git.sesse.net Git - plocate/blob - plocate-build.cpp

   1 #include "db.h"
   2 #include "vp4.h"
   3
   4 #include <algorithm>
   5 #include <arpa/inet.h>
   6 #include <assert.h>
   7 #include <chrono>
   8 #include <endian.h>
   9 #include <fcntl.h>
  10 #include <math.h>
  11 #include <memory>
  12 #include <stdio.h>
  13 #include <string.h>
  14 #include <string>
  15 #include <sys/stat.h>
  16 #include <sys/types.h>
  17 #include <unistd.h>
  18 #include <unordered_map>
  19 #include <vector>
  20 #include <zstd.h>
  21
  22 #define P4NENC_BOUND(n) ((n + 127) / 128 + (n + 32) * sizeof(uint32_t))
  23 #define dprintf(...)
  24 //#define dprintf(...) fprintf(stderr, __VA_ARGS__);
  25
  26 using namespace std;
  27 using namespace std::chrono;
  28
  29 string zstd_compress(const string &src, string *tempbuf);
  30
  31 constexpr unsigned num_overflow_slots = 16;
  32
  33 static inline uint32_t read_unigram(const string_view s, size_t idx)
  34 {
  35         if (idx < s.size()) {
  36                 return (unsigned char)s[idx];
  37         } else {
  38                 return 0;
  39         }
  40 }
  41
  42 static inline uint32_t read_trigram(const string_view s, size_t start)
  43 {
  44         return read_unigram(s, start) |
  45                 (read_unigram(s, start + 1) << 8) |
  46                 (read_unigram(s, start + 2) << 16);
  47 }
  48
  49 enum {
  50         DBE_NORMAL = 0, /* A non-directory file */
  51         DBE_DIRECTORY = 1, /* A directory */
  52         DBE_END = 2 /* End of directory contents; contains no name */
  53 };
  54
  55 // From mlocate.
  56 struct db_header {
  57         uint8_t magic[8];
  58         uint32_t conf_size;
  59         uint8_t version;
  60         uint8_t check_visibility;
  61         uint8_t pad[2];
  62 };
  63
  64 // From mlocate.
  65 struct db_directory {
  66         uint64_t time_sec;
  67         uint32_t time_nsec;
  68         uint8_t pad[4];
  69 };
  70
  71 class PostingListBuilder {
  72 public:
  73         void add_docid(uint32_t docid);
  74         void finish();
  75
  76         string encoded;
  77         size_t num_docids = 0;
  78
  79 private:
  80         void write_header(uint32_t docid);
  81         void append_block();
  82
  83         vector<uint32_t> pending_docids;
  84
  85         uint32_t last_block_end, last_docid = -1;
  86 };
  87
  88 void PostingListBuilder::add_docid(uint32_t docid)
  89 {
  90         // Deduplicate against the last inserted value, if any.
  91         if (docid == last_docid) {
  92                 return;
  93         }
  94
  95         if (num_docids == 0) {
  96                 // Very first docid.
  97                 write_header(docid);
  98                 ++num_docids;
  99                 last_block_end = last_docid = docid;
 100                 return;
 101         }
 102
 103         last_docid = docid;
 104         pending_docids.push_back(docid);
 105         if (pending_docids.size() == 128) {
 106                 append_block();
 107                 pending_docids.clear();
 108                 last_block_end = docid;
 109         }
 110         ++num_docids;
 111 }
 112
 113 void PostingListBuilder::finish()
 114 {
 115         if (pending_docids.empty()) {
 116                 return;
 117         }
 118
 119         assert(!encoded.empty());  // write_header() should already have run.
 120
 121         // No interleaving for partial blocks.
