llvm/utils/UnicodeData/UnicodeNameMappingGenerator.cpp - rust-lang/llvm-project - Git at Google

 //===--- UnicodeNameMappingGenerator.cpp - Unicode name data generator ---===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 //
 // This file is used to generate lib/Support/UnicodeNameToCodepointGenerated.cpp
 // using UnicodeData.txt and NameAliases.txt available at
 // https://unicode.org/Public/15.0.0/ucd/
 //===----------------------------------------------------------------------===//

 #include "llvm/ADT/STLExtras.h"
 #include "llvm/ADT/StringExtras.h"
 #include "llvm/ADT/StringRef.h"
 #include <algorithm>
 #include <array>
 #include <deque>
 #include <fstream>
 #include <memory>
 #include <optional>
 #include <set>
 #include <string>
 #include <unordered_map>
 #include <utility>
 #include <vector>

 static const llvm::StringRef Letters =
     " _-ABCDEFGHIJKLMNOPQRSTUVWXYZ0123456789";

 // Collect names UnicodeData.txt and AliasNames.txt
 // There may be multiple names per code points.
 static std::unordered_multimap<char32_t, std::string>
 loadDataFiles(const std::string &NamesFile, const std::string &AliasesFile) {
   std::unordered_multimap<char32_t, std::string> CollectedCharacters;
   auto FromFile = [&](const std::string &File, bool IsAliasFile = false) {
     std::ifstream InputFile(File);
     for (std::string Line; getline(InputFile, Line);) {
       if (Line.empty() || !isxdigit(Line[0]))
         continue;
       auto FirstSemiPos = Line.find(';');
       if (FirstSemiPos == std::string::npos)
         continue;
       auto SecondSemiPos = Line.find(';', FirstSemiPos + 1);
       if (SecondSemiPos == std::string::npos)
         continue;
       unsigned long long CodePoint;
       if (llvm::getAsUnsignedInteger(
               llvm::StringRef(Line.c_str(), FirstSemiPos), 16, CodePoint)) {
         continue;
       }

       std::string Name =
           Line.substr(FirstSemiPos + 1, SecondSemiPos - FirstSemiPos - 1);

       if (!Name.empty() && Name[0] == '<') {
         // Ignore ranges of characters, as their name is either absent or
         // generated.
         continue;
       }

       // Some aliases are ignored for compatibility with C++
       if (IsAliasFile) {
         std::string Kind = Line.substr(SecondSemiPos + 1);
         if (Kind != "control" && Kind != "correction" && Kind != "alternate")
           continue;
       }

       auto InsertUnique = [&](char32_t CP, std::string Name) {
         auto It = CollectedCharacters.find(CP);
         while (It != std::end(CollectedCharacters) && It->first == CP) {
           if (It->second == Name)
             return;
           ++It;
         }
         CollectedCharacters.insert({CP, std::move(Name)});
       };
       InsertUnique(CodePoint, std::move(Name));
     }
   };

   FromFile(NamesFile);
   FromFile(AliasesFile, true);
   return CollectedCharacters;
 }

 class Trie {
   struct Node;

 public:
   // When inserting named codepoint
   // We create a node per character in the name.
   // SPARKLE becomes S <- P <- A <- R <- K <- L <- E
   // Once all  characters are inserted, the tree is compacted
   void insert(llvm::StringRef Name, char32_t Codepoint) {
     Node *N = Root.get();
     for (auto Ch : Name) {
       std::string Label(1, Ch);
       auto It = llvm::find_if(N->Children,
                               [&](const auto &C) { return C->Name == Label; });
       if (It == N->Children.end()) {
         It = N->Children.insert(It, std::make_unique<Node>(Label, N));
       }
       N = It->get();
     }
     N->Value = Codepoint;
   }

   void compact() { compact(Root.get()); }

   // This creates 2 arrays of bytes from the tree:
   // A serialized dictionary of node labels,
   // And the nodes themselves.
   // The name of each label is found by indexing into the dictionary.
   // The longest names are inserted first into the dictionary,
   // in the hope it will contain shorter labels as substring,
   // thereby reducing duplication.
   // We could theorically be more clever by trying to minimizing the size
   // of the dictionary.
   std::pair<std::string, std::vector<uint8_t>> serialize() {
     std::set<std::string> Names = this->getNameFragments();
     std::vector<std::string> Sorted(Names.begin(), Names.end());
     llvm::sort(Sorted, [](const auto &a, const auto &b) {
       return a.size() > b.size();
     });
     std::string Dict(Letters.begin(), Letters.end());
     Dict.reserve(50000);
     for (const std::string &Name : Sorted) {
       if (Name.size() <= 1)
         continue;
       if (Dict.find(Name) != std::string::npos)
         continue;
       Dict += Name;
     }

     if (Dict.size() >= std::numeric_limits<uint16_t>::max()) {
       fprintf(stderr, "Dictionary too big  to be serialized");
       exit(1);
     }

     auto Bytes = dumpIndex(Dict);
     return {Dict, Bytes};
   }

   std::set<std::string> getNameFragments() {
     std::set<std::string> Keys;
     collectKeys(Root.get(), Keys);
     return Keys;
   }

