my $value = delete $args{'Value'}; # Can be 0
2514 $value = "" unless defined $value;
2515 $value{$addr} = $value;
2516
2517 $type{$addr} = delete $args{'Type'} || 0;
2518
2519 Carp::carp_extra_args(\%args) if main::DEBUG && %args;
2520
2521 if (! $type{$addr}) {
2522 $standard_form{$addr} = main::standardize($value);
2523 }
2524
2525 return $self;
2526 }
2527
2528 use overload
2529 fallback => 0,
2530 qw("") => "_operator_stringify",
2531 "." => \&main::_operator_dot,
2532 ;
2533
2534 sub _operator_stringify {
2535 my $self = shift;
ffe43484 2536 my $addr = do { no overloading; pack 'J', $self; };
99870f4d
KW
2537
2538 # Output it like '0041..0065 (value)'
2539 my $return = sprintf("%04X", $start{$addr})
2540 . '..'
2541 . sprintf("%04X", $end{$addr});
2542 my $value = $value{$addr};
2543 my $type = $type{$addr};
2544 $return .= ' (';
2545 $return .= "$value";
2546 $return .= ", Type=$type" if $type != 0;
2547 $return .= ')';
2548
2549 return $return;
2550 }
2551
2552 sub standard_form {
2553 # The standard form is the value itself if the standard form is
2554 # undefined (that is if the value is special)
2555
2556 my $self = shift;
2557 Carp::carp_extra_args(\@_) if main::DEBUG && @_;
2558
ffe43484 2559 my $addr = do { no overloading; pack 'J', $self; };
99870f4d
KW
2560
2561 return $standard_form{$addr} if defined $standard_form{$addr};
2562 return $value{$addr};
2563 }
2564
2565 sub dump {
2566 # Human, not machine readable. For machine readable, comment out this
2567 # entire routine and let the standard one take effect.
2568 my $self = shift;
2569 my $indent = shift;
2570 Carp::carp_extra_args(\@_) if main::DEBUG && @_;
2571
ffe43484 2572 my $addr = do { no overloading; pack 'J', $self; };
99870f4d
KW
2573
2574 my $return = $indent
2575 . sprintf("%04X", $start{$addr})
2576 . '..'
2577 . sprintf("%04X", $end{$addr})
2578 . " '$value{$addr}';";
2579 if (! defined $standard_form{$addr}) {
2580 $return .= "(type=$type{$addr})";
2581 }
2582 elsif ($standard_form{$addr} ne $value{$addr}) {
2583 $return .= "(standard '$standard_form{$addr}')";
2584 }
2585 return $return;
2586 }
2587} # End closure
2588
2589package _Range_List_Base;
2590
2591# Base class for range lists. A range list is simply an ordered list of
2592# ranges, so that the ranges with the lowest starting numbers are first in it.
2593#
2594# When a new range is added that is adjacent to an existing range that has the
2595# same value and type, it merges with it to form a larger range.
2596#
2597# Ranges generally do not overlap, except that there can be multiple entries
2598# of single code point ranges. This is because of NameAliases.txt.
2599#
2600# In this program, there is a standard value such that if two different
2601# values, have the same standard value, they are considered equivalent. This
2602# value was chosen so that it gives correct results on Unicode data
2603
2604# There are a number of methods to manipulate range lists, and some operators
2605# are overloaded to handle them.
2606
99870f4d
KW
2607sub trace { return main::trace(@_); }
2608
2609{ # Closure
2610
2611 our $addr;
2612
2613 main::setup_package();
2614
2615 my %ranges;
2616 # The list of ranges
2617 main::set_access('ranges', \%ranges, 'readable_array');
2618
2619 my %max;
2620 # The highest code point in the list. This was originally a method, but
2621 # actual measurements said it was used a lot.
2622 main::set_access('max', \%max, 'r');
2623
2624 my %each_range_iterator;
2625 # Iterator position for each_range()
2626 main::set_access('each_range_iterator', \%each_range_iterator);
2627
2628 my %owner_name_of;
2629 # Name of parent this is attached to, if any. Solely for better error
2630 # messages.
2631 main::set_access('owner_name_of', \%owner_name_of, 'p_r');
2632
2633 my %_search_ranges_cache;
2634 # A cache of the previous result from _search_ranges(), for better
2635 # performance
2636 main::set_access('_search_ranges_cache', \%_search_ranges_cache);
2637
2638 sub new {
2639 my $class = shift;
2640 my %args = @_;
2641
2642 # Optional initialization data for the range list.
2643 my $initialize = delete $args{'Initialize'};
2644
2645 my $self;
2646
2647 # Use _union() to initialize. _union() returns an object of this
2648 # class, which means that it will call this constructor recursively.
2649 # But it won't have this $initialize parameter so that it won't
2650 # infinitely loop on this.
2651 return _union($class, $initialize, %args) if defined $initialize;
2652
2653 $self = bless \do { my $anonymous_scalar }, $class;
ffe43484 2654 my $addr = do { no overloading; pack 'J', $self; };
99870f4d
KW
2655
2656 # Optional parent object, only for debug info.
2657 $owner_name_of{$addr} = delete $args{'Owner'};
2658 $owner_name_of{$addr} = "" if ! defined $owner_name_of{$addr};
2659
2660 # Stringify, in case it is an object.
2661 $owner_name_of{$addr} = "$owner_name_of{$addr}";
2662
2663 # This is used only for error messages, and so a colon is added
2664 $owner_name_of{$addr} .= ": " if $owner_name_of{$addr} ne "";
2665
2666 Carp::carp_extra_args(\%args) if main::DEBUG && %args;
2667
2668 # Max is initialized to a negative value that isn't adjacent to 0,
2669 # for simpler tests
2670 $max{$addr} = -2;
2671
2672 $_search_ranges_cache{$addr} = 0;
2673 $ranges{$addr} = [];
2674
2675 return $self;
2676 }
2677
2678 use overload
2679 fallback => 0,
2680 qw("") => "_operator_stringify",
2681 "." => \&main::_operator_dot,
2682 ;
2683
2684 sub _operator_stringify {
2685 my $self = shift;
ffe43484 2686 my $addr = do { no overloading; pack 'J', $self; };
99870f4d
KW
2687
2688 return "Range_List attached to '$owner_name_of{$addr}'"
2689 if $owner_name_of{$addr};
2690 return "anonymous Range_List " . \$self;
2691 }
2692
2693 sub _union {
2694 # Returns the union of the input code points. It can be called as
2695 # either a constructor or a method. If called as a method, the result
2696 # will be a new() instance of the calling object, containing the union
2697 # of that object with the other parameter's code points; if called as
2698 # a constructor, the first parameter gives the class the new object
2699 # should be, and the second parameter gives the code points to go into
2700 # it.
2701 # In either case, there are two parameters looked at by this routine;
2702 # any additional parameters are passed to the new() constructor.
2703 #
2704 # The code points can come in the form of some object that contains
2705 # ranges, and has a conventionally named method to access them; or
2706 # they can be an array of individual code points (as integers); or
2707 # just a single code point.
2708 #
2709 # If they are ranges, this routine doesn't make any effort to preserve
2710 # the range values of one input over the other. Therefore this base
2711 # class should not allow _union to be called from other than
2712 # initialization code, so as to prevent two tables from being added
2713 # together where the range values matter. The general form of this
2714 # routine therefore belongs in a derived class, but it was moved here
2715 # to avoid duplication of code. The failure to overload this in this
2716 # class keeps it safe.
