123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922923924925926927928929930931932933934935936937938939940941942943944945946947948949950951952953954955956957958959960961962963964965966967968969970971972973974975976977978979980981982983984985986987988989990991992993994995996997998999100010011002100310041005100610071008100910101011101210131014101510161017101810191020102110221023102410251026102710281029103010311032103310341035103610371038103910401041104210431044104510461047104810491050105110521053105410551056105710581059106010611062106310641065106610671068106910701071107210731074107510761077107810791080108110821083108410851086108710881089109010911092109310941095109610971098109911001101110211031104110511061107110811091110111111121113111411151116111711181119112011211122112311241125112611271128112911301131113211331134113511361137113811391140114111421143114411451146114711481149115011511152115311541155115611571158115911601161116211631164116511661167116811691170117111721173117411751176117711781179118011811182118311841185118611871188118911901191119211931194119511961197119811991200120112021203120412051206120712081209121012111212121312141215121612171218121912201221122212231224122512261227122812291230123112321233123412351236123712381239124012411242124312441245124612471248124912501251125212531254125512561257125812591260126112621263126412651266126712681269127012711272127312741275127612771278127912801281128212831284128512861287128812891290129112921293129412951296129712981299130013011302130313041305130613071308130913101311131213131314131513161317131813191320132113221323132413251326132713281329133013311332133313341335133613371338133913401341134213431344134513461347134813491350135113521353135413551356135713581359136013611362136313641365136613671368136913701371137213731374137513761377137813791380138113821383138413851386138713881389139013911392139313941395139613971398139914001401140214031404140514061407140814091410141114121413141414151416141714181419142014211422142314241425142614271428142914301431143214331434143514361437143814391440144114421443144414451446144714481449145014511452 |
- /*
- * Copyright (c) 2021, Tim Flynn <trflynn89@serenityos.org>
- *
- * SPDX-License-Identifier: BSD-2-Clause
- */
- #include "GeneratorUtil.h"
- #include <AK/AllOf.h>
- #include <AK/Array.h>
- #include <AK/CharacterTypes.h>
- #include <AK/Find.h>
- #include <AK/HashMap.h>
- #include <AK/Optional.h>
- #include <AK/QuickSort.h>
- #include <AK/SourceGenerator.h>
- #include <AK/String.h>
- #include <AK/StringUtils.h>
- #include <AK/Types.h>
- #include <AK/Vector.h>
- #include <LibCore/ArgsParser.h>
- #include <LibCore/Stream.h>
- using StringIndexType = u16;
- constexpr auto s_string_index_type = "u16"sv;
- // Some code points are excluded from UnicodeData.txt, and instead are part of a "range" of code
- // points, as indicated by the "name" field. For example:
- // 3400;<CJK Ideograph Extension A, First>;Lo;0;L;;;;;N;;;;;
- // 4DBF;<CJK Ideograph Extension A, Last>;Lo;0;L;;;;;N;;;;;
- struct CodePointRange {
- u32 first;
- u32 last;
- };
- // SpecialCasing source: https://www.unicode.org/Public/13.0.0/ucd/SpecialCasing.txt
- // Field descriptions: https://www.unicode.org/reports/tr44/tr44-13.html#SpecialCasing.txt
- struct SpecialCasing {
- u32 index { 0 };
- u32 code_point { 0 };
- Vector<u32> lowercase_mapping;
- Vector<u32> uppercase_mapping;
- Vector<u32> titlecase_mapping;
- String locale;
- String condition;
- };
- // Field descriptions: https://www.unicode.org/reports/tr44/#Character_Decomposition_Mappings
- struct CodePointDecomposition {
- // `tag` is a string since it's used for codegen as an enum value.
- String tag { "Canonical"sv };
- size_t decomposition_index { 0 };
- size_t decomposition_size { 0 };
- };
- // PropList source: https://www.unicode.org/Public/13.0.0/ucd/PropList.txt
- // Property descriptions: https://www.unicode.org/reports/tr44/tr44-13.html#PropList.txt
- using PropList = HashMap<String, Vector<CodePointRange>>;
- // Normalization source: https://www.unicode.org/Public/13.0.0/ucd/DerivedNormalizationProps.txt
- // Normalization descriptions: https://www.unicode.org/reports/tr44/#DerivedNormalizationProps.txt
- enum class QuickCheck {
- Yes,
- No,
- Maybe,
- };
- struct Normalization {
- CodePointRange code_point_range;
- Vector<u32> value;
- QuickCheck quick_check { QuickCheck::Yes };
- };
- using NormalizationProps = HashMap<String, Vector<Normalization>>;
- struct CodePointName {
- CodePointRange code_point_range;
- StringIndexType name { 0 };
- };
- // UnicodeData source: https://www.unicode.org/Public/13.0.0/ucd/UnicodeData.txt
- // Field descriptions: https://www.unicode.org/reports/tr44/tr44-13.html#UnicodeData.txt
- // https://www.unicode.org/reports/tr44/#General_Category_Values
- struct CodePointData {
- u32 code_point { 0 };
- String name;
- Optional<StringIndexType> abbreviation;
- u8 canonical_combining_class { 0 };
- String bidi_class;
- Optional<CodePointDecomposition> decomposition_mapping;
- Optional<i8> numeric_value_decimal;
- Optional<i8> numeric_value_digit;
- Optional<i8> numeric_value_numeric;
- bool bidi_mirrored { false };
- String unicode_1_name;
- String iso_comment;
- Optional<u32> simple_uppercase_mapping;
- Optional<u32> simple_lowercase_mapping;
- Optional<u32> simple_titlecase_mapping;
- Vector<u32> special_casing_indices;
- };
- struct BlockName {
- CodePointRange code_point_range;
- StringIndexType name { 0 };
- };
- struct UnicodeData {
- UniqueStringStorage<StringIndexType> unique_strings;
- u32 code_points_with_non_zero_combining_class { 0 };
- u32 code_points_with_decomposition_mapping { 0 };
- Vector<u32> decomposition_mappings;
- Vector<String> compatibility_tags;
- u32 simple_uppercase_mapping_size { 0 };
- u32 simple_lowercase_mapping_size { 0 };
- Vector<SpecialCasing> special_casing;
- u32 code_points_with_special_casing { 0 };
- u32 largest_casing_transform_size { 0 };
- u32 largest_special_casing_size { 0 };
- Vector<String> conditions;
- Vector<String> locales;
- Vector<CodePointData> code_point_data;
- HashMap<u32, StringIndexType> code_point_abbreviations;
- HashMap<u32, StringIndexType> code_point_display_name_aliases;
- Vector<CodePointName> code_point_display_names;
- PropList general_categories;
- Vector<Alias> general_category_aliases;
- // The Unicode standard defines additional properties (Any, Assigned, ASCII) which are not in
- // any UCD file. Assigned code point ranges are derived as this generator is executed.
- // https://unicode.org/reports/tr18/#General_Category_Property
- PropList prop_list {
- { "Any"sv, { { 0, 0x10ffff } } },
- { "Assigned"sv, {} },
- { "ASCII"sv, { { 0, 0x7f } } },
- };
- Vector<Alias> prop_aliases;
- PropList script_list {
- { "Unknown"sv, {} },
- };
- Vector<Alias> script_aliases;
- PropList script_extensions;
- PropList block_list {
- { "No_Block"sv, {} },
- };
- Vector<Alias> block_aliases;
- Vector<BlockName> block_display_names;
- // FIXME: We are not yet doing anything with this data. It will be needed for String.prototype.normalize.