   // Maps a valid char in an Unicode character name
   // To a 6 bits index.
   static uint8_t letter(char C) {
     auto Pos = Letters.find(C);
     assert(Pos != std::string::npos &&
            "Invalid letter in Unicode character name");
     return Pos;
   }

   // clang-format off
   // +================+============+======================+=============+========+===+==============+===============+
   // | 0          | 1             | 2-7 (6)              | 8-23        | 24-44  |    | 46           | 47            |
   // +================+============+======================+=============+========+===+==============+===============+
   // | Has Value |  Has Long Name | Letter OR Name Size  | Dict Index  | Value  |    | Has Sibling  | Has Children  |
   // +----------------+------------+----------------------+-------------+--------+---+--------------+---------------+
   // clang-format on

   std::vector<uint8_t> dumpIndex(const std::string &Dict) {
     struct ChildrenOffset {
       Node *FirstChild;
       std::size_t Offset;
       bool HasValue;
     };

     // Keep track of the start of each node
     // position in the serialized data.
     std::unordered_map<Node *, int32_t> Offsets;

     // Keep track of where to write the index
     // of the first children
     std::vector<ChildrenOffset> ChildrenOffsets;
     std::unordered_map<Node *, bool> SiblingTracker;
     std::deque<Node *> AllNodes;
     std::vector<uint8_t> Bytes;
     Bytes.reserve(250'000);
     // This leading byte is used by the reading code to detect the root node.
     Bytes.push_back(0);

     auto CollectChildren = [&SiblingTracker, &AllNodes](const auto &Children) {
       for (std::size_t Index = 0; Index < Children.size(); Index++) {
         const std::unique_ptr<Node> &Child = Children[Index];
         AllNodes.push_back(Child.get());
         if (Index != Children.size() - 1)
           SiblingTracker[Child.get()] = true;
       }
     };
     CollectChildren(Root->Children);

     while (!AllNodes.empty()) {
       const std::size_t Offset = Bytes.size();
       Node *const N = AllNodes.front();
       AllNodes.pop_front();

       assert(!N->Name.empty());
       Offsets[N] = Offset;

       uint8_t FirstByte = (!!N->Value) ? 0x80 : 0;
       // Single letter node are indexed in 6 bits
       if (N->Name.size() == 1) {
         FirstByte |= letter(N->Name[0]);
         Bytes.push_back(FirstByte);
       } else {
         // Otherwise we use a 16 bits index
         FirstByte = FirstByte | uint8_t(N->Name.size()) | 0x40;
         Bytes.push_back(FirstByte);
         auto PosInDict = Dict.find(N->Name);
         assert(PosInDict != std::string::npos);
         uint8_t Low = PosInDict;
         uint8_t High = ((PosInDict >> 8) & 0xFF);
         Bytes.push_back(High);
         Bytes.push_back(Low);
       }

       const bool HasSibling = SiblingTracker.count(N) != 0;
       const bool HasChildren = N->Children.size() != 0;

       if (!!N->Value) {
         uint32_t Value = (*(N->Value) << 3);
         uint8_t H = ((Value >> 16) & 0xFF);
         uint8_t M = ((Value >> 8) & 0xFF);
         uint8_t L = (Value & 0xFF) | uint8_t(HasSibling ? 0x01 : 0) |
                     uint8_t(HasChildren ? 0x02 : 0);

         Bytes.push_back(H);
         Bytes.push_back(M);
         Bytes.push_back(L);

         if (HasChildren) {
           ChildrenOffsets.push_back(
               ChildrenOffset{N->Children[0].get(), Bytes.size(), true});
           // index of the first children
           Bytes.push_back(0x00);
           Bytes.push_back(0x00);
           Bytes.push_back(0x00);
         }
       } else {
         // When there is no value (that's most intermediate nodes)
         // Dispense of the 3 values bytes, and only store
         // 1 byte to track whether the node has sibling and children
         // + 2 bytes for the index of the first children if necessary.
         // That index also uses bytes 0-6 of the previous byte.
         uint8_t Byte =
             uint8_t(HasSibling ? 0x80 : 0) | uint8_t(HasChildren ? 0x40 : 0);
         Bytes.push_back(Byte);
         if (HasChildren) {
           ChildrenOffsets.emplace_back(
               ChildrenOffset{N->Children[0].get(), Bytes.size() - 1, false});
           Bytes.push_back(0x00);
           Bytes.push_back(0x00);
         }
       }
       CollectChildren(N->Children);
     }