2717 #
2718
2719 my $self;
2720 my @args; # Arguments to pass to the constructor
2721
2722 my $class = shift;
2723
2724 # If a method call, will start the union with the object itself, and
2725 # the class of the new object will be the same as self.
2726 if (ref $class) {
2727 $self = $class;
2728 $class = ref $self;
2729 push @args, $self;
2730 }
2731
2732 # Add the other required parameter.
2733 push @args, shift;
2734 # Rest of parameters are passed on to the constructor
2735
2736 # Accumulate all records from both lists.
2737 my @records;
2738 for my $arg (@args) {
2739 #local $to_trace = 0 if main::DEBUG;
2740 trace "argument = $arg" if main::DEBUG && $to_trace;
2741 if (! defined $arg) {
2742 my $message = "";
2743 if (defined $self) {
f998e60c 2744 no overloading;
051df77b 2745 $message .= $owner_name_of{pack 'J', $self};
99870f4d
KW
2746 }
2747 Carp::my_carp_bug($message .= "Undefined argument to _union. No union done.");
2748 return;
2749 }
2750 $arg = [ $arg ] if ! ref $arg;
2751 my $type = ref $arg;
2752 if ($type eq 'ARRAY') {
2753 foreach my $element (@$arg) {
2754 push @records, Range->new($element, $element);
2755 }
2756 }
2757 elsif ($arg->isa('Range')) {
2758 push @records, $arg;
2759 }
2760 elsif ($arg->can('ranges')) {
2761 push @records, $arg->ranges;
2762 }
2763 else {
2764 my $message = "";
2765 if (defined $self) {
f998e60c 2766 no overloading;
051df77b 2767 $message .= $owner_name_of{pack 'J', $self};
99870f4d
KW
2768 }
2769 Carp::my_carp_bug($message . "Cannot take the union of a $type. No union done.");
2770 return;
2771 }
2772 }
2773
2774 # Sort with the range containing the lowest ordinal first, but if
2775 # two ranges start at the same code point, sort with the bigger range
2776 # of the two first, because it takes fewer cycles.
2777 @records = sort { ($a->start <=> $b->start)
2778 or
2779 # if b is shorter than a, b->end will be
2780 # less than a->end, and we want to select
2781 # a, so want to return -1
2782 ($b->end <=> $a->end)
2783 } @records;
2784
2785 my $new = $class->new(@_);
2786
2787 # Fold in records so long as they add new information.
2788 for my $set (@records) {
2789 my $start = $set->start;
2790 my $end = $set->end;
2791 my $value = $set->value;
2792 if ($start > $new->max) {
2793 $new->_add_delete('+', $start, $end, $value);
2794 }
2795 elsif ($end > $new->max) {
2796 $new->_add_delete('+', $new->max +1, $end, $value);
2797 }
2798 }
2799
2800 return $new;
2801 }
2802
2803 sub range_count { # Return the number of ranges in the range list
2804 my $self = shift;
2805 Carp::carp_extra_args(\@_) if main::DEBUG && @_;
2806
f998e60c 2807 no overloading;
051df77b 2808 return scalar @{$ranges{pack 'J', $self}};
99870f4d
KW
2809 }
2810
2811 sub min {
2812 # Returns the minimum code point currently in the range list, or if
2813 # the range list is empty, 2 beyond the max possible. This is a
2814 # method because used so rarely, that not worth saving between calls,
2815 # and having to worry about changing it as ranges are added and
2816 # deleted.
2817
2818 my $self = shift;
2819 Carp::carp_extra_args(\@_) if main::DEBUG && @_;
2820
ffe43484 2821 my $addr = do { no overloading; pack 'J', $self; };
99870f4d
KW
2822
2823 # If the range list is empty, return a large value that isn't adjacent
2824 # to any that could be in the range list, for simpler tests
2825 return $LAST_UNICODE_CODEPOINT + 2 unless scalar @{$ranges{$addr}};
2826 return $ranges{$addr}->[0]->start;
2827 }
2828
2829 sub contains {
2830 # Boolean: Is argument in the range list? If so returns $i such that:
2831 # range[$i]->end < $codepoint <= range[$i+1]->end
2832 # which is one beyond what you want; this is so that the 0th range
2833 # doesn't return false
2834 my $self = shift;
2835 my $codepoint = shift;
2836 Carp::carp_extra_args(\@_) if main::DEBUG && @_;
2837
99870f4d
KW
2838 my $i = $self->_search_ranges($codepoint);
2839 return 0 unless defined $i;
2840
2841 # The search returns $i, such that
2842 # range[$i-1]->end < $codepoint <= range[$i]->end
2843 # So is in the table if and only iff it is at least the start position
2844 # of range $i.
f998e60c 2845 no overloading;
051df77b 2846 return 0 if $ranges{pack 'J', $self}->[$i]->start > $codepoint;
99870f4d
KW
2847 return $i + 1;
2848 }
2849
2850 sub value_of {
2851 # Returns the value associated with the code point, undef if none
2852
2853 my $self = shift;
2854 my $codepoint = shift;
2855 Carp::carp_extra_args(\@_) if main::DEBUG && @_;
2856
99870f4d
KW
2857 my $i = $self->contains($codepoint);
2858 return unless $i;
2859
2860 # contains() returns 1 beyond where we should look
f998e60c 2861 no overloading;
051df77b 2862 return $ranges{pack 'J', $self}->[$i-1]->value;
99870f4d
KW
2863 }
2864
2865 sub _search_ranges {
2866 # Find the range in the list which contains a code point, or where it
2867 # should go if were to add it. That is, it returns $i, such that:
2868 # range[$i-1]->end < $codepoint <= range[$i]->end
2869 # Returns undef if no such $i is possible (e.g. at end of table), or
2870 # if there is an error.
2871
2872 my $self = shift;
2873 my $code_point = shift;
2874 Carp::carp_extra_args(\@_) if main::DEBUG && @_;
2875
ffe43484 2876 my $addr = do { no overloading; pack 'J', $self; };
99870f4d
KW
2877
2878 return if $code_point > $max{$addr};
2879 my $r = $ranges{$addr}; # The current list of ranges
2880 my $range_list_size = scalar @$r;
2881 my $i;
2882
2883 use integer; # want integer division
2884
2885 # Use the cached result as the starting guess for this one, because,
2886 # an experiment on 5.1 showed that 90% of the time the cache was the
2887 # same as the result on the next call (and 7% it was one less).
2888 $i = $_search_ranges_cache{$addr};
2889 $i = 0 if $i >= $range_list_size; # Reset if no longer valid (prob.
2890 # from an intervening deletion
2891 #local $to_trace = 1 if main::DEBUG;
2892 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);
2893 return $i if $code_point <= $r->[$i]->end
2894 && ($i == 0 || $r->[$i-1]->end < $code_point);
2895
2896 # Here the cache doesn't yield the correct $i. Try adding 1.
2897 if ($i < $range_list_size - 1
2898 && $r->[$i]->end < $code_point &&
2899 $code_point <= $r->[$i+1]->end)
2900 {
2901 $i++;
2902 trace "next \$i is correct: $i" if main::DEBUG && $to_trace;
2903 $_search_ranges_cache{$addr} = $i;
2904 return $i;
2905 }
2906
2907 # Here, adding 1 also didn't work. We do a binary search to
2908 # find the correct position, starting with current $i
2909 my $lower = 0;
2910 my $upper = $range_list_size - 1;
2911 while (1) {
2912 trace "top of loop i=$i:", sprintf("%04X", $r->[$lower]->start), "[$lower] .. ", sprintf("%04X", $r->[$i]->start), "[$i] .. ", sprintf("%04X", $r->[$upper]->start), "[$upper]" if main::DEBUG && $to_trace;
2913
2914 if ($code_point <= $r->[$i]->end) {
2915
2916 # Here we have met the upper constraint. We can quit if we
2917 # also meet the lower one.