- NormalizationProps normalization_props;
- PropList grapheme_break_props;
- PropList word_break_props;
- PropList sentence_break_props;
- };
- static String sanitize_entry(String const& entry)
- {
- auto sanitized = entry.replace("-"sv, "_"sv, ReplaceMode::All);
- sanitized = sanitized.replace(" "sv, "_"sv, ReplaceMode::All);
- StringBuilder builder;
- bool next_is_upper = true;
- for (auto ch : sanitized) {
- if (next_is_upper)
- builder.append_code_point(to_ascii_uppercase(ch));
- else
- builder.append_code_point(ch);
- next_is_upper = ch == '_';
- }
- return builder.to_string();
- }
- static Vector<u32> parse_code_point_list(StringView list)
- {
- Vector<u32> code_points;
- auto segments = list.split_view(' ');
- for (auto const& code_point : segments)
- code_points.append(AK::StringUtils::convert_to_uint_from_hex<u32>(code_point).value());
- return code_points;
- }
- static CodePointRange parse_code_point_range(StringView list)
- {
- CodePointRange code_point_range {};
- if (list.contains(".."sv)) {
- auto segments = list.split_view(".."sv);
- VERIFY(segments.size() == 2);
- auto begin = AK::StringUtils::convert_to_uint_from_hex<u32>(segments[0]).value();
- auto end = AK::StringUtils::convert_to_uint_from_hex<u32>(segments[1]).value();
- code_point_range = { begin, end };
- } else {
- auto code_point = AK::StringUtils::convert_to_uint_from_hex<u32>(list).value();
- code_point_range = { code_point, code_point };
- }
- return code_point_range;
- }
- static ErrorOr<void> parse_special_casing(Core::Stream::BufferedFile& file, UnicodeData& unicode_data)
- {
- Array<u8, 1024> buffer;
- while (TRY(file.can_read_line())) {
- auto line = TRY(file.read_line(buffer));
- if (line.is_empty() || line.starts_with('#'))
- continue;
- if (auto index = line.find('#'); index.has_value())
- line = line.substring_view(0, *index);
- auto segments = line.split_view(';', SplitBehavior::KeepEmpty);
- VERIFY(segments.size() == 5 || segments.size() == 6);
- SpecialCasing casing {};
- casing.code_point = AK::StringUtils::convert_to_uint_from_hex<u32>(segments[0]).value();
- casing.lowercase_mapping = parse_code_point_list(segments[1]);
- casing.titlecase_mapping = parse_code_point_list(segments[2]);
- casing.uppercase_mapping = parse_code_point_list(segments[3]);
- if (auto condition = segments[4].trim_whitespace(); !condition.is_empty()) {
- auto conditions = condition.split_view(' ', SplitBehavior::KeepEmpty);
- VERIFY(conditions.size() == 1 || conditions.size() == 2);
- if (conditions.size() == 2) {
- casing.locale = conditions[0];
- casing.condition = conditions[1];
- } else if (all_of(conditions[0], is_ascii_lower_alpha)) {
- casing.locale = conditions[0];
- } else {
- casing.condition = conditions[0];
- }
- if (!casing.locale.is_empty()) {
- casing.locale = String::formatted("{:c}{}", to_ascii_uppercase(casing.locale[0]), casing.locale.substring_view(1));
- if (!unicode_data.locales.contains_slow(casing.locale))
- unicode_data.locales.append(casing.locale);
- }
- casing.condition = casing.condition.replace("_"sv, ""sv, ReplaceMode::All);
- if (!casing.condition.is_empty() && !unicode_data.conditions.contains_slow(casing.condition))
- unicode_data.conditions.append(casing.condition);
- }
- unicode_data.largest_casing_transform_size = max(unicode_data.largest_casing_transform_size, casing.lowercase_mapping.size());
- unicode_data.largest_casing_transform_size = max(unicode_data.largest_casing_transform_size, casing.titlecase_mapping.size());
- unicode_data.largest_casing_transform_size = max(unicode_data.largest_casing_transform_size, casing.uppercase_mapping.size());
- unicode_data.special_casing.append(move(casing));
- }
- quick_sort(unicode_data.special_casing, [](auto const& lhs, auto const& rhs) {
- if (lhs.code_point != rhs.code_point)
- return lhs.code_point < rhs.code_point;
- if (lhs.locale.is_empty() && !rhs.locale.is_empty())
- return false;
- if (!lhs.locale.is_empty() && rhs.locale.is_empty())
- return true;
- return lhs.locale < rhs.locale;
- });
- for (u32 i = 0; i < unicode_data.special_casing.size(); ++i)
- unicode_data.special_casing[i].index = i;
- return {};
- }
- static ErrorOr<void> parse_prop_list(Core::Stream::BufferedFile& file, PropList& prop_list, bool multi_value_property = false, bool sanitize_property = false)
- {
- Array<u8, 1024> buffer;
- while (TRY(file.can_read_line())) {
- auto line = TRY(file.read_line(buffer));
- if (line.is_empty() || line.starts_with('#'))
- continue;
- if (auto index = line.find('#'); index.has_value())
- line = line.substring_view(0, *index);
- auto segments = line.split_view(';', SplitBehavior::KeepEmpty);
- VERIFY(segments.size() == 2);
- auto code_point_range = parse_code_point_range(segments[0].trim_whitespace());
- Vector<StringView> properties;
- if (multi_value_property)
- properties = segments[1].trim_whitespace().split_view(' ');
- else
- properties = { segments[1].trim_whitespace() };
- for (auto& property : properties) {
- auto& code_points = prop_list.ensure(sanitize_property ? sanitize_entry(property).trim_whitespace().view() : property.trim_whitespace());
- code_points.append(code_point_range);
- }
- }
- return {};
- }
- static ErrorOr<void> parse_alias_list(Core::Stream::BufferedFile& file, PropList const& prop_list, Vector<Alias>& prop_aliases)
- {
- String current_property;
- Array<u8, 1024> buffer;
- auto append_alias = [&](auto alias, auto property) {
- // Note: The alias files contain lines such as "Hyphen = Hyphen", which we should just skip.
- if (alias == property)
- return;
- // FIXME: We will, eventually, need to find where missing properties are located and parse them.
- if (!prop_list.contains(property))
- return;
- prop_aliases.append({ property, alias });
- };
- while (TRY(file.can_read_line())) {
- auto line = TRY(file.read_line(buffer));
- if (line.is_empty() || line.starts_with('#')) {
- if (line.ends_with("Properties"sv))
- current_property = line.substring_view(2);
- continue;
- }
- // Note: For now, we only care about Binary Property aliases for Unicode property escapes.
- if (current_property != "Binary Properties"sv)
- continue;
- auto segments = line.split_view(';', SplitBehavior::KeepEmpty);
- VERIFY((segments.size() == 2) || (segments.size() == 3));
- auto alias = segments[0].trim_whitespace();
- auto property = segments[1].trim_whitespace();
- append_alias(alias, property);
- if (segments.size() == 3) {
- alias = segments[2].trim_whitespace();
- append_alias(alias, property);
- }
- }
- return {};
- }
- static ErrorOr<void> parse_name_aliases(Core::Stream::BufferedFile& file, UnicodeData& unicode_data)
- {
- Array<u8, 1024> buffer;
- while (TRY(file.can_read_line())) {
- auto line = TRY(file.read_line(buffer));
- if (line.is_empty() || line.starts_with('#'))
- continue;
- auto segments = line.split_view(';', SplitBehavior::KeepEmpty);
- VERIFY(segments.size() == 3);
- auto code_point = AK::StringUtils::convert_to_uint_from_hex<u32>(segments[0].trim_whitespace());
- auto alias = segments[1].trim_whitespace();
- auto reason = segments[2].trim_whitespace();
- if (reason == "abbreviation"sv) {
- auto index = unicode_data.unique_strings.ensure(alias);
- unicode_data.code_point_abbreviations.set(*code_point, index);
- } else if (reason.is_one_of("correction"sv, "control"sv)) {
- if (!unicode_data.code_point_display_name_aliases.contains(*code_point)) {
- auto index = unicode_data.unique_strings.ensure(alias);
- unicode_data.code_point_display_name_aliases.set(*code_point, index);
- }
- }
- }
- return {};
- }
- static ErrorOr<void> parse_value_alias_list(Core::Stream::BufferedFile& file, StringView desired_category, Vector<String> const& value_list, Vector<Alias>& prop_aliases, bool primary_value_is_first = true, bool sanitize_alias = false)
- {
- TRY(file.seek(0, Core::Stream::SeekMode::SetPosition));
- Array<u8, 1024> buffer;
- auto append_alias = [&](auto alias, auto value) {
- // Note: The value alias file contains lines such as "Ahom = Ahom", which we should just skip.