 122         unsigned char buf[P4NENC_BOUND(128)];
 123         unsigned char *end = p4d1enc32(pending_docids.data(), pending_docids.size(), buf, last_block_end);
 124         encoded.append(reinterpret_cast<char *>(buf), reinterpret_cast<char *>(end));
 125 }
 126
 127 void PostingListBuilder::append_block()
 128 {
 129         unsigned char buf[P4NENC_BOUND(128)];
 130         assert(pending_docids.size() == 128);
 131         unsigned char *end = p4d1enc128v32(pending_docids.data(), 128, buf, last_block_end);
 132         encoded.append(reinterpret_cast<char *>(buf), reinterpret_cast<char *>(end));
 133 }
 134
 135 void PostingListBuilder::write_header(uint32_t docid)
 136 {
 137         unsigned char buf[P4NENC_BOUND(1)];
 138         size_t bytes = p4nd1enc128v32(&docid, 1, buf);
 139         encoded.append(reinterpret_cast<char *>(buf), bytes);
 140 }
 141
 142 class Corpus {
 143 public:
 144         Corpus(FILE *outfp, size_t block_size)
 145                 : outfp(outfp), block_size(block_size) {}
 146         void add_file(string filename);
 147         void flush_block();
 148
 149         vector<uint64_t> filename_blocks;
 150         unordered_map<uint32_t, PostingListBuilder> invindex;
 151         size_t num_files = 0, num_files_in_block = 0, num_blocks = 0;
 152
 153 private:
 154         FILE *outfp;
 155         string current_block;
 156         string tempbuf;
 157         const size_t block_size;
 158 };
 159
 160 void Corpus::add_file(string filename)
 161 {
 162         ++num_files;
 163         if (!current_block.empty()) {
 164                 current_block.push_back('\0');
 165         }
 166         current_block += filename;
 167         if (++num_files_in_block == block_size) {
 168                 flush_block();
 169         }
 170 }
 171
 172 void Corpus::flush_block()
 173 {
 174         if (current_block.empty()) {
 175                 return;
 176         }
 177
 178         uint32_t docid = num_blocks;
 179
 180         // Create trigrams.
 181         const char *ptr = current_block.c_str();
 182         while (ptr < current_block.c_str() + current_block.size()) {
 183                 string_view s(ptr);
 184                 if (s.size() >= 3) {
 185                         for (size_t j = 0; j < s.size() - 2; ++j) {
 186                                 uint32_t trgm = read_trigram(s, j);
 187                                 invindex[trgm].add_docid(docid);
 188                         }
 189                 }
 190                 ptr += s.size() + 1;
 191         }
 192
 193         // Compress and add the filename block.
 194         filename_blocks.push_back(ftell(outfp));
 195         string compressed = zstd_compress(current_block, &tempbuf);
 196         if (fwrite(compressed.data(), compressed.size(), 1, outfp) != 1) {
 197                 perror("fwrite()");
 198                 exit(1);
 199         }
 200
 201         current_block.clear();
 202         num_files_in_block = 0;
 203         ++num_blocks;
 204 }
 205
 206 string read_cstr(FILE *fp)
 207 {
 208         string ret;
 209         for ( ;; ) {
 210                 int ch = getc(fp);
 211                 if (ch == -1) {
 212                         perror("getc");
 213                         exit(1);
 214                 }
 215                 if (ch == 0) {
 216                         return ret;
 217                 }
 218                 ret.push_back(ch);
 219         }
 220 }
 221
 222 void handle_directory(FILE *fp, Corpus *corpus)
 223 {
 224         db_directory dummy;
 225         if (fread(&dummy, sizeof(dummy), 1, fp) != 1) {
 226                 if (feof(fp)) {
 227                         return;
 228                 } else {
 229                         perror("fread");
 230                 }
 231         }
 232
 233         string dir_path = read_cstr(fp);
 234         if (dir_path == "/") {
 235                 dir_path = "";
 236         }
 237
 238         for (;;) {
 239                 int type = getc(fp);
 240                 if (type == DBE_NORMAL) {
 241                         string filename = read_cstr(fp);
 242                         corpus->add_file(dir_path + "/" + filename);
 243                 } else if (type == DBE_DIRECTORY) {
 244                         string dirname = read_cstr(fp);
 245                         corpus->add_file(dir_path + "/" + dirname);
 246                 } else {
 247                         return;  // Probably end.