2918 last if $i == 0 || $r->[$i-1]->end < $code_point;
2919
2920 $upper = $i; # Still too high.
2921
2922 }
2923 else {
2924
2925 # Here, $r[$i]->end < $code_point, so look higher up.
2926 $lower = $i;
2927 }
2928
2929 # Split search domain in half to try again.
2930 my $temp = ($upper + $lower) / 2;
2931
2932 # No point in continuing unless $i changes for next time
2933 # in the loop.
2934 if ($temp == $i) {
2935
2936 # We can't reach the highest element because of the averaging.
2937 # So if one below the upper edge, force it there and try one
2938 # more time.
2939 if ($i == $range_list_size - 2) {
2940
2941 trace "Forcing to upper edge" if main::DEBUG && $to_trace;
2942 $i = $range_list_size - 1;
2943
2944 # Change $lower as well so if fails next time through,
2945 # taking the average will yield the same $i, and we will
2946 # quit with the error message just below.
2947 $lower = $i;
2948 next;
2949 }
2950 Carp::my_carp_bug("$owner_name_of{$addr}Can't find where the range ought to go. No action taken.");
2951 return;
2952 }
2953 $i = $temp;
2954 } # End of while loop
2955
2956 if (main::DEBUG && $to_trace) {
2957 trace 'i-1=[', $i-1, ']', $r->[$i-1] if $i;
2958 trace "i= [ $i ]", $r->[$i];
2959 trace 'i+1=[', $i+1, ']', $r->[$i+1] if $i < $range_list_size - 1;
2960 }
2961
2962 # Here we have found the offset. Cache it as a starting point for the
2963 # next call.
2964 $_search_ranges_cache{$addr} = $i;
2965 return $i;
2966 }
2967
2968 sub _add_delete {
2969 # Add, replace or delete ranges to or from a list. The $type
2970 # parameter gives which:
2971 # '+' => insert or replace a range, returning a list of any changed
2972 # ranges.
2973 # '-' => delete a range, returning a list of any deleted ranges.
2974 #
2975 # The next three parameters give respectively the start, end, and
2976 # value associated with the range. 'value' should be null unless the
2977 # operation is '+';
2978 #
2979 # The range list is kept sorted so that the range with the lowest
2980 # starting position is first in the list, and generally, adjacent
c1739a4a 2981 # ranges with the same values are merged into a single larger one (see
99870f4d
KW
2982 # exceptions below).
2983 #
c1739a4a 2984 # There are more parameters; all are key => value pairs:
99870f4d
KW
2985 # Type gives the type of the value. It is only valid for '+'.
2986 # All ranges have types; if this parameter is omitted, 0 is
2987 # assumed. Ranges with type 0 are assumed to obey the
2988 # Unicode rules for casing, etc; ranges with other types are
2989 # not. Otherwise, the type is arbitrary, for the caller's
2990 # convenience, and looked at only by this routine to keep
2991 # adjacent ranges of different types from being merged into
2992 # a single larger range, and when Replace =>
2993 # $IF_NOT_EQUIVALENT is specified (see just below).
2994 # Replace determines what to do if the range list already contains
2995 # ranges which coincide with all or portions of the input
2996 # range. It is only valid for '+':
2997 # => $NO means that the new value is not to replace
2998 # any existing ones, but any empty gaps of the
2999 # range list coinciding with the input range
3000 # will be filled in with the new value.
3001 # => $UNCONDITIONALLY means to replace the existing values with
3002 # this one unconditionally. However, if the
3003 # new and old values are identical, the
3004 # replacement is skipped to save cycles
3005 # => $IF_NOT_EQUIVALENT means to replace the existing values
3006 # with this one if they are not equivalent.
3007 # Ranges are equivalent if their types are the
c1739a4a 3008 # same, and they are the same string; or if
99870f4d
KW
3009 # both are type 0 ranges, if their Unicode
3010 # standard forms are identical. In this last
3011 # case, the routine chooses the more "modern"
3012 # one to use. This is because some of the
3013 # older files are formatted with values that
3014 # are, for example, ALL CAPs, whereas the
3015 # derived files have a more modern style,
3016 # which looks better. By looking for this
3017 # style when the pre-existing and replacement
3018 # standard forms are the same, we can move to
3019 # the modern style
3020 # => $MULTIPLE means that if this range duplicates an
3021 # existing one, but has a different value,
3022 # don't replace the existing one, but insert
3023 # this, one so that the same range can occur
3024 # multiple times.
3025 # => anything else is the same as => $IF_NOT_EQUIVALENT
3026 #
c1739a4a
KW
3027 # "same value" means identical for non-type-0 ranges, and it means
3028 # having the same standard forms for type-0 ranges.
99870f4d
KW
3029
3030 return Carp::carp_too_few_args(\@_, 5) if main::DEBUG && @_ < 5;
3031
3032 my $self = shift;
3033 my $operation = shift; # '+' for add/replace; '-' for delete;
3034 my $start = shift;
3035 my $end = shift;
3036 my $value = shift;
3037
3038 my %args = @_;
3039
3040 $value = "" if not defined $value; # warning: $value can be "0"
3041
3042 my $replace = delete $args{'Replace'};
3043 $replace = $IF_NOT_EQUIVALENT unless defined $replace;
3044
3045 my $type = delete $args{'Type'};
3046 $type = 0 unless defined $type;
3047
3048 Carp::carp_extra_args(\%args) if main::DEBUG && %args;
3049
ffe43484 3050 my $addr = do { no overloading; pack 'J', $self; };
99870f4d
KW
3051
3052 if ($operation ne '+' && $operation ne '-') {
3053 Carp::my_carp_bug("$owner_name_of{$addr}First parameter to _add_delete must be '+' or '-'. No action taken.");
3054 return;
3055 }
3056 unless (defined $start && defined $end) {
3057 Carp::my_carp_bug("$owner_name_of{$addr}Undefined start and/or end to _add_delete. No action taken.");
3058 return;
3059 }
3060 unless ($end >= $start) {
3061 Carp::my_carp_bug("$owner_name_of{$addr}End of range (" . sprintf("%04X", $end) . ") must not be before start (" . sprintf("%04X", $start) . "). No action taken.");
3062 return;
3063 }
3064 #local $to_trace = 1 if main::DEBUG;
3065
3066 if ($operation eq '-') {
3067 if ($replace != $IF_NOT_EQUIVALENT) {
3068 Carp::my_carp_bug("$owner_name_of{$addr}Replace => \$IF_NOT_EQUIVALENT is required when deleting a range from a range list. Assuming Replace => \$IF_NOT_EQUIVALENT.");
3069 $replace = $IF_NOT_EQUIVALENT;
3070 }
3071 if ($type) {
3072 Carp::my_carp_bug("$owner_name_of{$addr}Type => 0 is required when deleting a range from a range list. Assuming Type => 0.");
3073 $type = 0;
3074 }
3075 if ($value ne "") {
3076 Carp::my_carp_bug("$owner_name_of{$addr}Value => \"\" is required when deleting a range from a range list. Assuming Value => \"\".");
3077 $value = "";
3078 }
3079 }
3080
3081 my $r = $ranges{$addr}; # The current list of ranges
3082 my $range_list_size = scalar @$r; # And its size
3083 my $max = $max{$addr}; # The current high code point in
3084 # the list of ranges
3085
3086 # Do a special case requiring fewer machine cycles when the new range
3087 # starts after the current highest point. The Unicode input data is
3088 # structured so this is common.