- if (alias == value)
- return;
- // FIXME: We will, eventually, need to find where missing properties are located and parse them.
- if (!value_list.contains_slow(value))
- return;
- prop_aliases.append({ value, alias });
- };
- while (TRY(file.can_read_line())) {
- auto line = TRY(file.read_line(buffer));
- if (line.is_empty() || line.starts_with('#'))
- continue;
- if (auto index = line.find('#'); index.has_value())
- line = line.substring_view(0, *index);
- auto segments = line.split_view(';', SplitBehavior::KeepEmpty);
- auto category = segments[0].trim_whitespace();
- if (category != desired_category)
- continue;
- VERIFY((segments.size() == 3) || (segments.size() == 4));
- auto value = primary_value_is_first ? segments[1].trim_whitespace() : segments[2].trim_whitespace();
- auto alias = primary_value_is_first ? segments[2].trim_whitespace() : segments[1].trim_whitespace();
- append_alias(sanitize_alias ? sanitize_entry(alias).view() : alias, value);
- if (segments.size() == 4) {
- alias = segments[3].trim_whitespace();
- append_alias(sanitize_alias ? sanitize_entry(alias).view() : alias, value);
- }
- }
- return {};
- }
- static ErrorOr<void> parse_normalization_props(Core::Stream::BufferedFile& file, UnicodeData& unicode_data)
- {
- Array<u8, 1024> buffer;
- while (TRY(file.can_read_line())) {
- auto line = TRY(file.read_line(buffer));
- if (line.is_empty() || line.starts_with('#'))
- continue;
- if (auto index = line.find('#'); index.has_value())
- line = line.substring_view(0, *index);
- auto segments = line.split_view(';', SplitBehavior::KeepEmpty);
- VERIFY((segments.size() == 2) || (segments.size() == 3));
- auto code_point_range = parse_code_point_range(segments[0].trim_whitespace());
- auto property = segments[1].trim_whitespace().to_string();
- Vector<u32> value;
- QuickCheck quick_check = QuickCheck::Yes;
- if (segments.size() == 3) {
- auto value_or_quick_check = segments[2].trim_whitespace();
- if ((value_or_quick_check == "N"sv))
- quick_check = QuickCheck::No;
- else if ((value_or_quick_check == "M"sv))
- quick_check = QuickCheck::Maybe;
- else
- value = parse_code_point_list(value_or_quick_check);
- }
- auto& normalizations = unicode_data.normalization_props.ensure(property);
- normalizations.append({ code_point_range, move(value), quick_check });
- auto& prop_list = unicode_data.prop_list.ensure(property);
- prop_list.append(move(code_point_range));
- }
- return {};
- }
- static void add_canonical_code_point_name(CodePointRange range, StringView name, UnicodeData& unicode_data)
- {
- // https://www.unicode.org/versions/Unicode15.0.0/ch04.pdf#G142981
- // FIXME: Implement the NR1 rules for Hangul syllables.
- struct CodePointNameFormat {
- CodePointRange code_point_range;
- StringView name;
- };
- // These code point ranges are the NR2 set of name replacements defined by Table 4-8.
- constexpr Array<CodePointNameFormat, 16> s_ideographic_replacements { {
- { { 0x3400, 0x4DBF }, "CJK UNIFIED IDEOGRAPH-{:X}"sv },
- { { 0x4E00, 0x9FFF }, "CJK UNIFIED IDEOGRAPH-{:X}"sv },
- { { 0xF900, 0xFA6D }, "CJK COMPATIBILITY IDEOGRAPH-{:X}"sv },
- { { 0xFA70, 0xFAD9 }, "CJK COMPATIBILITY IDEOGRAPH-{:X}"sv },
- { { 0x17000, 0x187F7 }, "TANGUT IDEOGRAPH-{:X}"sv },
- { { 0x18B00, 0x18CD5 }, "KHITAN SMALL SCRIPT CHARACTER-{:X}"sv },
- { { 0x18D00, 0x18D08 }, "TANGUT IDEOGRAPH-{:X}"sv },
- { { 0x1B170, 0x1B2FB }, "NUSHU CHARACTER-{:X}"sv },
- { { 0x20000, 0x2A6DF }, "CJK UNIFIED IDEOGRAPH-{:X}"sv },
- { { 0x2A700, 0x2B739 }, "CJK UNIFIED IDEOGRAPH-{:X}"sv },
- { { 0x2B740, 0x2B81D }, "CJK UNIFIED IDEOGRAPH-{:X}"sv },
- { { 0x2B820, 0x2CEA1 }, "CJK UNIFIED IDEOGRAPH-{:X}"sv },
- { { 0x2CEB0, 0x2EBE0 }, "CJK UNIFIED IDEOGRAPH-{:X}"sv },
- { { 0x2F800, 0x2FA1D }, "CJK COMPATIBILITY IDEOGRAPH-{:X}"sv },
- { { 0x30000, 0x3134A }, "CJK UNIFIED IDEOGRAPH-{:X}"sv },
- { { 0x31350, 0x323AF }, "CJK UNIFIED IDEOGRAPH-{:X}"sv },
- } };
- auto it = find_if(s_ideographic_replacements.begin(), s_ideographic_replacements.end(),
- [&](auto const& replacement) {
- return replacement.code_point_range.first == range.first;
- });
- if (it != s_ideographic_replacements.end()) {
- auto index = unicode_data.unique_strings.ensure(it->name);
- unicode_data.code_point_display_names.append({ it->code_point_range, index });
- return;
- }
- it = find_if(s_ideographic_replacements.begin(), s_ideographic_replacements.end(),
- [&](auto const& replacement) {
- return (replacement.code_point_range.first <= range.first) && (range.first <= replacement.code_point_range.last);
- });
- if (it != s_ideographic_replacements.end()) {
- // Drop code points that will have been captured by a range defined by the ideographic replacements.
- return;
- }
- if (auto alias = unicode_data.code_point_display_name_aliases.get(range.first); alias.has_value()) {
- // NR4 states that control code points have a null string as their name. Our implementation
- // uses the control code's alias as its display name.