     // Once all the nodes are in the inndex
     // Fill the bytes we left to indicate the position
     // of the children
     for (const ChildrenOffset &Parent : ChildrenOffsets) {
       const auto It = Offsets.find(Parent.FirstChild);
       assert(It != Offsets.end());
       std::size_t Pos = It->second;
       if (Parent.HasValue) {
         Bytes[Parent.Offset] = ((Pos >> 16) & 0xFF);
       } else {
         Bytes[Parent.Offset] =
             Bytes[Parent.Offset] | uint8_t((Pos >> 16) & 0xFF);
       }
       Bytes[Parent.Offset + 1] = ((Pos >> 8) & 0xFF);
       Bytes[Parent.Offset + 2] = Pos & 0xFF;
     }

     // Add some padding so that the deserialization code
     // doesn't try to read past the enf of the array.
     Bytes.push_back(0);
     Bytes.push_back(0);
     Bytes.push_back(0);
     Bytes.push_back(0);
     Bytes.push_back(0);
     Bytes.push_back(0);

     return Bytes;
   }

 private:
   void collectKeys(Node *N, std::set<std::string> &Keys) {
     Keys.insert(N->Name);
     for (const std::unique_ptr<Node> &Child : N->Children) {
       collectKeys(Child.get(), Keys);
     }
   }

   // Merge sequences of 1-character nodes
   // This greatly reduce the total number of nodes,
   // and therefore the size of the index.
   // When the tree gets serialized, we only have 5 bytes to store the
   // size of a name. Overlong names (>32 characters) are therefore
   // kep into separate nodes
   void compact(Node *N) {
     for (auto &&Child : N->Children) {
       compact(Child.get());
     }
     if (N->Parent && N->Parent->Children.size() == 1 && !N->Parent->Value &&
         (N->Parent->Name.size() + N->Name.size() <= 32)) {
       N->Parent->Value = N->Value;
       N->Parent->Name += N->Name;
       N->Parent->Children = std::move(N->Children);
       for (std::unique_ptr<Node> &c : N->Parent->Children) {
         c->Parent = N->Parent;
       }
     }
   }
   struct Node {
     Node(std::string Name, Node *Parent = nullptr)
         : Name(Name), Parent(Parent) {}

     std::vector<std::unique_ptr<Node>> Children;
     std::string Name;
     Node *Parent = nullptr;
     std::optional<char32_t> Value;
   };

   std::unique_ptr<Node> Root = std::make_unique<Node>("");
 };

 extern const char *UnicodeLicense;

 int main(int argc, char **argv) {
   printf("Unicode name -> codepoint mapping generator\n"
          "Usage: %s UnicodeData.txt NameAliases.txt output\n\n",
          argv[0]);
   printf("NameAliases.txt can be found at "
          "https://unicode.org/Public/15.0.0/ucd/NameAliases.txt\n"
          "UnicodeData.txt can be found at "
          "https://unicode.org/Public/15.0.0/ucd/UnicodeData.txt\n\n");

   if (argc != 4)
     return EXIT_FAILURE;

   FILE *Out = fopen(argv[3], "w");
   if (!Out) {
     printf("Error creating output file.\n");
     return EXIT_FAILURE;
   }

   Trie T;
   uint32_t NameCount = 0;
   std::size_t LongestName = 0;
   auto Entries = loadDataFiles(argv[1], argv[2]);
   for (const std::pair<const char32_t, std::string> &Entry : Entries) {
     char32_t Codepoint = Entry.first;
     const std::string &Name = Entry.second;
     // Ignore names which are not valid.
     if (Name.empty() ||
         !llvm::all_of(Name, [](char C) { return Letters.contains(C); })) {
       continue;
     }
     printf("%06x: %s\n", static_cast<unsigned int>(Codepoint), Name.c_str());
     T.insert(Name, Codepoint);
     LongestName =
         std::max(LongestName, std::size_t(llvm::count_if(Name, llvm::isAlnum)));
     NameCount++;
   }
   T.compact();

   std::pair<std::string, std::vector<uint8_t>> Data = T.serialize();
   const std::string &Dict = Data.first;
   const std::vector<uint8_t> &Tree = Data.second;

   fprintf(Out, R"(
 //===------------- Support/UnicodeNameToCodepointGenerated.cpp ------------===//
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 //
 // This file implements mapping the name of a unicode code point to its value.
 //
 // This file was generated using %s.
 // Do not edit manually.
 //
 //===----------------------------------------------------------------------===//
 %s