3089 if ($start > $max) {
3090
3091 trace "$owner_name_of{$addr} $operation", sprintf("%04X", $start) . '..' . sprintf("%04X", $end) . " ($value) type=$type" if main::DEBUG && $to_trace;
3092 return if $operation eq '-'; # Deleting a non-existing range is a
3093 # no-op
3094
3095 # If the new range doesn't logically extend the current final one
3096 # in the range list, create a new range at the end of the range
3097 # list. (max cleverly is initialized to a negative number not
3098 # adjacent to 0 if the range list is empty, so even adding a range
3099 # to an empty range list starting at 0 will have this 'if'
3100 # succeed.)
3101 if ($start > $max + 1 # non-adjacent means can't extend.
3102 || @{$r}[-1]->value ne $value # values differ, can't extend.
3103 || @{$r}[-1]->type != $type # types differ, can't extend.
3104 ) {
3105 push @$r, Range->new($start, $end,
3106 Value => $value,
3107 Type => $type);
3108 }
3109 else {
3110
3111 # Here, the new range starts just after the current highest in
3112 # the range list, and they have the same type and value.
3113 # Extend the current range to incorporate the new one.
3114 @{$r}[-1]->set_end($end);
3115 }
3116
3117 # This becomes the new maximum.
3118 $max{$addr} = $end;
3119
3120 return;
3121 }
3122 #local $to_trace = 0 if main::DEBUG;
3123
3124 trace "$owner_name_of{$addr} $operation", sprintf("%04X", $start) . '..' . sprintf("%04X", $end) . " ($value) replace=$replace" if main::DEBUG && $to_trace;
3125
3126 # Here, the input range isn't after the whole rest of the range list.
3127 # Most likely 'splice' will be needed. The rest of the routine finds
3128 # the needed splice parameters, and if necessary, does the splice.
3129 # First, find the offset parameter needed by the splice function for
3130 # the input range. Note that the input range may span multiple
3131 # existing ones, but we'll worry about that later. For now, just find
3132 # the beginning. If the input range is to be inserted starting in a
3133 # position not currently in the range list, it must (obviously) come
3134 # just after the range below it, and just before the range above it.
3135 # Slightly less obviously, it will occupy the position currently
3136 # occupied by the range that is to come after it. More formally, we
3137 # are looking for the position, $i, in the array of ranges, such that:
3138 #
3139 # r[$i-1]->start <= r[$i-1]->end < $start < r[$i]->start <= r[$i]->end
3140 #
3141 # (The ordered relationships within existing ranges are also shown in
3142 # the equation above). However, if the start of the input range is
3143 # within an existing range, the splice offset should point to that
3144 # existing range's position in the list; that is $i satisfies a
3145 # somewhat different equation, namely:
3146 #
3147 #r[$i-1]->start <= r[$i-1]->end < r[$i]->start <= $start <= r[$i]->end
3148 #
3149 # More briefly, $start can come before or after r[$i]->start, and at
3150 # this point, we don't know which it will be. However, these
3151 # two equations share these constraints:
3152 #
3153 # r[$i-1]->end < $start <= r[$i]->end
3154 #
3155 # And that is good enough to find $i.
3156
3157 my $i = $self->_search_ranges($start);
3158 if (! defined $i) {
3159 Carp::my_carp_bug("Searching $self for range beginning with $start unexpectedly returned undefined. Operation '$operation' not performed");
3160 return;
3161 }
3162
3163 # The search function returns $i such that:
3164 #
3165 # r[$i-1]->end < $start <= r[$i]->end
3166 #
3167 # That means that $i points to the first range in the range list
3168 # that could possibly be affected by this operation. We still don't
3169 # know if the start of the input range is within r[$i], or if it
3170 # points to empty space between r[$i-1] and r[$i].
3171 trace "[$i] is the beginning splice point. Existing range there is ", $r->[$i] if main::DEBUG && $to_trace;
3172
3173 # Special case the insertion of data that is not to replace any
3174 # existing data.
3175 if ($replace == $NO) { # If $NO, has to be operation '+'
3176 #local $to_trace = 1 if main::DEBUG;
3177 trace "Doesn't replace" if main::DEBUG && $to_trace;
3178
3179 # Here, the new range is to take effect only on those code points
3180 # that aren't already in an existing range. This can be done by
3181 # looking through the existing range list and finding the gaps in
3182 # the ranges that this new range affects, and then calling this
3183 # function recursively on each of those gaps, leaving untouched
3184 # anything already in the list. Gather up a list of the changed
3185 # gaps first so that changes to the internal state as new ranges
3186 # are added won't be a problem.
3187 my @gap_list;
3188
3189 # First, if the starting point of the input range is outside an
3190 # existing one, there is a gap from there to the beginning of the
3191 # existing range -- add a span to fill the part that this new
3192 # range occupies
3193 if ($start < $r->[$i]->start) {
3194 push @gap_list, Range->new($start,
3195 main::min($end,
3196 $r->[$i]->start - 1),
3197 Type => $type);
3198 trace "gap before $r->[$i] [$i], will add", $gap_list[-1] if main::DEBUG && $to_trace;
3199 }
3200
3201 # Then look through the range list for other gaps until we reach
3202 # the highest range affected by the input one.
3203 my $j;
3204 for ($j = $i+1; $j < $range_list_size; $j++) {
3205 trace "j=[$j]", $r->[$j] if main::DEBUG && $to_trace;
3206 last if $end < $r->[$j]->start;
3207
3208 # If there is a gap between when this range starts and the
3209 # previous one ends, add a span to fill it. Note that just
3210 # because there are two ranges doesn't mean there is a
3211 # non-zero gap between them. It could be that they have
3212 # different values or types
3213 if ($r->[$j-1]->end + 1 != $r->[$j]->start) {
3214 push @gap_list,
3215 Range->new($r->[$j-1]->end + 1,
3216 $r->[$j]->start - 1,
3217 Type => $type);
3218 trace "gap between $r->[$j-1] and $r->[$j] [$j], will add: $gap_list[-1]" if main::DEBUG && $to_trace;
3219 }
3220 }
3221
3222 # Here, we have either found an existing range in the range list,
3223 # beyond the area affected by the input one, or we fell off the
3224 # end of the loop because the input range affects the whole rest
3225 # of the range list. In either case, $j is 1 higher than the
3226 # highest affected range. If $j == $i, it means that there are no
3227 # affected ranges, that the entire insertion is in the gap between
3228 # r[$i-1], and r[$i], which we already have taken care of before
3229 # the loop.
3230 # On the other hand, if there are affected ranges, it might be
3231 # that there is a gap that needs filling after the final such
3232 # range to the end of the input range
3233 if ($r->[$j-1]->end < $end) {
3234 push @gap_list, Range->new(main::max($start,
3235 $r->[$j-1]->end + 1),
3236 $end,
3237 Type => $type);
3238 trace "gap after $r->[$j-1], will add $gap_list[-1]" if main::DEBUG && $to_trace;
3239 }
3240
3241 # Call recursively to fill in all the gaps.
3242 foreach my $gap (@gap_list) {
3243 $self->_add_delete($operation,
3244 $gap->start,
3245 $gap->end,
3246 $value,
3247 Type => $type);
3248 }
3249
3250 return;
3251 }
3252
3253 # Here, we have taken care of the case where $replace is $NO, which
3254 # means that whatever action we now take is done unconditionally. It
3255 # still could be that this call will result in a no-op, if duplicates
3256 # aren't allowed, and we are inserting a range that merely duplicates
3257 # data already in the range list; or also if deleting a non-existent
3258 # range.