- unicode_data.code_point_display_names.append({ range, *alias });
- return;
- }
- auto index = unicode_data.unique_strings.ensure(name);
- unicode_data.code_point_display_names.append({ range, index });
- }
- static Optional<CodePointDecomposition> parse_decomposition_mapping(StringView string, UnicodeData& unicode_data)
- {
- if (string.is_empty())
- return {};
- CodePointDecomposition mapping;
- auto parts = string.split_view(' ');
- VERIFY(parts.size() > 0);
- if (parts.first().starts_with('<')) {
- auto const tag = parts.take_first().trim("<>"sv);
- mapping.tag = String::formatted("{:c}{}", to_ascii_uppercase(tag[0]), tag.substring_view(1));
- if (!unicode_data.compatibility_tags.contains_slow(mapping.tag))
- unicode_data.compatibility_tags.append(mapping.tag);
- }
- mapping.decomposition_index = unicode_data.decomposition_mappings.size();
- mapping.decomposition_size = parts.size();
- for (auto part : parts) {
- unicode_data.decomposition_mappings.append(AK::StringUtils::convert_to_uint_from_hex<u32>(part).value());
- }
- return mapping;
- }
- static ErrorOr<void> parse_block_display_names(Core::Stream::BufferedFile& file, UnicodeData& unicode_data)
- {
- Array<u8, 1024> buffer;
- while (TRY(file.can_read_line())) {
- auto line = TRY(file.read_line(buffer));
- if (line.is_empty() || line.starts_with('#'))
- continue;
- auto segments = line.split_view(';', SplitBehavior::KeepEmpty);
- VERIFY(segments.size() == 2);
- auto code_point_range = parse_code_point_range(segments[0].trim_whitespace());
- auto display_name = segments[1].trim_whitespace();
- auto index = unicode_data.unique_strings.ensure(display_name);
- unicode_data.block_display_names.append({ code_point_range, index });
- }
- TRY(file.seek(0, Core::Stream::SeekMode::SetPosition));
- return {};
- }
- static ErrorOr<void> parse_unicode_data(Core::Stream::BufferedFile& file, UnicodeData& unicode_data)
- {
- Optional<u32> code_point_range_start;
- auto& assigned_code_points = unicode_data.prop_list.find("Assigned"sv)->value;
- Optional<u32> assigned_code_point_range_start = 0;
- u32 previous_code_point = 0;
- Array<u8, 1024> buffer;
- while (TRY(file.can_read_line())) {
- auto line = TRY(file.read_line(buffer));
- if (line.is_empty())
- continue;
- auto segments = line.split_view(';', SplitBehavior::KeepEmpty);
- VERIFY(segments.size() == 15);
- CodePointData data {};
- data.code_point = AK::StringUtils::convert_to_uint_from_hex<u32>(segments[0]).value();
- data.name = segments[1];
- data.canonical_combining_class = AK::StringUtils::convert_to_uint<u8>(segments[3]).value();
- data.bidi_class = segments[4];
- data.decomposition_mapping = parse_decomposition_mapping(segments[5], unicode_data);
- data.numeric_value_decimal = AK::StringUtils::convert_to_int<i8>(segments[6]);
- data.numeric_value_digit = AK::StringUtils::convert_to_int<i8>(segments[7]);
- data.numeric_value_numeric = AK::StringUtils::convert_to_int<i8>(segments[8]);
- data.bidi_mirrored = segments[9] == "Y"sv;
- data.unicode_1_name = segments[10];
- data.iso_comment = segments[11];
- data.simple_uppercase_mapping = AK::StringUtils::convert_to_uint_from_hex<u32>(segments[12]);
- data.simple_lowercase_mapping = AK::StringUtils::convert_to_uint_from_hex<u32>(segments[13]);
- data.simple_titlecase_mapping = AK::StringUtils::convert_to_uint_from_hex<u32>(segments[14]);
- if (auto abbreviation = unicode_data.code_point_abbreviations.get(data.code_point); abbreviation.has_value())
- data.abbreviation = *abbreviation;
- if (!assigned_code_point_range_start.has_value())
- assigned_code_point_range_start = data.code_point;
- if (data.name.starts_with("<"sv) && data.name.ends_with(", First>"sv)) {
- VERIFY(!code_point_range_start.has_value() && assigned_code_point_range_start.has_value());
- code_point_range_start = data.code_point;
- data.name = data.name.substring(1, data.name.length() - 9);
- assigned_code_points.append({ *assigned_code_point_range_start, previous_code_point });
- assigned_code_point_range_start.clear();
- } else if (data.name.starts_with("<"sv) && data.name.ends_with(", Last>"sv)) {
- VERIFY(code_point_range_start.has_value());
- CodePointRange code_point_range { *code_point_range_start, data.code_point };
- assigned_code_points.append(code_point_range);
- data.name = data.name.substring(1, data.name.length() - 8);
- code_point_range_start.clear();
- add_canonical_code_point_name(code_point_range, data.name, unicode_data);
- } else {
- add_canonical_code_point_name({ data.code_point, data.code_point }, data.name, unicode_data);
- if ((data.code_point > 0) && (data.code_point - previous_code_point) != 1) {
- VERIFY(assigned_code_point_range_start.has_value());
- assigned_code_points.append({ *assigned_code_point_range_start, previous_code_point });
- assigned_code_point_range_start = data.code_point;
- }
- }
- bool has_special_casing { false };
- for (auto const& casing : unicode_data.special_casing) {
- if (casing.code_point == data.code_point) {
- data.special_casing_indices.append(casing.index);
- has_special_casing = true;
- }
- }
- unicode_data.code_points_with_non_zero_combining_class += data.canonical_combining_class != 0;
- unicode_data.simple_uppercase_mapping_size += data.simple_uppercase_mapping.has_value();
- unicode_data.simple_lowercase_mapping_size += data.simple_lowercase_mapping.has_value();
- unicode_data.code_points_with_decomposition_mapping += data.decomposition_mapping.has_value();
- unicode_data.code_points_with_special_casing += has_special_casing;
- unicode_data.largest_special_casing_size = max(unicode_data.largest_special_casing_size, data.special_casing_indices.size());
- previous_code_point = data.code_point;
- unicode_data.code_point_data.append(move(data));
- }
- return {};
- }
- static ErrorOr<void> generate_unicode_data_header(Core::Stream::BufferedFile& file, UnicodeData& unicode_data)
- {
- StringBuilder builder;
- SourceGenerator generator { builder };
- generator.set("casing_transform_size", String::number(unicode_data.largest_casing_transform_size));
- auto generate_enum = [&](StringView name, StringView default_, Vector<String> values, Vector<Alias> aliases = {}) {
- quick_sort(values);
- quick_sort(aliases, [](auto& alias1, auto& alias2) { return alias1.alias < alias2.alias; });
- generator.set("name", name);
- generator.set("underlying", String::formatted("{}UnderlyingType", name));
- generator.set("type", ((values.size() + !default_.is_empty()) < 256) ? "u8"sv : "u16"sv);
- generator.append(R"~~~(
- using @underlying@ = @type@;
- enum class @name@ : @underlying@ {)~~~");
- if (!default_.is_empty()) {
- generator.set("default", default_);
- generator.append(R"~~~(
- @default@,)~~~");
- }
- for (auto const& value : values) {
- generator.set("value", value);
- generator.append(R"~~~(
- @value@,)~~~");
- }
- for (auto const& alias : aliases) {
- generator.set("alias", alias.alias);
- generator.set("value", alias.name);
- generator.append(R"~~~(
- @alias@ = @value@,)~~~");
- }
- generator.append(R"~~~(
- };
- )~~~");
- };
- generator.append(R"~~~(
- #pragma once
- #include <AK/Types.h>
- #include <LibUnicode/Forward.h>
- namespace Unicode {
- )~~~");