 #include "llvm/Support/Compiler.h"
 #include <cstddef>
 #include <cstdint>
 )",
           argv[0], UnicodeLicense);

   fprintf(Out,
           "namespace llvm { namespace sys { namespace unicode { \n"
           "extern const char *UnicodeNameToCodepointDict;\n"
           "extern const uint8_t *UnicodeNameToCodepointIndex;\n"
           "extern const std::size_t UnicodeNameToCodepointIndexSize;\n"
           "extern const std::size_t UnicodeNameToCodepointLargestNameSize;\n");

   fprintf(Out, "const char* UnicodeNameToCodepointDict = \"%s\";\n",
           Dict.c_str());

   fprintf(Out, "uint8_t UnicodeNameToCodepointIndex_[%zu] = {\n",
           Tree.size() + 1);

   for (auto Byte : Tree) {
     fprintf(Out, "0x%02x,", Byte);
   }

   fprintf(Out, "0};");
   fprintf(Out, "const uint8_t* UnicodeNameToCodepointIndex = "
                "UnicodeNameToCodepointIndex_; \n");
   fprintf(Out, "const std::size_t UnicodeNameToCodepointIndexSize = %zu;\n",
           Tree.size() + 1);
   fprintf(Out,
           "const std::size_t UnicodeNameToCodepointLargestNameSize = %zu;\n",
           LongestName);
   fprintf(Out, "\n}}}\n");
   fclose(Out);
   printf("Generated %s: %u Files.\nIndex: %f kB, Dictionary: %f kB.\nDone\n\n",
          argv[3], NameCount, Tree.size() / 1024.0, Dict.size() / 1024.0);
 }

 const char *UnicodeLicense = R"(
 /*
 UNICODE, INC. LICENSE AGREEMENT - DATA FILES AND SOFTWARE

 See Terms of Use <https://www.unicode.org/copyright.html>
 for definitions of Unicode Inc.’s Data Files and Software.

 NOTICE TO USER: Carefully read the following legal agreement.
 BY DOWNLOADING, INSTALLING, COPYING OR OTHERWISE USING UNICODE INC.'S
 DATA FILES ("DATA FILES"), AND/OR SOFTWARE ("SOFTWARE"),
 YOU UNEQUIVOCALLY ACCEPT, AND AGREE TO BE BOUND BY, ALL OF THE
 TERMS AND CONDITIONS OF THIS AGREEMENT.
 IF YOU DO NOT AGREE, DO NOT DOWNLOAD, INSTALL, COPY, DISTRIBUTE OR USE
 THE DATA FILES OR SOFTWARE.

 COPYRIGHT AND PERMISSION NOTICE

 Copyright © 1991-2022 Unicode, Inc. All rights reserved.
 Distributed under the Terms of Use in https://www.unicode.org/copyright.html.

 Permission is hereby granted, free of charge, to any person obtaining
 a copy of the Unicode data files and any associated documentation
 (the "Data Files") or Unicode software and any associated documentation
 (the "Software") to deal in the Data Files or Software
 without restriction, including without limitation the rights to use,
 copy, modify, merge, publish, distribute, and/or sell copies of
 the Data Files or Software, and to permit persons to whom the Data Files
 or Software are furnished to do so, provided that either
 (a) this copyright and permission notice appear with all copies
 of the Data Files or Software, or
 (b) this copyright and permission notice appear in associated
 Documentation.

 THE DATA FILES AND SOFTWARE ARE PROVIDED "AS IS", WITHOUT WARRANTY OF
 ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE
 WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
 NONINFRINGEMENT OF THIRD PARTY RIGHTS.
 IN NO EVENT SHALL THE COPYRIGHT HOLDER OR HOLDERS INCLUDED IN THIS
 NOTICE BE LIABLE FOR ANY CLAIM, OR ANY SPECIAL INDIRECT OR CONSEQUENTIAL
 DAMAGES, OR ANY DAMAGES WHATSOEVER RESULTING FROM LOSS OF USE,
 DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE OR OTHER
 TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR
 PERFORMANCE OF THE DATA FILES OR SOFTWARE.