3259 # $i still points to the first potential affected range. Now find the
3260 # highest range affected, which will determine the length parameter to
3261 # splice. (The input range can span multiple existing ones.) While
3262 # we are looking through the range list, see also if this is an
3263 # insertion that will change the values of at least one of the
3264 # affected ranges. We don't need to do this check unless this is an
3265 # insertion of non-multiples, and also since this is a boolean, we
3266 # don't need to do it if have already determined that it will make a
3267 # change; just unconditionally change them. $cdm is created to be 1
3268 # if either of these is true. (The 'c' in the name comes from below)
3269 my $cdm = ($operation eq '-' || $replace == $MULTIPLE);
3270 my $j; # This will point to the highest affected range
3271
3272 # For non-zero types, the standard form is the value itself;
3273 my $standard_form = ($type) ? $value : main::standardize($value);
3274
3275 for ($j = $i; $j < $range_list_size; $j++) {
3276 trace "Looking for highest affected range; the one at $j is ", $r->[$j] if main::DEBUG && $to_trace;
3277
3278 # If find a range that it doesn't overlap into, we can stop
3279 # searching
3280 last if $end < $r->[$j]->start;
3281
3282 # Here, overlaps the range at $j. If the value's don't match,
3283 # and this is supposedly an insertion, it becomes a change
3284 # instead. This is what the 'c' stands for in $cdm.
3285 if (! $cdm) {
3286 if ($r->[$j]->standard_form ne $standard_form) {
3287 $cdm = 1;
3288 }
3289 else {
3290
3291 # Here, the two values are essentially the same. If the
3292 # two are actually identical, replacing wouldn't change
3293 # anything so skip it.
3294 my $pre_existing = $r->[$j]->value;
3295 if ($pre_existing ne $value) {
3296
3297 # Here the new and old standardized values are the
3298 # same, but the non-standardized values aren't. If
3299 # replacing unconditionally, then replace
3300 if( $replace == $UNCONDITIONALLY) {
3301 $cdm = 1;
3302 }
3303 else {
3304
3305 # Here, are replacing conditionally. Decide to
3306 # replace or not based on which appears to look
3307 # the "nicest". If one is mixed case and the
3308 # other isn't, choose the mixed case one.
3309 my $new_mixed = $value =~ /[A-Z]/
3310 && $value =~ /[a-z]/;
3311 my $old_mixed = $pre_existing =~ /[A-Z]/
3312 && $pre_existing =~ /[a-z]/;
3313
3314 if ($old_mixed != $new_mixed) {
3315 $cdm = 1 if $new_mixed;
3316 if (main::DEBUG && $to_trace) {
3317 if ($cdm) {
3318 trace "Replacing $pre_existing with $value";
3319 }
3320 else {
3321 trace "Retaining $pre_existing over $value";
3322 }
3323 }
3324 }
3325 else {
3326
3327 # Here casing wasn't different between the two.
3328 # If one has hyphens or underscores and the
3329 # other doesn't, choose the one with the
3330 # punctuation.
3331 my $new_punct = $value =~ /[-_]/;
3332 my $old_punct = $pre_existing =~ /[-_]/;
3333
3334 if ($old_punct != $new_punct) {
3335 $cdm = 1 if $new_punct;
3336 if (main::DEBUG && $to_trace) {
3337 if ($cdm) {
3338 trace "Replacing $pre_existing with $value";
3339 }
3340 else {
3341 trace "Retaining $pre_existing over $value";
3342 }
3343 }
3344 } # else existing one is just as "good";
3345 # retain it to save cycles.
3346 }
3347 }
3348 }
3349 }
3350 }
3351 } # End of loop looking for highest affected range.
3352
3353 # Here, $j points to one beyond the highest range that this insertion
3354 # affects (hence to beyond the range list if that range is the final
3355 # one in the range list).
3356
3357 # The splice length is all the affected ranges. Get it before
3358 # subtracting, for efficiency, so we don't have to later add 1.
3359 my $length = $j - $i;
3360
3361 $j--; # $j now points to the highest affected range.
3362 trace "Final affected range is $j: $r->[$j]" if main::DEBUG && $to_trace;
3363
3364 # If inserting a multiple record, this is where it goes, after all the
3365 # existing ones for this range. This implies an insertion, and no
3366 # change to any existing ranges. Note that $j can be -1 if this new
3367 # range doesn't actually duplicate any existing, and comes at the
3368 # beginning of the list, in which case we can handle it like any other
3369 # insertion, and is easier to do so.
3370 if ($replace == $MULTIPLE && $j >= 0) {
3371
3372 # This restriction could be remedied with a little extra work, but
3373 # it won't hopefully ever be necessary
3374 if ($r->[$j]->start != $r->[$j]->end) {
3375 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.");
3376 return;
3377 }
3378
3379 # Don't add an exact duplicate, as it isn't really a multiple
3380 return if $value eq $r->[$j]->value && $type eq $r->[$j]->type;
3381
3382 trace "Adding multiple record at $j+1 with $start..$end, $value" if main::DEBUG && $to_trace;
3383 my @return = splice @$r,
3384 $j+1,
3385 0,
3386 Range->new($start,
3387 $end,
3388 Value => $value,
3389 Type => $type);
3390 if (main::DEBUG && $to_trace) {
3391 trace "After splice:";
3392 trace 'j-2=[', $j-2, ']', $r->[$j-2] if $j >= 2;
3393 trace 'j-1=[', $j-1, ']', $r->[$j-1] if $j >= 1;
3394 trace "j =[", $j, "]", $r->[$j] if $j >= 0;
3395 trace 'j+1=[', $j+1, ']', $r->[$j+1] if $j < @$r - 1;
3396 trace 'j+2=[', $j+2, ']', $r->[$j+2] if $j < @$r - 2;
3397 trace 'j+3=[', $j+3, ']', $r->[$j+3] if $j < @$r - 3;
3398 }
3399 return @return;
3400 }
3401
3402 # Here, have taken care of $NO and $MULTIPLE replaces.
3403 # $j points to the highest affected range. But it can be < $i or even
3404 # -1. These happen only if the insertion is entirely in the gap
3405 # between r[$i-1] and r[$i]. Here's why: j < i means that the j loop
3406 # above exited first time through with $end < $r->[$i]->start. (And
3407 # then we subtracted one from j) This implies also that $start <
3408 # $r->[$i]->start, but we know from above that $r->[$i-1]->end <
3409 # $start, so the entire input range is in the gap.
3410 if ($j < $i) {
3411
3412 # Here the entire input range is in the gap before $i.
3413
3414 if (main::DEBUG && $to_trace) {
3415 if ($i) {
3416 trace "Entire range is between $r->[$i-1] and $r->[$i]";
3417 }
3418 else {
3419 trace "Entire range is before $r->[$i]";
3420 }
3421 }
3422 return if $operation ne '+'; # Deletion of a non-existent range is
3423 # a no-op
3424 }
3425 else {
3426
3427 # Here the entire input range is not in the gap before $i. There
3428 # is an affected one, and $j points to the highest such one.
3429
3430 # At this point, here is the situation:
3431 # This is not an insertion of a multiple, nor of tentative ($NO)
3432 # data.
3433 # $i points to the first element in the current range list that
3434 # may be affected by this operation. In fact, we know
3435 # that the range at $i is affected because we are in
3436 # the else branch of this 'if'
3437 # $j points to the highest affected range.