- generate_enum("Locale"sv, "None"sv, unicode_data.locales);
- generate_enum("Condition"sv, "None"sv, move(unicode_data.conditions));
- generate_enum("GeneralCategory"sv, {}, unicode_data.general_categories.keys(), unicode_data.general_category_aliases);
- generate_enum("Property"sv, {}, unicode_data.prop_list.keys(), unicode_data.prop_aliases);
- generate_enum("Script"sv, {}, unicode_data.script_list.keys(), unicode_data.script_aliases);
- generate_enum("Block"sv, {}, unicode_data.block_list.keys(), unicode_data.block_aliases);
- generate_enum("GraphemeBreakProperty"sv, {}, unicode_data.grapheme_break_props.keys());
- generate_enum("WordBreakProperty"sv, {}, unicode_data.word_break_props.keys());
- generate_enum("SentenceBreakProperty"sv, {}, unicode_data.sentence_break_props.keys());
- generate_enum("CompatibilityFormattingTag"sv, "Canonical"sv, unicode_data.compatibility_tags);
- generator.append(R"~~~(
- struct SpecialCasing {
- u32 code_point { 0 };
- u32 lowercase_mapping[@casing_transform_size@];
- u32 lowercase_mapping_size { 0 };
- u32 uppercase_mapping[@casing_transform_size@];
- u32 uppercase_mapping_size { 0 };
- u32 titlecase_mapping[@casing_transform_size@];
- u32 titlecase_mapping_size { 0 };
- Locale locale { Locale::None };
- Condition condition { Condition::None };
- };
- struct CodePointDecomposition {
- u32 code_point { 0 };
- CompatibilityFormattingTag tag { CompatibilityFormattingTag::Canonical };
- Span<u32 const> decomposition;
- };
- Optional<Locale> locale_from_string(StringView locale);
- }
- )~~~");
- TRY(file.write(generator.as_string_view().bytes()));
- return {};
- }
- static ErrorOr<void> generate_unicode_data_implementation(Core::Stream::BufferedFile& file, UnicodeData const& unicode_data)
- {
- StringBuilder builder;
- SourceGenerator generator { builder };
- generator.set("string_index_type"sv, s_string_index_type);
- generator.set("largest_special_casing_size", String::number(unicode_data.largest_special_casing_size));
- generator.set("special_casing_size", String::number(unicode_data.special_casing.size()));
- generator.append(R"~~~(
- #include <AK/Array.h>
- #include <AK/BinarySearch.h>
- #include <AK/CharacterTypes.h>
- #include <AK/Optional.h>
- #include <AK/Span.h>
- #include <AK/String.h>
- #include <AK/StringView.h>
- #include <LibUnicode/CharacterTypes.h>
- #include <LibUnicode/UnicodeData.h>
- #include <LibUnicode/Normalize.h>
- namespace Unicode {
- )~~~");
- unicode_data.unique_strings.generate(generator);
- auto append_list_and_size = [&](auto const& list, StringView format) {
- if (list.is_empty()) {
- generator.append(", {}, 0");
- return;
- }
- bool first = true;
- generator.append(", {");
- for (auto const& item : list) {
- generator.append(first ? " "sv : ", "sv);
- generator.append(String::formatted(format, item));
- first = false;
- }
- generator.append(String::formatted(" }}, {}", list.size()));
- };
- generator.append(R"~~~(
- static constexpr Array<SpecialCasing, @special_casing_size@> s_special_casing { {)~~~");
- for (auto const& casing : unicode_data.special_casing) {
- generator.set("code_point", String::formatted("{:#x}", casing.code_point));
- generator.append(R"~~~(
- { @code_point@)~~~");
- constexpr auto format = "0x{:x}"sv;
- append_list_and_size(casing.lowercase_mapping, format);
- append_list_and_size(casing.uppercase_mapping, format);
- append_list_and_size(casing.titlecase_mapping, format);
- generator.set("locale", casing.locale.is_empty() ? "None" : casing.locale);
- generator.append(", Locale::@locale@");
- generator.set("condition", casing.condition.is_empty() ? "None" : casing.condition);
- generator.append(", Condition::@condition@");
- generator.append(" },");
- }
- generator.append(R"~~~(
- } };
- struct CodePointMapping {
- u32 code_point { 0 };
- u32 mapping { 0 };
- };
- struct SpecialCaseMapping {
- u32 code_point { 0 };
- Array<SpecialCasing const*, @largest_special_casing_size@> special_casing {};
- u32 special_casing_size { 0 };
- };
- struct CodePointAbbreviation {
- u32 code_point { 0 };
- @string_index_type@ abbreviation { 0 };
- };
- template<typename MappingType>
- struct CodePointComparator {
- constexpr int operator()(u32 code_point, MappingType const& mapping)
- {
- return code_point - mapping.code_point;
- }
- };
- struct CodePointRangeComparator {
- constexpr int operator()(u32 code_point, CodePointRange const& range)
- {
- return (code_point > range.last) - (code_point < range.first);
- }
- };
- struct BlockNameData {
- CodePointRange code_point_range {};
- @string_index_type@ display_name { 0 };
- };
- struct BlockNameComparator : public CodePointRangeComparator {
- constexpr int operator()(u32 code_point, BlockNameData const& name)
- {
- return CodePointRangeComparator::operator()(code_point, name.code_point_range);
- }
- };
- struct CodePointName {
- CodePointRange code_point_range {};
- @string_index_type@ display_name { 0 };
- };
- struct CodePointNameComparator : public CodePointRangeComparator {
- constexpr int operator()(u32 code_point, CodePointName const& name)
- {
- return CodePointRangeComparator::operator()(code_point, name.code_point_range);
- }
- };
- )~~~");
- generator.set("decomposition_mappings_size", String::number(unicode_data.decomposition_mappings.size()));
- generator.append("\nstatic constexpr Array<u32, @decomposition_mappings_size@> s_decomposition_mappings_data { ");
- generator.append(String::join(", "sv, unicode_data.decomposition_mappings, "{:#x}"sv));
- generator.append(" };\n");
- auto append_code_point_mappings = [&](StringView name, StringView mapping_type, u32 size, auto mapping_getter) {
- generator.set("name", name);
- generator.set("mapping_type", mapping_type);
- generator.set("size", String::number(size));
- generator.append(R"~~~(
- static constexpr Array<@mapping_type@, @size@> s_@name@_mappings { {
- )~~~");
- constexpr size_t max_mappings_per_row = 20;
- size_t mappings_in_current_row = 0;
- for (auto const& data : unicode_data.code_point_data) {
- auto mapping = mapping_getter(data);
- if constexpr (requires { mapping.has_value(); }) {
- if (!mapping.has_value())
- continue;
- } else {
- if (mapping.is_empty())
- continue;
- }
- if (mappings_in_current_row++ > 0)
- generator.append(" ");
- generator.set("code_point", String::formatted("{:#x}", data.code_point));
- generator.append("{ @code_point@");
- if constexpr (IsSame<decltype(mapping), Optional<u32>> || IsSame<decltype(mapping), Optional<StringIndexType>>) {
- generator.set("mapping", String::formatted("{:#x}", *mapping));
- generator.append(", @mapping@ },");
- } else if constexpr (IsSame<decltype(mapping), Optional<CodePointDecomposition>>) {
- generator.set("tag", mapping->tag);
- generator.set("start", String::number(mapping->decomposition_index));
- generator.set("size", String::number(mapping->decomposition_size));
- generator.append(", CompatibilityFormattingTag::@tag@, Span<u32 const> { s_decomposition_mappings_data.data() + @start@, @size@ } },");
- } else {
- append_list_and_size(data.special_casing_indices, "&s_special_casing[{}]"sv);
- generator.append(" },");
- }
- if (mappings_in_current_row == max_mappings_per_row) {
- mappings_in_current_row = 0;