 Except as contained in this notice, the name of a copyright holder
 shall not be used in advertising or otherwise to promote the sale,
 use or other dealings in these Data Files or Software without prior
 written authorization of the copyright holder.
 */
 )";
	//===--- UnicodeNameMappingGenerator.cpp - Unicode name data generator ---===//
	//
	// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
	// See https://llvm.org/LICENSE.txt for license information.
	// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
	//
	//===----------------------------------------------------------------------===//
	//
	// This file is used to generate lib/Support/UnicodeNameToCodepointGenerated.cpp
	// using UnicodeData.txt and NameAliases.txt available at
	// https://unicode.org/Public/15.0.0/ucd/
	//===----------------------------------------------------------------------===//

	#include "llvm/ADT/STLExtras.h"
	#include "llvm/ADT/StringExtras.h"
	#include "llvm/ADT/StringRef.h"
	#include <algorithm>
	#include <array>
	#include <deque>
	#include <fstream>
	#include <memory>
	#include <optional>
	#include <set>
	#include <string>
	#include <unordered_map>
	#include <utility>
	#include <vector>

	static const llvm::StringRef Letters =
	" _-ABCDEFGHIJKLMNOPQRSTUVWXYZ0123456789";

	// Collect names UnicodeData.txt and AliasNames.txt
	// There may be multiple names per code points.
	static std::unordered_multimap<char32_t, std::string>
	loadDataFiles(const std::string &NamesFile, const std::string &AliasesFile) {
	std::unordered_multimap<char32_t, std::string> CollectedCharacters;
	auto FromFile = [&](const std::string &File, bool IsAliasFile = false) {
	std::ifstream InputFile(File);
	for (std::string Line; getline(InputFile, Line);) {
	if (Line.empty() \|\| !isxdigit(Line[0]))
	continue;
	auto FirstSemiPos = Line.find(';');
	if (FirstSemiPos == std::string::npos)
	continue;
	auto SecondSemiPos = Line.find(';', FirstSemiPos + 1);
	if (SecondSemiPos == std::string::npos)
	continue;
	unsigned long long CodePoint;
	if (llvm::getAsUnsignedInteger(
	llvm::StringRef(Line.c_str(), FirstSemiPos), 16, CodePoint)) {
	continue;
	}

	std::string Name =
	Line.substr(FirstSemiPos + 1, SecondSemiPos - FirstSemiPos - 1);

	if (!Name.empty() && Name[0] == '<') {
	// Ignore ranges of characters, as their name is either absent or
	// generated.
	continue;
	}

	// Some aliases are ignored for compatibility with C++
	if (IsAliasFile) {
	std::string Kind = Line.substr(SecondSemiPos + 1);
	if (Kind != "control" && Kind != "correction" && Kind != "alternate")
	continue;
	}

	auto InsertUnique = [&](char32_t CP, std::string Name) {
	auto It = CollectedCharacters.find(CP);
	while (It != std::end(CollectedCharacters) && It->first == CP) {
	if (It->second == Name)
	return;
	++It;
	}
	CollectedCharacters.insert({CP, std::move(Name)});
	};
	InsertUnique(CodePoint, std::move(Name));
	}
	};

	FromFile(NamesFile);
	FromFile(AliasesFile, true);
	return CollectedCharacters;
	}

	class Trie {
	struct Node;

	public:
	// When inserting named codepoint
	// We create a node per character in the name.
	// SPARKLE becomes S <- P <- A <- R <- K <- L <- E
	// Once all characters are inserted, the tree is compacted
	void insert(llvm::StringRef Name, char32_t Codepoint) {
	Node *N = Root.get();
	for (auto Ch : Name) {
	std::string Label(1, Ch);
	auto It = llvm::find_if(N->Children,
	[&](const auto &C) { return C->Name == Label; });
	if (It == N->Children.end()) {
	It = N->Children.insert(It, std::make_unique<Node>(Label, N));
	}
	N = It->get();
	}
	N->Value = Codepoint;
	}

	void compact() { compact(Root.get()); }

	// This creates 2 arrays of bytes from the tree:
	// A serialized dictionary of node labels,
	// And the nodes themselves.
	// The name of each label is found by indexing into the dictionary.
	// The longest names are inserted first into the dictionary,
	// in the hope it will contain shorter labels as substring,
	// thereby reducing duplication.
	// We could theorically be more clever by trying to minimizing the size
	// of the dictionary.
	std::pair<std::string, std::vector<uint8_t>> serialize() {
	std::set<std::string> Names = this->getNameFragments();
	std::vector<std::string> Sorted(Names.begin(), Names.end());
	llvm::sort(Sorted, [](const auto &a, const auto &b) {
	return a.size() > b.size();
	});
	std::string Dict(Letters.begin(), Letters.end());
	Dict.reserve(50000);
	for (const std::string &Name : Sorted) {
	if (Name.size() <= 1)
	continue;
	if (Dict.find(Name) != std::string::npos)
	continue;
	Dict += Name;
	}

	if (Dict.size() >= std::numeric_limits<uint16_t>::max()) {
	fprintf(stderr, "Dictionary too big to be serialized");
	exit(1);
	}

	auto Bytes = dumpIndex(Dict);
	return {Dict, Bytes};
	}

	std::set<std::string> getNameFragments() {
	std::set<std::string> Keys;
	collectKeys(Root.get(), Keys);
	return Keys;
	}