3438 # In other words,
3439 # r[$i-1]->end < $start <= r[$i]->end
3440 # And:
3441 # r[$i-1]->end < $start <= $end <= r[$j]->end
3442 #
3443 # Also:
3444 # $cdm is a boolean which is set true if and only if this is a
3445 # change or deletion (multiple was handled above). In
3446 # other words, it could be renamed to be just $cd.
3447
3448 # We now have enough information to decide if this call is a no-op
3449 # or not. It is a no-op if it is a deletion of a non-existent
3450 # range, or an insertion of already existing data.
3451
3452 if (main::DEBUG && $to_trace && ! $cdm
3453 && $i == $j
3454 && $start >= $r->[$i]->start)
3455 {
3456 trace "no-op";
3457 }
3458 return if ! $cdm # change or delete => not no-op
3459 && $i == $j # more than one affected range => not no-op
3460
3461 # Here, r[$i-1]->end < $start <= $end <= r[$i]->end
3462 # Further, $start and/or $end is >= r[$i]->start
3463 # The test below hence guarantees that
3464 # r[$i]->start < $start <= $end <= r[$i]->end
3465 # This means the input range is contained entirely in
3466 # the one at $i, so is a no-op
3467 && $start >= $r->[$i]->start;
3468 }
3469
3470 # Here, we know that some action will have to be taken. We have
3471 # calculated the offset and length (though adjustments may be needed)
3472 # for the splice. Now start constructing the replacement list.
3473 my @replacement;
3474 my $splice_start = $i;
3475
3476 my $extends_below;
3477 my $extends_above;
3478
3479 # See if should extend any adjacent ranges.
3480 if ($operation eq '-') { # Don't extend deletions
3481 $extends_below = $extends_above = 0;
3482 }
3483 else { # Here, should extend any adjacent ranges. See if there are
3484 # any.
3485 $extends_below = ($i > 0
3486 # can't extend unless adjacent
3487 && $r->[$i-1]->end == $start -1
3488 # can't extend unless are same standard value
3489 && $r->[$i-1]->standard_form eq $standard_form
3490 # can't extend unless share type
3491 && $r->[$i-1]->type == $type);
3492 $extends_above = ($j+1 < $range_list_size
3493 && $r->[$j+1]->start == $end +1
3494 && $r->[$j+1]->standard_form eq $standard_form
3495 && $r->[$j-1]->type == $type);
3496 }
3497 if ($extends_below && $extends_above) { # Adds to both
3498 $splice_start--; # start replace at element below
3499 $length += 2; # will replace on both sides
3500 trace "Extends both below and above ranges" if main::DEBUG && $to_trace;
3501
3502 # The result will fill in any gap, replacing both sides, and
3503 # create one large range.
3504 @replacement = Range->new($r->[$i-1]->start,
3505 $r->[$j+1]->end,
3506 Value => $value,
3507 Type => $type);
3508 }
3509 else {
3510
3511 # Here we know that the result won't just be the conglomeration of
3512 # a new range with both its adjacent neighbors. But it could
3513 # extend one of them.
3514
3515 if ($extends_below) {
3516
3517 # Here the new element adds to the one below, but not to the
3518 # one above. If inserting, and only to that one range, can
3519 # just change its ending to include the new one.
3520 if ($length == 0 && ! $cdm) {
3521 $r->[$i-1]->set_end($end);
3522 trace "inserted range extends range to below so it is now $r->[$i-1]" if main::DEBUG && $to_trace;
3523 return;
3524 }
3525 else {
3526 trace "Changing inserted range to start at ", sprintf("%04X", $r->[$i-1]->start), " instead of ", sprintf("%04X", $start) if main::DEBUG && $to_trace;
3527 $splice_start--; # start replace at element below
3528 $length++; # will replace the element below
3529 $start = $r->[$i-1]->start;
3530 }
3531 }
3532 elsif ($extends_above) {
3533
3534 # Here the new element adds to the one above, but not below.
3535 # Mirror the code above
3536 if ($length == 0 && ! $cdm) {
3537 $r->[$j+1]->set_start($start);
3538 trace "inserted range extends range to above so it is now $r->[$j+1]" if main::DEBUG && $to_trace;
3539 return;
3540 }
3541 else {
3542 trace "Changing inserted range to end at ", sprintf("%04X", $r->[$j+1]->end), " instead of ", sprintf("%04X", $end) if main::DEBUG && $to_trace;
3543 $length++; # will replace the element above
3544 $end = $r->[$j+1]->end;
3545 }
3546 }
3547
3548 trace "Range at $i is $r->[$i]" if main::DEBUG && $to_trace;
3549
3550 # Finally, here we know there will have to be a splice.
3551 # If the change or delete affects only the highest portion of the
3552 # first affected range, the range will have to be split. The
3553 # splice will remove the whole range, but will replace it by a new
3554 # range containing just the unaffected part. So, in this case,
3555 # add to the replacement list just this unaffected portion.
3556 if (! $extends_below
3557 && $start > $r->[$i]->start && $start <= $r->[$i]->end)
3558 {
3559 push @replacement,
3560 Range->new($r->[$i]->start,
3561 $start - 1,
3562 Value => $r->[$i]->value,
3563 Type => $r->[$i]->type);
3564 }
3565
3566 # In the case of an insert or change, but not a delete, we have to
3567 # put in the new stuff; this comes next.
3568 if ($operation eq '+') {
3569 push @replacement, Range->new($start,
3570 $end,
3571 Value => $value,
3572 Type => $type);
3573 }
3574
3575 trace "Range at $j is $r->[$j]" if main::DEBUG && $to_trace && $j != $i;
3576 #trace "$end >=", $r->[$j]->start, " && $end <", $r->[$j]->end if main::DEBUG && $to_trace;
3577
3578 # And finally, if we're changing or deleting only a portion of the
3579 # highest affected range, it must be split, as the lowest one was.
3580 if (! $extends_above
3581 && $j >= 0 # Remember that j can be -1 if before first
3582 # current element
3583 && $end >= $r->[$j]->start
3584 && $end < $r->[$j]->end)
3585 {
3586 push @replacement,
3587 Range->new($end + 1,
3588 $r->[$j]->end,
3589 Value => $r->[$j]->value,
3590 Type => $r->[$j]->type);
3591 }
3592 }
3593
3594 # And do the splice, as calculated above
3595 if (main::DEBUG && $to_trace) {
3596 trace "replacing $length element(s) at $i with ";
3597 foreach my $replacement (@replacement) {
3598 trace " $replacement";
3599 }
3600 trace "Before splice:";
3601 trace 'i-2=[', $i-2, ']', $r->[$i-2] if $i >= 2;
3602 trace 'i-1=[', $i-1, ']', $r->[$i-1] if $i >= 1;
3603 trace "i =[", $i, "]", $r->[$i];
3604 trace 'i+1=[', $i+1, ']', $r->[$i+1] if $i < @$r - 1;
3605 trace 'i+2=[', $i+2, ']', $r->[$i+2] if $i < @$r - 2;
3606 }
3607
3608 my @return = splice @$r, $splice_start, $length, @replacement;
3609
3610 if (main::DEBUG && $to_trace) {
3611 trace "After splice:";
3612 trace 'i-2=[', $i-2, ']', $r->[$i-2] if $i >= 2;
3613 trace 'i-1=[', $i-1, ']', $r->[$i-1] if $i >= 1;
3614 trace "i =[", $i, "]", $r->[$i];
3615 trace 'i+1=[', $i+1, ']', $r->[$i+1] if $i < @$r - 1;
3616 trace 'i+2=[', $i+2, ']', $r->[$i+2] if $i < @$r - 2;
3617 trace "removed @return";
3618 }
3619
3620 # An actual deletion could have changed the maximum in the list.