 248                 }
 249         }
 250 }
 251
 252 void read_mlocate(const char *filename, Corpus *corpus)
 253 {
 254         FILE *fp = fopen(filename, "rb");
 255         if (fp == nullptr) {
 256                 perror(filename);
 257                 exit(1);
 258         }
 259
 260         db_header hdr;
 261         if (fread(&hdr, sizeof(hdr), 1, fp) != 1) {
 262                 perror("short read");
 263                 exit(1);
 264         }
 265
 266         // TODO: Care about the base path.
 267         string path = read_cstr(fp);
 268         while (!feof(fp)) {
 269                 handle_directory(fp, corpus);
 270         }
 271         fclose(fp);
 272 }
 273
 274 string zstd_compress(const string &src, string *tempbuf)
 275 {
 276         size_t max_size = ZSTD_compressBound(src.size());
 277         if (tempbuf->size() < max_size) {
 278                 tempbuf->resize(max_size);
 279         }
 280         size_t size = ZSTD_compress(&(*tempbuf)[0], max_size, src.data(), src.size(), /*level=*/6);
 281         return string(tempbuf->data(), size);
 282 }
 283
 284 bool is_prime(uint32_t x)
 285 {
 286         if ((x % 2) == 0 || (x % 3) == 0) {
 287                 return false;
 288         }
 289         uint32_t limit = ceil(sqrt(x));
 290         for (uint32_t factor = 5; factor <= limit; ++factor) {
 291                 if ((x % factor) == 0) {
 292                         return false;
 293                 }
 294         }
 295         return true;
 296 }
 297
 298 uint32_t next_prime(uint32_t x)
 299 {
 300         if ((x % 2) == 0) {
 301                 ++x;
 302         }
 303         while (!is_prime(x)) {
 304                 x += 2;
 305         }
 306         return x;
 307 }
 308
 309 unique_ptr<Trigram[]> create_hashtable(const Corpus &corpus, const vector<uint32_t> &all_trigrams, uint32_t ht_size, uint32_t num_overflow_slots)
 310 {
 311         unique_ptr<Trigram[]> ht(new Trigram[ht_size + num_overflow_slots + 1]);  // 1 for the sentinel element at the end.
 312         for (unsigned i = 0; i < ht_size + num_overflow_slots + 1; ++i) {
 313                 ht[i].trgm = uint32_t(-1);
 314                 ht[i].num_docids = 0;
 315                 ht[i].offset = 0;
 316         }
 317         for (uint32_t trgm : all_trigrams) {
 318                 // We don't know offset yet, so set it to zero.
 319                 Trigram to_insert{ trgm, uint32_t(corpus.invindex.find(trgm)->second.num_docids), 0 };
 320
 321                 uint32_t bucket = hash_trigram(trgm, ht_size);
 322                 unsigned distance = 0;
 323                 while (ht[bucket].num_docids != 0) {
 324                         // Robin Hood hashing; reduces the longest distance by a lot.
 325                         unsigned other_distance = bucket - hash_trigram(ht[bucket].trgm, ht_size);
 326                         if (distance > other_distance) {
 327                                 swap(to_insert, ht[bucket]);
 328                                 distance = other_distance;
 329                         }
 330
 331                         ++bucket, ++distance;
 332                         if (distance > num_overflow_slots) {
 333                                 return nullptr;
 334                         }
 335                 }
 336                 ht[bucket] = to_insert;
 337         }
 338         return ht;
 339 }
 340
 341 void do_build(const char *infile, const char *outfile, int block_size)
 342 {
 343         steady_clock::time_point start __attribute__((unused)) = steady_clock::now();
 344
 345         umask(0027);
 346         FILE *outfp = fopen(outfile, "wb");
 347
 348         // Write the header.