	// Maps a valid char in an Unicode character name
	// To a 6 bits index.
	static uint8_t letter(char C) {
	auto Pos = Letters.find(C);
	assert(Pos != std::string::npos &&
	"Invalid letter in Unicode character name");
	return Pos;
	}

	// clang-format off
	// +================+============+======================+=============+========+===+==============+===============+
	// \| 0 \| 1 \| 2-7 (6) \| 8-23 \| 24-44 \| \| 46 \| 47 \|
	// +================+============+======================+=============+========+===+==============+===============+
	// \| Has Value \| Has Long Name \| Letter OR Name Size \| Dict Index \| Value \| \| Has Sibling \| Has Children \|
	// +----------------+------------+----------------------+-------------+--------+---+--------------+---------------+
	// clang-format on

	std::vector<uint8_t> dumpIndex(const std::string &Dict) {
	struct ChildrenOffset {
	Node *FirstChild;
	std::size_t Offset;
	bool HasValue;
	};

	// Keep track of the start of each node
	// position in the serialized data.
	std::unordered_map<Node *, int32_t> Offsets;

	// Keep track of where to write the index
	// of the first children
	std::vector<ChildrenOffset> ChildrenOffsets;
	std::unordered_map<Node *, bool> SiblingTracker;
	std::deque<Node *> AllNodes;
	std::vector<uint8_t> Bytes;
	Bytes.reserve(250'000);
	// This leading byte is used by the reading code to detect the root node.
	Bytes.push_back(0);

	auto CollectChildren = [&SiblingTracker, &AllNodes](const auto &Children) {
	for (std::size_t Index = 0; Index < Children.size(); Index++) {
	const std::unique_ptr<Node> &Child = Children[Index];
	AllNodes.push_back(Child.get());
	if (Index != Children.size() - 1)
	SiblingTracker[Child.get()] = true;
	}
	};
	CollectChildren(Root->Children);

	while (!AllNodes.empty()) {
	const std::size_t Offset = Bytes.size();
	Node *const N = AllNodes.front();
	AllNodes.pop_front();

	assert(!N->Name.empty());
	Offsets[N] = Offset;

	uint8_t FirstByte = (!!N->Value) ? 0x80 : 0;
	// Single letter node are indexed in 6 bits
	if (N->Name.size() == 1) {
	FirstByte \|= letter(N->Name[0]);
	Bytes.push_back(FirstByte);
	} else {
	// Otherwise we use a 16 bits index
	FirstByte = FirstByte \| uint8_t(N->Name.size()) \| 0x40;
	Bytes.push_back(FirstByte);
	auto PosInDict = Dict.find(N->Name);
	assert(PosInDict != std::string::npos);
	uint8_t Low = PosInDict;
	uint8_t High = ((PosInDict >> 8) & 0xFF);
	Bytes.push_back(High);
	Bytes.push_back(Low);
	}

	const bool HasSibling = SiblingTracker.count(N) != 0;
	const bool HasChildren = N->Children.size() != 0;

	if (!!N->Value) {
	uint32_t Value = (*(N->Value) << 3);
	uint8_t H = ((Value >> 16) & 0xFF);
	uint8_t M = ((Value >> 8) & 0xFF);
	uint8_t L = (Value & 0xFF) \| uint8_t(HasSibling ? 0x01 : 0) \|
	uint8_t(HasChildren ? 0x02 : 0);

	Bytes.push_back(H);
	Bytes.push_back(M);
	Bytes.push_back(L);

	if (HasChildren) {
	ChildrenOffsets.push_back(
	ChildrenOffset{N->Children[0].get(), Bytes.size(), true});
	// index of the first children
	Bytes.push_back(0x00);
	Bytes.push_back(0x00);
	Bytes.push_back(0x00);
	}
	} else {
	// When there is no value (that's most intermediate nodes)
	// Dispense of the 3 values bytes, and only store
	// 1 byte to track whether the node has sibling and children
	// + 2 bytes for the index of the first children if necessary.
	// That index also uses bytes 0-6 of the previous byte.
	uint8_t Byte =
	uint8_t(HasSibling ? 0x80 : 0) \| uint8_t(HasChildren ? 0x40 : 0);
	Bytes.push_back(Byte);
	if (HasChildren) {
	ChildrenOffsets.emplace_back(
	ChildrenOffset{N->Children[0].get(), Bytes.size() - 1, false});
	Bytes.push_back(0x00);
	Bytes.push_back(0x00);
	}
	}
	CollectChildren(N->Children);
	}