3621 # There was no deletion if the splice didn't return something, but
3622 # otherwise recalculate it. This is done too rarely to worry about
3623 # performance.
3624 if ($operation eq '-' && @return) {
3625 $max{$addr} = $r->[-1]->end;
3626 }
3627 return @return;
3628 }
3629
3630 sub reset_each_range { # reset the iterator for each_range();
3631 my $self = shift;
3632 Carp::carp_extra_args(\@_) if main::DEBUG && @_;
3633
f998e60c 3634 no overloading;
051df77b 3635 undef $each_range_iterator{pack 'J', $self};
99870f4d
KW
3636 return;
3637 }
3638
3639 sub each_range {
3640 # Iterate over each range in a range list. Results are undefined if
3641 # the range list is changed during the iteration.
3642
3643 my $self = shift;
3644 Carp::carp_extra_args(\@_) if main::DEBUG && @_;
3645
ffe43484 3646 my $addr = do { no overloading; pack 'J', $self; };
99870f4d
KW
3647
3648 return if $self->is_empty;
3649
3650 $each_range_iterator{$addr} = -1
3651 if ! defined $each_range_iterator{$addr};
3652 $each_range_iterator{$addr}++;
3653 return $ranges{$addr}->[$each_range_iterator{$addr}]
3654 if $each_range_iterator{$addr} < @{$ranges{$addr}};
3655 undef $each_range_iterator{$addr};
3656 return;
3657 }
3658
3659 sub count { # Returns count of code points in range list
3660 my $self = shift;
3661 Carp::carp_extra_args(\@_) if main::DEBUG && @_;
3662
ffe43484 3663 my $addr = do { no overloading; pack 'J', $self; };
99870f4d
KW
3664
3665 my $count = 0;
3666 foreach my $range (@{$ranges{$addr}}) {
3667 $count += $range->end - $range->start + 1;
3668 }
3669 return $count;
3670 }
3671
3672 sub delete_range { # Delete a range
3673 my $self = shift;
3674 my $start = shift;
3675 my $end = shift;
3676
3677 Carp::carp_extra_args(\@_) if main::DEBUG && @_;
3678
3679 return $self->_add_delete('-', $start, $end, "");
3680 }
3681
3682 sub is_empty { # Returns boolean as to if a range list is empty
3683 my $self = shift;
3684 Carp::carp_extra_args(\@_) if main::DEBUG && @_;
3685
f998e60c 3686 no overloading;
051df77b 3687 return scalar @{$ranges{pack 'J', $self}} == 0;
99870f4d
KW
3688 }
3689
3690 sub hash {
3691 # Quickly returns a scalar suitable for separating tables into
3692 # buckets, i.e. it is a hash function of the contents of a table, so
3693 # there are relatively few conflicts.
3694
3695 my $self = shift;
3696 Carp::carp_extra_args(\@_) if main::DEBUG && @_;
3697
ffe43484 3698 my $addr = do { no overloading; pack 'J', $self; };
99870f4d
KW
3699
3700 # These are quickly computable. Return looks like 'min..max;count'
3701 return $self->min . "..$max{$addr};" . scalar @{$ranges{$addr}};
3702 }
3703} # End closure for _Range_List_Base
3704
3705package Range_List;
3706use base '_Range_List_Base';
3707
3708# A Range_List is a range list for match tables; i.e. the range values are
3709# not significant. Thus a number of operations can be safely added to it,
3710# such as inversion, intersection. Note that union is also an unsafe
3711# operation when range values are cared about, and that method is in the base
3712# class, not here. But things are set up so that that method is callable only
3713# during initialization. Only in this derived class, is there an operation
3714# that combines two tables. A Range_Map can thus be used to initialize a
3715# Range_List, and its mappings will be in the list, but are not significant to
3716# this class.
3717
3718sub trace { return main::trace(@_); }
3719
3720{ # Closure
3721
3722 use overload
3723 fallback => 0,
3724 '+' => sub { my $self = shift;
3725 my $other = shift;
3726
3727 return $self->_union($other)
3728 },
3729 '&' => sub { my $self = shift;
3730 my $other = shift;
3731
3732 return $self->_intersect($other, 0);
3733 },
3734 '~' => "_invert",
3735 '-' => "_subtract",
3736 ;
3737
3738 sub _invert {
3739 # Returns a new Range_List that gives all code points not in $self.
3740
3741 my $self = shift;
3742
3743 my $new = Range_List->new;
3744
3745 # Go through each range in the table, finding the gaps between them
3746 my $max = -1; # Set so no gap before range beginning at 0
3747 for my $range ($self->ranges) {
3748 my $start = $range->start;
3749 my $end = $range->end;
3750
3751 # If there is a gap before this range, the inverse will contain
3752 # that gap.
3753 if ($start > $max + 1) {
3754 $new->add_range($max + 1, $start - 1);
3755 }
3756 $max = $end;
3757 }
3758
3759 # And finally, add the gap from the end of the table to the max
3760 # possible code point
3761 if ($max < $LAST_UNICODE_CODEPOINT) {
3762 $new->add_range($max + 1, $LAST_UNICODE_CODEPOINT);
3763 }
3764 return $new;
3765 }
3766
3767 sub _subtract {
3768 # Returns a new Range_List with the argument deleted from it. The
3769 # argument can be a single code point, a range, or something that has
3770 # a range, with the _range_list() method on it returning them
3771
3772 my $self = shift;
3773 my $other = shift;
3774 my $reversed = shift;
3775 Carp::carp_extra_args(\@_) if main::DEBUG && @_;
3776
3777 if ($reversed) {
3778 Carp::my_carp_bug("Can't cope with a "
3779 . __PACKAGE__
3780 . " being the second parameter in a '-'. Subtraction ignored.");
3781 return $self;
3782 }
3783
3784 my $new = Range_List->new(Initialize => $self);
3785
3786 if (! ref $other) { # Single code point
3787 $new->delete_range($other, $other);
3788 }
3789 elsif ($other->isa('Range')) {
3790 $new->delete_range($other->start, $other->end);
3791 }
3792 elsif ($other->can('_range_list')) {
3793 foreach my $range ($other->_range_list->ranges) {
3794 $new->delete_range($range->start, $range->end);
3795 }
3796 }
3797 else {
3798 Carp::my_carp_bug("Can't cope with a "
3799 . ref($other)
3800 . " argument to '-'. Subtraction ignored."
3801 );
3802 return $self;
3803 }
3804
3805 return $new;
3806 }
3807
3808 sub _intersect {
3809 # Returns either a boolean giving whether the two inputs' range lists
3810 # intersect (overlap), or a new Range_List containing the intersection
3811 # of the two lists. The optional final parameter being true indicates
3812 # to do the check instead of the intersection.