- generator.append("\n ");
- }
- }
- generator.append(R"~~~(
- } };
- )~~~");
- };
- append_code_point_mappings("combining_class"sv, "CodePointMapping"sv, unicode_data.code_points_with_non_zero_combining_class,
- [](auto const& data) -> Optional<u32> {
- if (data.canonical_combining_class == 0)
- return {};
- return data.canonical_combining_class;
- });
- append_code_point_mappings("uppercase"sv, "CodePointMapping"sv, unicode_data.simple_uppercase_mapping_size, [](auto const& data) { return data.simple_uppercase_mapping; });
- append_code_point_mappings("lowercase"sv, "CodePointMapping"sv, unicode_data.simple_lowercase_mapping_size, [](auto const& data) { return data.simple_lowercase_mapping; });
- append_code_point_mappings("special_case"sv, "SpecialCaseMapping"sv, unicode_data.code_points_with_special_casing, [](auto const& data) { return data.special_casing_indices; });
- append_code_point_mappings("abbreviation"sv, "CodePointAbbreviation"sv, unicode_data.code_point_abbreviations.size(), [](auto const& data) { return data.abbreviation; });
- append_code_point_mappings("decomposition"sv, "CodePointDecomposition"sv, unicode_data.code_points_with_decomposition_mapping,
- [](auto const& data) {
- return data.decomposition_mapping;
- });
- auto append_code_point_range_list = [&](String name, Vector<CodePointRange> const& ranges) {
- generator.set("name", name);
- generator.set("size", String::number(ranges.size()));
- generator.append(R"~~~(
- static constexpr Array<CodePointRange, @size@> @name@ { {
- )~~~");
- constexpr size_t max_ranges_per_row = 20;
- size_t ranges_in_current_row = 0;
- for (auto const& range : ranges) {
- if (ranges_in_current_row++ > 0)
- generator.append(" ");
- generator.set("first", String::formatted("{:#x}", range.first));
- generator.set("last", String::formatted("{:#x}", range.last));
- generator.append("{ @first@, @last@ },");
- if (ranges_in_current_row == max_ranges_per_row) {
- ranges_in_current_row = 0;
- generator.append("\n ");
- }
- }
- generator.append(R"~~~(
- } };
- )~~~");
- };
- auto append_prop_list = [&](StringView collection_name, StringView property_format, PropList const& property_list) {
- for (auto const& property : property_list) {
- auto name = String::formatted(property_format, property.key);
- append_code_point_range_list(move(name), property.value);
- }
- auto property_names = property_list.keys();
- quick_sort(property_names);
- generator.set("name", collection_name);
- generator.set("size", String::number(property_names.size()));
- generator.append(R"~~~(
- static constexpr Array<Span<CodePointRange const>, @size@> @name@ { {)~~~");
- for (auto const& property_name : property_names) {
- generator.set("name", String::formatted(property_format, property_name));
- generator.append(R"~~~(
- @name@.span(),)~~~");
- }
- generator.append(R"~~~(
- } };
- )~~~");
- };
- append_prop_list("s_general_categories"sv, "s_general_category_{}"sv, unicode_data.general_categories);
- append_prop_list("s_properties"sv, "s_property_{}"sv, unicode_data.prop_list);
- append_prop_list("s_scripts"sv, "s_script_{}"sv, unicode_data.script_list);
- append_prop_list("s_script_extensions"sv, "s_script_extension_{}"sv, unicode_data.script_extensions);
- append_prop_list("s_blocks"sv, "s_block_{}"sv, unicode_data.block_list);
- append_prop_list("s_grapheme_break_properties"sv, "s_grapheme_break_property_{}"sv, unicode_data.grapheme_break_props);
- append_prop_list("s_word_break_properties"sv, "s_word_break_property_{}"sv, unicode_data.word_break_props);
- append_prop_list("s_sentence_break_properties"sv, "s_sentence_break_property_{}"sv, unicode_data.sentence_break_props);
- auto append_code_point_display_names = [&](StringView type, StringView name, auto const& display_names) {
- constexpr size_t max_values_per_row = 30;
- size_t values_in_current_row = 0;
- generator.set("type", type);
- generator.set("name", name);
- generator.set("size", String::number(display_names.size()));
- generator.append(R"~~~(
- static constexpr Array<@type@, @size@> @name@ { {
- )~~~");
- for (auto const& display_name : display_names) {
- if (values_in_current_row++ > 0)
- generator.append(", ");
- generator.set("first", String::formatted("{:#x}", display_name.code_point_range.first));
- generator.set("last", String::formatted("{:#x}", display_name.code_point_range.last));
- generator.set("name", String::number(display_name.name));
- generator.append("{ { @first@, @last@ }, @name@ }");
- if (values_in_current_row == max_values_per_row) {
- values_in_current_row = 0;
- generator.append(",\n ");
- }
- }
- generator.append(R"~~~(
- } };
- )~~~");
- };
- append_code_point_display_names("BlockNameData"sv, "s_block_display_names"sv, unicode_data.block_display_names);
- append_code_point_display_names("CodePointName"sv, "s_code_point_display_names"sv, unicode_data.code_point_display_names);
- generator.append(R"~~~(
- Optional<StringView> code_point_block_display_name(u32 code_point)
- {
- if (auto const* entry = binary_search(s_block_display_names, code_point, nullptr, BlockNameComparator {}))
- return decode_string(entry->display_name);
- return {};
- }
- Span<BlockName const> block_display_names()
- {
- static auto display_names = []() {
- Array<BlockName, s_block_display_names.size()> display_names;
- for (size_t i = 0; i < s_block_display_names.size(); ++i) {
- auto const& display_name = s_block_display_names[i];
- display_names[i] = { display_name.code_point_range, decode_string(display_name.display_name) };
- }
- return display_names;
- }();
- return display_names.span();
- }
- Optional<String> code_point_display_name(u32 code_point)
- {
- if (auto const* entry = binary_search(s_code_point_display_names, code_point, nullptr, CodePointNameComparator {})) {
- auto display_name = decode_string(entry->display_name);
- if (display_name.ends_with("{:X}"sv))
- return String::formatted(display_name, code_point);
- return display_name;
- }
- return {};
- }
- )~~~");
- auto append_code_point_mapping_search = [&](StringView method, StringView mappings, StringView fallback) {
- generator.set("method", method);
- generator.set("mappings", mappings);
- generator.set("fallback", fallback);
- generator.append(R"~~~(
- u32 @method@(u32 code_point)
- {
- auto const* mapping = binary_search(@mappings@, code_point, nullptr, CodePointComparator<CodePointMapping> {});
- return mapping ? mapping->mapping : @fallback@;
- }
- )~~~");
- };
- append_code_point_mapping_search("canonical_combining_class"sv, "s_combining_class_mappings"sv, "0"sv);
- append_code_point_mapping_search("to_unicode_uppercase"sv, "s_uppercase_mappings"sv, "code_point"sv);
- append_code_point_mapping_search("to_unicode_lowercase"sv, "s_lowercase_mappings"sv, "code_point"sv);
- generator.append(R"~~~(
- Span<SpecialCasing const* const> special_case_mapping(u32 code_point)
- {
- auto const* mapping = binary_search(s_special_case_mappings, code_point, nullptr, CodePointComparator<SpecialCaseMapping> {});
- if (mapping == nullptr)
- return {};
- return mapping->special_casing.span().slice(0, mapping->special_casing_size);
- }
- Optional<StringView> code_point_abbreviation(u32 code_point)
- {