	// Once all the nodes are in the inndex
	// Fill the bytes we left to indicate the position
	// of the children
	for (const ChildrenOffset &Parent : ChildrenOffsets) {
	const auto It = Offsets.find(Parent.FirstChild);
	assert(It != Offsets.end());
	std::size_t Pos = It->second;
	if (Parent.HasValue) {
	Bytes[Parent.Offset] = ((Pos >> 16) & 0xFF);
	} else {
	Bytes[Parent.Offset] =
	Bytes[Parent.Offset] \| uint8_t((Pos >> 16) & 0xFF);
	}
	Bytes[Parent.Offset + 1] = ((Pos >> 8) & 0xFF);
	Bytes[Parent.Offset + 2] = Pos & 0xFF;
	}

	// Add some padding so that the deserialization code
	// doesn't try to read past the enf of the array.
	Bytes.push_back(0);
	Bytes.push_back(0);
	Bytes.push_back(0);
	Bytes.push_back(0);
	Bytes.push_back(0);
	Bytes.push_back(0);

	return Bytes;
	}

	private:
	void collectKeys(Node *N, std::set<std::string> &Keys) {
	Keys.insert(N->Name);
	for (const std::unique_ptr<Node> &Child : N->Children) {
	collectKeys(Child.get(), Keys);
	}
	}

	// Merge sequences of 1-character nodes
	// This greatly reduce the total number of nodes,
	// and therefore the size of the index.
	// When the tree gets serialized, we only have 5 bytes to store the
	// size of a name. Overlong names (>32 characters) are therefore
	// kep into separate nodes
	void compact(Node *N) {
	for (auto &&Child : N->Children) {
	compact(Child.get());
	}
	if (N->Parent && N->Parent->Children.size() == 1 && !N->Parent->Value &&
	(N->Parent->Name.size() + N->Name.size() <= 32)) {
	N->Parent->Value = N->Value;
	N->Parent->Name += N->Name;
	N->Parent->Children = std::move(N->Children);
	for (std::unique_ptr<Node> &c : N->Parent->Children) {
	c->Parent = N->Parent;
	}
	}
	}
	struct Node {
	Node(std::string Name, Node *Parent = nullptr)
	: Name(Name), Parent(Parent) {}

	std::vector<std::unique_ptr<Node>> Children;
	std::string Name;
	Node *Parent = nullptr;
	std::optional<char32_t> Value;
	};

	std::unique_ptr<Node> Root = std::make_unique<Node>("");
	};

	extern const char *UnicodeLicense;

	int main(int argc, char **argv) {
	printf("Unicode name -> codepoint mapping generator\n"
	"Usage: %s UnicodeData.txt NameAliases.txt output\n\n",
	argv[0]);
	printf("NameAliases.txt can be found at "
	"https://unicode.org/Public/15.0.0/ucd/NameAliases.txt\n"
	"UnicodeData.txt can be found at "
	"https://unicode.org/Public/15.0.0/ucd/UnicodeData.txt\n\n");

	if (argc != 4)
	return EXIT_FAILURE;

	FILE *Out = fopen(argv[3], "w");
	if (!Out) {
	printf("Error creating output file.\n");
	return EXIT_FAILURE;
	}

	Trie T;
	uint32_t NameCount = 0;
	std::size_t LongestName = 0;
	auto Entries = loadDataFiles(argv[1], argv[2]);
	for (const std::pair<const char32_t, std::string> &Entry : Entries) {
	char32_t Codepoint = Entry.first;
	const std::string &Name = Entry.second;
	// Ignore names which are not valid.
	if (Name.empty() \|\|
	!llvm::all_of(Name, [](char C) { return Letters.contains(C); })) {
	continue;
	}
	printf("%06x: %s\n", static_cast<unsigned int>(Codepoint), Name.c_str());
	T.insert(Name, Codepoint);
	LongestName =
	std::max(LongestName, std::size_t(llvm::count_if(Name, llvm::isAlnum)));
	NameCount++;
	}
	T.compact();

	std::pair<std::string, std::vector<uint8_t>> Data = T.serialize();
	const std::string &Dict = Data.first;
	const std::vector<uint8_t> &Tree = Data.second;

	fprintf(Out, R"(
	//===------------- Support/UnicodeNameToCodepointGenerated.cpp ------------===//
	// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
	// See https://llvm.org/LICENSE.txt for license information.
	// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
	//
	//===----------------------------------------------------------------------===//
	//
	// This file implements mapping the name of a unicode code point to its value.
	//
	// This file was generated using %s.
	// Do not edit manually.
	//
	//===----------------------------------------------------------------------===//
	%s