3813
3814 my $a_object = shift;
3815 my $b_object = shift;
3816 my $check_if_overlapping = shift;
3817 $check_if_overlapping = 0 unless defined $check_if_overlapping;
3818 Carp::carp_extra_args(\@_) if main::DEBUG && @_;
3819
3820 if (! defined $b_object) {
3821 my $message = "";
3822 $message .= $a_object->_owner_name_of if defined $a_object;
3823 Carp::my_carp_bug($message .= "Called with undefined value. Intersection not done.");
3824 return;
3825 }
3826
3827 # a & b = !(!a | !b), or in our terminology = ~ ( ~a + -b )
3828 # Thus the intersection could be much more simply be written:
3829 # return ~(~$a_object + ~$b_object);
3830 # But, this is slower, and when taking the inverse of a large
3831 # range_size_1 table, back when such tables were always stored that
3832 # way, it became prohibitively slow, hence the code was changed to the
3833 # below
3834
3835 if ($b_object->isa('Range')) {
3836 $b_object = Range_List->new(Initialize => $b_object,
3837 Owner => $a_object->_owner_name_of);
3838 }
3839 $b_object = $b_object->_range_list if $b_object->can('_range_list');
3840
3841 my @a_ranges = $a_object->ranges;
3842 my @b_ranges = $b_object->ranges;
3843
3844 #local $to_trace = 1 if main::DEBUG;
3845 trace "intersecting $a_object with ", scalar @a_ranges, "ranges and $b_object with", scalar @b_ranges, " ranges" if main::DEBUG && $to_trace;
3846
3847 # Start with the first range in each list
3848 my $a_i = 0;
3849 my $range_a = $a_ranges[$a_i];
3850 my $b_i = 0;
3851 my $range_b = $b_ranges[$b_i];
3852
3853 my $new = __PACKAGE__->new(Owner => $a_object->_owner_name_of)
3854 if ! $check_if_overlapping;
3855
3856 # If either list is empty, there is no intersection and no overlap
3857 if (! defined $range_a || ! defined $range_b) {
3858 return $check_if_overlapping ? 0 : $new;
3859 }
3860 trace "range_a[$a_i]=$range_a; range_b[$b_i]=$range_b" if main::DEBUG && $to_trace;
3861
3862 # Otherwise, must calculate the intersection/overlap. Start with the
3863 # very first code point in each list
3864 my $a = $range_a->start;
3865 my $b = $range_b->start;
3866
3867 # Loop through all the ranges of each list; in each iteration, $a and
3868 # $b are the current code points in their respective lists
3869 while (1) {
3870
3871 # If $a and $b are the same code point, ...
3872 if ($a == $b) {
3873
3874 # it means the lists overlap. If just checking for overlap
3875 # know the answer now,
3876 return 1 if $check_if_overlapping;
3877
3878 # The intersection includes this code point plus anything else
3879 # common to both current ranges.
3880 my $start = $a;
3881 my $end = main::min($range_a->end, $range_b->end);
3882 if (! $check_if_overlapping) {
3883 trace "adding intersection range ", sprintf("%04X", $start) . ".." . sprintf("%04X", $end) if main::DEBUG && $to_trace;
3884 $new->add_range($start, $end);
3885 }
3886
3887 # Skip ahead to the end of the current intersect
3888 $a = $b = $end;
3889
3890 # If the current intersect ends at the end of either range (as
3891 # it must for at least one of them), the next possible one
3892 # will be the beginning code point in it's list's next range.
3893 if ($a == $range_a->end) {
3894 $range_a = $a_ranges[++$a_i];
3895 last unless defined $range_a;
3896 $a = $range_a->start;
3897 }
3898 if ($b == $range_b->end) {
3899 $range_b = $b_ranges[++$b_i];
3900 last unless defined $range_b;
3901 $b = $range_b->start;
3902 }
3903
3904 trace "range_a[$a_i]=$range_a; range_b[$b_i]=$range_b" if main::DEBUG && $to_trace;
3905 }
3906 elsif ($a < $b) {
3907
3908 # Not equal, but if the range containing $a encompasses $b,
3909 # change $a to be the middle of the range where it does equal
3910 # $b, so the next iteration will get the intersection
3911 if ($range_a->end >= $b) {
3912 $a = $b;
3913 }
3914 else {
3915
3916 # Here, the current range containing $a is entirely below
3917 # $b. Go try to find a range that could contain $b.
3918 $a_i = $a_object->_search_ranges($b);
3919
3920 # If no range found, quit.
3921 last unless defined $a_i;
3922
3923 # The search returns $a_i, such that
3924 # range_a[$a_i-1]->end < $b <= range_a[$a_i]->end
3925 # Set $a to the beginning of this new range, and repeat.
3926 $range_a = $a_ranges[$a_i];
3927 $a = $range_a->start;
3928 }
3929 }
3930 else { # Here, $b < $a.
3931
3932 # Mirror image code to the leg just above
3933 if ($range_b->end >= $a) {
3934 $b = $a;
3935 }
3936 else {
3937 $b_i = $b_object->_search_ranges($a);
3938 last unless defined $b_i;
3939 $range_b = $b_ranges[$b_i];
3940 $b = $range_b->start;
3941 }
3942 }
3943 } # End of looping through ranges.
3944
3945 # Intersection fully computed, or now know that there is no overlap
3946 return $check_if_overlapping ? 0 : $new;
3947 }
3948
3949 sub overlaps {
3950 # Returns boolean giving whether the two arguments overlap somewhere
3951
3952 my $self = shift;
3953 my $other = shift;
3954 Carp::carp_extra_args(\@_) if main::DEBUG && @_;
3955
3956 return $self->_intersect($other, 1);
3957 }
3958
3959 sub add_range {
3960 # Add a range to the list.
3961
3962 my $self = shift;
3963 my $start = shift;
3964 my $end = shift;
3965 Carp::carp_extra_args(\@_) if main::DEBUG && @_;
3966
3967 return $self->_add_delete('+', $start, $end, "");
3968 }
3969
99870f4d
KW
3970 sub is_code_point_usable {
3971 # This used only for making the test script. See if the input
3972 # proposed trial code point is one that Perl will handle. If second
3973 # parameter is 0, it won't select some code points for various
3974 # reasons, noted below.
3975
3976 my $code = shift;
3977 my $try_hard = shift;
3978 Carp::carp_extra_args(\@_) if main::DEBUG && @_;
3979
3980 return 0 if $code < 0; # Never use a negative
3981
99870f4d
KW
3982 # shun null. I'm (khw) not sure why this was done, but NULL would be
3983 # the character very frequently used.
3984 return $try_hard if $code == 0x0000;
3985
3986 return 0 if $try_hard; # XXX Temporary until fix utf8.c
3987
3988 # shun non-character code points.
3989 return $try_hard if $code >= 0xFDD0 && $code <= 0xFDEF;
3990 return $try_hard if ($code & 0xFFFE) == 0xFFFE; # includes FFFF
3991
3992 return $try_hard if $code > $LAST_UNICODE_CODEPOINT; # keep in range
3993 return $try_hard if $code >= 0xD800 && $code <= 0xDFFF; # no surrogate
3994
3995 return 1;
3996 }
3997
3998 sub get_valid_code_point {
3999 # Return a code point that's part of the range list. Returns nothing
4000 # if the table is empty or we can't find a suitable code point. This
4001 # used only for making the test script.
4002
4003 my $self = shift;
4004 Carp::carp_extra_args(\@_) if main::DEBUG && @_;
4005
ffe43484 4006 my $addr = do { no overloading; pack 'J', $self; };
99870f4d
KW
4007
4008 # On first pass, don't choose less desirable code points; if no good
4009 # one is found, repeat, allowing a less desirable one to be selected.
4010 for my $try_hard (0, 1) {
4011
4012 # Look through all the ranges for a usable code point.
4013 for my $set ($self->ranges) {
4014
4015 # Try the edge cases first, starting with the end point of the
4016 # range.
4017 my $end = $set->end;
4018 return $end if is_code_point_usable($end, $try_hard);
4019
4020 # End point didn't, work. Start at the beginning and try
4021 # every one until find one that does work.
4022 for my $trial ($set->start .. $end - 1) {
4023 return $trial if is_code_point_usable($trial, $try_hard);
4024 }
4025 }
4026 }
4027 return (); # If none found, give up.
4028 }