- auto const* mapping = binary_search(s_abbreviation_mappings, code_point, nullptr, CodePointComparator<CodePointAbbreviation> {});
- if (mapping == nullptr)
- return {};
- if (mapping->abbreviation == 0)
- return {};
- return decode_string(mapping->abbreviation);
- }
- Optional<CodePointDecomposition const&> code_point_decomposition(u32 code_point)
- {
- auto const* mapping = binary_search(s_decomposition_mappings, code_point, nullptr, CodePointComparator<CodePointDecomposition> {});
- if (mapping == nullptr)
- return {};
- return *mapping;
- }
- Span<CodePointDecomposition const> code_point_decompositions()
- {
- return s_decomposition_mappings;
- }
- )~~~");
- auto append_prop_search = [&](StringView enum_title, StringView enum_snake, StringView collection_name) {
- generator.set("enum_title", enum_title);
- generator.set("enum_snake", enum_snake);
- generator.set("collection_name", collection_name);
- generator.append(R"~~~(
- bool code_point_has_@enum_snake@(u32 code_point, @enum_title@ @enum_snake@)
- {
- auto index = static_cast<@enum_title@UnderlyingType>(@enum_snake@);
- auto const& ranges = @collection_name@.at(index);
- auto const* range = binary_search(ranges, code_point, nullptr, CodePointRangeComparator {});
- return range != nullptr;
- }
- )~~~");
- };
- auto append_from_string = [&](StringView enum_title, StringView enum_snake, auto const& prop_list, Vector<Alias> const& aliases) {
- HashValueMap<StringView> hashes;
- hashes.ensure_capacity(prop_list.size() + aliases.size());
- ValueFromStringOptions options {};
- for (auto const& prop : prop_list) {
- if constexpr (IsSame<RemoveCVReference<decltype(prop)>, String>) {
- hashes.set(CaseInsensitiveStringViewTraits::hash(prop), prop);
- options.sensitivity = CaseSensitivity::CaseInsensitive;
- } else {
- hashes.set(prop.key.hash(), prop.key);
- }
- }
- for (auto const& alias : aliases)
- hashes.set(alias.alias.hash(), alias.alias);
- generate_value_from_string(generator, "{}_from_string"sv, enum_title, enum_snake, move(hashes), options);
- };
- append_from_string("Locale"sv, "locale"sv, unicode_data.locales, {});
- append_prop_search("GeneralCategory"sv, "general_category"sv, "s_general_categories"sv);
- append_from_string("GeneralCategory"sv, "general_category"sv, unicode_data.general_categories, unicode_data.general_category_aliases);
- append_prop_search("Property"sv, "property"sv, "s_properties"sv);
- append_from_string("Property"sv, "property"sv, unicode_data.prop_list, unicode_data.prop_aliases);
- append_prop_search("Script"sv, "script"sv, "s_scripts"sv);
- append_prop_search("Script"sv, "script_extension"sv, "s_script_extensions"sv);
- append_from_string("Script"sv, "script"sv, unicode_data.script_list, unicode_data.script_aliases);
- append_prop_search("Block"sv, "block"sv, "s_blocks"sv);
- append_from_string("Block"sv, "block"sv, unicode_data.block_list, unicode_data.block_aliases);
- append_prop_search("GraphemeBreakProperty"sv, "grapheme_break_property"sv, "s_grapheme_break_properties"sv);
- append_prop_search("WordBreakProperty"sv, "word_break_property"sv, "s_word_break_properties"sv);
- append_prop_search("SentenceBreakProperty"sv, "sentence_break_property"sv, "s_sentence_break_properties"sv);
- generator.append(R"~~~(
- }
- )~~~");
- TRY(file.write(generator.as_string_view().bytes()));
- return {};
- }
- static Vector<u32> flatten_code_point_ranges(Vector<CodePointRange> const& code_points)
- {
- Vector<u32> flattened;
- for (auto const& range : code_points) {
- flattened.grow_capacity(range.last - range.first);
- for (u32 code_point = range.first; code_point <= range.last; ++code_point)
- flattened.append(code_point);
- }
- return flattened;
- }
- static Vector<CodePointRange> form_code_point_ranges(Vector<u32> code_points)
- {
- Vector<CodePointRange> ranges;
- u32 range_start = code_points[0];
- u32 range_end = range_start;
- for (size_t i = 1; i < code_points.size(); ++i) {
- u32 code_point = code_points[i];
- if ((code_point - range_end) == 1) {
- range_end = code_point;
- } else {
- ranges.append({ range_start, range_end });
- range_start = code_point;
- range_end = code_point;
- }
- }
- ranges.append({ range_start, range_end });
- return ranges;
- }
- static void sort_and_merge_code_point_ranges(Vector<CodePointRange>& code_points)
- {
- quick_sort(code_points, [](auto const& range1, auto const& range2) {
- return range1.first < range2.first;
- });
- for (size_t i = 0; i < code_points.size() - 1;) {
- if (code_points[i].last >= code_points[i + 1].first) {
- code_points[i].last = max(code_points[i].last, code_points[i + 1].last);
- code_points.remove(i + 1);
- } else {
- ++i;
- }
- }
- auto all_code_points = flatten_code_point_ranges(code_points);
- code_points = form_code_point_ranges(all_code_points);
- }
- static void populate_general_category_unions(PropList& general_categories)
- {
- // The Unicode standard defines General Category values which are not in any UCD file. These
- // values are simply unions of other values.
- // https://www.unicode.org/reports/tr44/#GC_Values_Table
- auto populate_union = [&](auto alias, auto categories) {
- auto& code_points = general_categories.ensure(alias);
- for (auto const& category : categories)
- code_points.extend(general_categories.find(category)->value);
- sort_and_merge_code_point_ranges(code_points);
- };
- populate_union("LC"sv, Array { "Ll"sv, "Lu"sv, "Lt"sv });
- populate_union("L"sv, Array { "Lu"sv, "Ll"sv, "Lt"sv, "Lm"sv, "Lo"sv });
- populate_union("M"sv, Array { "Mn"sv, "Mc"sv, "Me"sv });
- populate_union("N"sv, Array { "Nd"sv, "Nl"sv, "No"sv });
- populate_union("P"sv, Array { "Pc"sv, "Pd"sv, "Ps"sv, "Pe"sv, "Pi"sv, "Pf"sv, "Po"sv });
- populate_union("S"sv, Array { "Sm"sv, "Sc"sv, "Sk"sv, "So"sv });
- populate_union("Z"sv, Array { "Zs"sv, "Zl"sv, "Zp"sv });
- populate_union("C"sv, Array { "Cc"sv, "Cf"sv, "Cs"sv, "Co"sv, "Cn"sv });
- }
- static void normalize_script_extensions(PropList& script_extensions, PropList const& script_list, Vector<Alias> const& script_aliases)
- {
- // The ScriptExtensions UCD file lays out its code point ranges rather uniquely compared to
- // other files. The Script listed on each line may either be a full Script string or an aliased
- // abbreviation. Further, the extensions may or may not include the base Script list. Normalize
- // the extensions here to be keyed by the full Script name and always include the base list.
- auto extensions = move(script_extensions);
- script_extensions = script_list;
- for (auto const& extension : extensions) {
- auto it = find_if(script_aliases.begin(), script_aliases.end(), [&](auto const& alias) { return extension.key == alias.alias; });
- auto const& key = (it == script_aliases.end()) ? extension.key : it->name;
- auto& code_points = script_extensions.find(key)->value;
- code_points.extend(extension.value);
- sort_and_merge_code_point_ranges(code_points);
- }
- // Lastly, the Common and Inherited script extensions are special. They must not contain any
- // code points which appear in other script extensions. The ScriptExtensions UCD file does not
- // list these extensions, therefore this peculiarity must be handled programmatically.