	#include "llvm/Support/Compiler.h"
	#include <cstddef>
	#include <cstdint>
	)",
	argv[0], UnicodeLicense);

	fprintf(Out,
	"namespace llvm { namespace sys { namespace unicode { \n"
	"extern const char *UnicodeNameToCodepointDict;\n"
	"extern const uint8_t *UnicodeNameToCodepointIndex;\n"
	"extern const std::size_t UnicodeNameToCodepointIndexSize;\n"
	"extern const std::size_t UnicodeNameToCodepointLargestNameSize;\n");

	fprintf(Out, "const char* UnicodeNameToCodepointDict = \"%s\";\n",
	Dict.c_str());

	fprintf(Out, "uint8_t UnicodeNameToCodepointIndex_[%zu] = {\n",
	Tree.size() + 1);

	for (auto Byte : Tree) {
	fprintf(Out, "0x%02x,", Byte);
	}

	fprintf(Out, "0};");
	fprintf(Out, "const uint8_t* UnicodeNameToCodepointIndex = "
	"UnicodeNameToCodepointIndex_; \n");
	fprintf(Out, "const std::size_t UnicodeNameToCodepointIndexSize = %zu;\n",
	Tree.size() + 1);
	fprintf(Out,
	"const std::size_t UnicodeNameToCodepointLargestNameSize = %zu;\n",
	LongestName);
	fprintf(Out, "\n}}}\n");
	fclose(Out);
	printf("Generated %s: %u Files.\nIndex: %f kB, Dictionary: %f kB.\nDone\n\n",
	argv[3], NameCount, Tree.size() / 1024.0, Dict.size() / 1024.0);
	}

	const char *UnicodeLicense = R"(
	/*
	UNICODE, INC. LICENSE AGREEMENT - DATA FILES AND SOFTWARE

	See Terms of Use <https://www.unicode.org/copyright.html>
	for definitions of Unicode Inc.’s Data Files and Software.

	NOTICE TO USER: Carefully read the following legal agreement.
	BY DOWNLOADING, INSTALLING, COPYING OR OTHERWISE USING UNICODE INC.'S
	DATA FILES ("DATA FILES"), AND/OR SOFTWARE ("SOFTWARE"),
	YOU UNEQUIVOCALLY ACCEPT, AND AGREE TO BE BOUND BY, ALL OF THE
	TERMS AND CONDITIONS OF THIS AGREEMENT.
	IF YOU DO NOT AGREE, DO NOT DOWNLOAD, INSTALL, COPY, DISTRIBUTE OR USE
	THE DATA FILES OR SOFTWARE.

	COPYRIGHT AND PERMISSION NOTICE

	Copyright © 1991-2022 Unicode, Inc. All rights reserved.
	Distributed under the Terms of Use in https://www.unicode.org/copyright.html.

	Permission is hereby granted, free of charge, to any person obtaining
	a copy of the Unicode data files and any associated documentation
	(the "Data Files") or Unicode software and any associated documentation
	(the "Software") to deal in the Data Files or Software
	without restriction, including without limitation the rights to use,
	copy, modify, merge, publish, distribute, and/or sell copies of
	the Data Files or Software, and to permit persons to whom the Data Files
	or Software are furnished to do so, provided that either
	(a) this copyright and permission notice appear with all copies
	of the Data Files or Software, or
	(b) this copyright and permission notice appear in associated
	Documentation.

	THE DATA FILES AND SOFTWARE ARE PROVIDED "AS IS", WITHOUT WARRANTY OF
	ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE
	WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
	NONINFRINGEMENT OF THIRD PARTY RIGHTS.
	IN NO EVENT SHALL THE COPYRIGHT HOLDER OR HOLDERS INCLUDED IN THIS
	NOTICE BE LIABLE FOR ANY CLAIM, OR ANY SPECIAL INDIRECT OR CONSEQUENTIAL
	DAMAGES, OR ANY DAMAGES WHATSOEVER RESULTING FROM LOSS OF USE,
	DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE OR OTHER
	TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR
	PERFORMANCE OF THE DATA FILES OR SOFTWARE.

	Except as contained in this notice, the name of a copyright holder
	shall not be used in advertising or otherwise to promote the sale,
	use or other dealings in these Data Files or Software without prior
	written authorization of the copyright holder.
	*/
	)";