- // https://www.unicode.org/reports/tr24/#Assignment_ScriptX_Values
- auto code_point_has_other_extension = [&](StringView key, u32 code_point) {
- for (auto const& extension : extensions) {
- if (extension.key == key)
- continue;
- if (any_of(extension.value, [&](auto const& r) { return (r.first <= code_point) && (code_point <= r.last); }))
- return true;
- }
- return false;
- };
- auto get_code_points_without_other_extensions = [&](StringView key) {
- auto code_points = flatten_code_point_ranges(script_list.find(key)->value);
- code_points.remove_all_matching([&](u32 c) { return code_point_has_other_extension(key, c); });
- return code_points;
- };
- auto common_code_points = get_code_points_without_other_extensions("Common"sv);
- script_extensions.set("Common"sv, form_code_point_ranges(common_code_points));
- auto inherited_code_points = get_code_points_without_other_extensions("Inherited"sv);
- script_extensions.set("Inherited"sv, form_code_point_ranges(inherited_code_points));
- }
- ErrorOr<int> serenity_main(Main::Arguments arguments)
- {
- StringView generated_header_path;
- StringView generated_implementation_path;
- StringView unicode_data_path;
- StringView special_casing_path;
- StringView derived_general_category_path;
- StringView prop_list_path;
- StringView derived_core_prop_path;
- StringView derived_binary_prop_path;
- StringView prop_alias_path;
- StringView prop_value_alias_path;
- StringView name_alias_path;
- StringView scripts_path;
- StringView script_extensions_path;
- StringView blocks_path;
- StringView emoji_data_path;
- StringView normalization_path;
- StringView grapheme_break_path;
- StringView word_break_path;
- StringView sentence_break_path;
- Core::ArgsParser args_parser;
- args_parser.add_option(generated_header_path, "Path to the Unicode Data header file to generate", "generated-header-path", 'h', "generated-header-path");
- args_parser.add_option(generated_implementation_path, "Path to the Unicode Data implementation file to generate", "generated-implementation-path", 'c', "generated-implementation-path");
- args_parser.add_option(unicode_data_path, "Path to UnicodeData.txt file", "unicode-data-path", 'u', "unicode-data-path");
- args_parser.add_option(special_casing_path, "Path to SpecialCasing.txt file", "special-casing-path", 's', "special-casing-path");
- args_parser.add_option(derived_general_category_path, "Path to DerivedGeneralCategory.txt file", "derived-general-category-path", 'g', "derived-general-category-path");
- args_parser.add_option(prop_list_path, "Path to PropList.txt file", "prop-list-path", 'p', "prop-list-path");
- args_parser.add_option(derived_core_prop_path, "Path to DerivedCoreProperties.txt file", "derived-core-prop-path", 'd', "derived-core-prop-path");
- args_parser.add_option(derived_binary_prop_path, "Path to DerivedBinaryProperties.txt file", "derived-binary-prop-path", 'b', "derived-binary-prop-path");
- args_parser.add_option(prop_alias_path, "Path to PropertyAliases.txt file", "prop-alias-path", 'a', "prop-alias-path");
- args_parser.add_option(prop_value_alias_path, "Path to PropertyValueAliases.txt file", "prop-value-alias-path", 'v', "prop-value-alias-path");
- args_parser.add_option(name_alias_path, "Path to NameAliases.txt file", "name-alias-path", 'm', "name-alias-path");
- args_parser.add_option(scripts_path, "Path to Scripts.txt file", "scripts-path", 'r', "scripts-path");
- args_parser.add_option(script_extensions_path, "Path to ScriptExtensions.txt file", "script-extensions-path", 'x', "script-extensions-path");
- args_parser.add_option(blocks_path, "Path to Blocks.txt file", "blocks-path", 'k', "blocks-path");
- args_parser.add_option(emoji_data_path, "Path to emoji-data.txt file", "emoji-data-path", 'e', "emoji-data-path");
- args_parser.add_option(normalization_path, "Path to DerivedNormalizationProps.txt file", "normalization-path", 'n', "normalization-path");
- args_parser.add_option(grapheme_break_path, "Path to GraphemeBreakProperty.txt file", "grapheme-break-path", 'f', "grapheme-break-path");
- args_parser.add_option(word_break_path, "Path to WordBreakProperty.txt file", "word-break-path", 'w', "word-break-path");
- args_parser.add_option(sentence_break_path, "Path to SentenceBreakProperty.txt file", "sentence-break-path", 'i', "sentence-break-path");
- args_parser.parse(arguments);
- auto generated_header_file = TRY(open_file(generated_header_path, Core::Stream::OpenMode::Write));
- auto generated_implementation_file = TRY(open_file(generated_implementation_path, Core::Stream::OpenMode::Write));
- auto unicode_data_file = TRY(open_file(unicode_data_path, Core::Stream::OpenMode::Read));
- auto derived_general_category_file = TRY(open_file(derived_general_category_path, Core::Stream::OpenMode::Read));
- auto special_casing_file = TRY(open_file(special_casing_path, Core::Stream::OpenMode::Read));
- auto prop_list_file = TRY(open_file(prop_list_path, Core::Stream::OpenMode::Read));
- auto derived_core_prop_file = TRY(open_file(derived_core_prop_path, Core::Stream::OpenMode::Read));
- auto derived_binary_prop_file = TRY(open_file(derived_binary_prop_path, Core::Stream::OpenMode::Read));
- auto prop_alias_file = TRY(open_file(prop_alias_path, Core::Stream::OpenMode::Read));
- auto prop_value_alias_file = TRY(open_file(prop_value_alias_path, Core::Stream::OpenMode::Read));
- auto name_alias_file = TRY(open_file(name_alias_path, Core::Stream::OpenMode::Read));
- auto scripts_file = TRY(open_file(scripts_path, Core::Stream::OpenMode::Read));
- auto script_extensions_file = TRY(open_file(script_extensions_path, Core::Stream::OpenMode::Read));
- auto blocks_file = TRY(open_file(blocks_path, Core::Stream::OpenMode::Read));
- auto emoji_data_file = TRY(open_file(emoji_data_path, Core::Stream::OpenMode::Read));
- auto normalization_file = TRY(open_file(normalization_path, Core::Stream::OpenMode::Read));
- auto grapheme_break_file = TRY(open_file(grapheme_break_path, Core::Stream::OpenMode::Read));
- auto word_break_file = TRY(open_file(word_break_path, Core::Stream::OpenMode::Read));
- auto sentence_break_file = TRY(open_file(sentence_break_path, Core::Stream::OpenMode::Read));
- UnicodeData unicode_data {};
- TRY(parse_special_casing(*special_casing_file, unicode_data));
- TRY(parse_prop_list(*derived_general_category_file, unicode_data.general_categories));
- TRY(parse_prop_list(*prop_list_file, unicode_data.prop_list));
- TRY(parse_prop_list(*derived_core_prop_file, unicode_data.prop_list));
- TRY(parse_prop_list(*derived_binary_prop_file, unicode_data.prop_list));
- TRY(parse_prop_list(*emoji_data_file, unicode_data.prop_list));
- TRY(parse_normalization_props(*normalization_file, unicode_data));
- TRY(parse_alias_list(*prop_alias_file, unicode_data.prop_list, unicode_data.prop_aliases));
- TRY(parse_prop_list(*scripts_file, unicode_data.script_list));
- TRY(parse_prop_list(*script_extensions_file, unicode_data.script_extensions, true));
- TRY(parse_block_display_names(*blocks_file, unicode_data));
- TRY(parse_prop_list(*blocks_file, unicode_data.block_list, false, true));
- TRY(parse_name_aliases(*name_alias_file, unicode_data));
- TRY(parse_prop_list(*grapheme_break_file, unicode_data.grapheme_break_props));
- TRY(parse_prop_list(*word_break_file, unicode_data.word_break_props));
- TRY(parse_prop_list(*sentence_break_file, unicode_data.sentence_break_props));
- populate_general_category_unions(unicode_data.general_categories);
- TRY(parse_unicode_data(*unicode_data_file, unicode_data));
- TRY(parse_value_alias_list(*prop_value_alias_file, "gc"sv, unicode_data.general_categories.keys(), unicode_data.general_category_aliases));
- TRY(parse_value_alias_list(*prop_value_alias_file, "sc"sv, unicode_data.script_list.keys(), unicode_data.script_aliases, false));
- TRY(parse_value_alias_list(*prop_value_alias_file, "blk"sv, unicode_data.block_list.keys(), unicode_data.block_aliases, false, true));
- normalize_script_extensions(unicode_data.script_extensions, unicode_data.script_list, unicode_data.script_aliases);
- TRY(generate_unicode_data_header(*generated_header_file, unicode_data));
- TRY(generate_unicode_data_implementation(*generated_implementation_file, unicode_data));
- return 0;
- }
|