 349         Header hdr;
 350         memcpy(hdr.magic, "\0plocate", 8);
 351         hdr.version = -1;  // Mark as broken.
 352         hdr.hashtable_size = 0;  // Not known yet.
 353         hdr.extra_ht_slots = num_overflow_slots;
 354         hdr.hash_table_offset_bytes = -1;  // We don't know these offsets yet.
 355         hdr.filename_index_offset_bytes = -1;
 356         fwrite(&hdr, sizeof(hdr), 1, outfp);
 357
 358         Corpus corpus(outfp, block_size);
 359
 360         read_mlocate(infile, &corpus);
 361         if (false) {  // To read a plain text file.
 362                 FILE *fp = fopen(infile, "r");
 363                 while (!feof(fp)) {
 364                         char buf[1024];
 365                         if (fgets(buf, 1024, fp) == nullptr || feof(fp)) {
 366                                 break;
 367                         }
 368                         string s(buf);
 369                         if (s.back() == '\n')
 370                                 s.pop_back();
 371                         corpus.add_file(move(s));
 372                 }
 373                 fclose(fp);
 374         }
 375         corpus.flush_block();
 376         dprintf("Read %zu files from %s\n", corpus.num_files, infile);
 377         hdr.num_docids = corpus.filename_blocks.size();
 378
 379         // Stick an empty block at the end as sentinel.
 380         corpus.filename_blocks.push_back(ftell(outfp));
 381         const size_t bytes_for_filenames = corpus.filename_blocks.back() - corpus.filename_blocks.front();
 382
 383         // Write the offsets to the filenames.
 384         hdr.filename_index_offset_bytes = ftell(outfp);
 385         const size_t bytes_for_filename_index = corpus.filename_blocks.size() * sizeof(uint64_t);
 386         fwrite(corpus.filename_blocks.data(), corpus.filename_blocks.size(), sizeof(uint64_t), outfp);
 387         corpus.filename_blocks.clear();
 388         corpus.filename_blocks.shrink_to_fit();
 389
 390         // Finish up encoding the posting lists.
 391         size_t trigrams = 0, longest_posting_list = 0;
 392         size_t bytes_for_posting_lists = 0;
 393         for (auto &[trigram, pl_builder] : corpus.invindex) {
 394                 pl_builder.finish();
 395                 longest_posting_list = max(longest_posting_list, pl_builder.num_docids);
 396                 trigrams += pl_builder.num_docids;
 397                 bytes_for_posting_lists += pl_builder.encoded.size();
 398         }
 399         dprintf("%zu files, %zu different trigrams, %zu entries, avg len %.2f, longest %zu\n",
 400                 corpus.num_files, corpus.invindex.size(), trigrams, double(trigrams) / corpus.invindex.size(), longest_posting_list);
 401         dprintf("%zu bytes used for posting lists (%.2f bits/entry)\n", bytes_for_posting_lists, 8 * bytes_for_posting_lists / double(trigrams));
 402
 403         dprintf("Building posting lists took %.1f ms.\n\n", 1e3 * duration<float>(steady_clock::now() - start).count());
 404
 405         // Sort the trigrams, mostly to get a consistent result every time
 406         // (the hash table will put things in random order anyway).
 407         vector<uint32_t> all_trigrams;
 408         for (auto &[trigram, pl_builder] : corpus.invindex) {
 409                 all_trigrams.push_back(trigram);
 410         }
 411         sort(all_trigrams.begin(), all_trigrams.end());
 412
 413         // Create the hash table.
 414         unique_ptr<Trigram[]> hashtable;
 415         uint32_t ht_size = next_prime(all_trigrams.size());
 416         for (;;) {
 417                 hashtable = create_hashtable(corpus, all_trigrams, ht_size, num_overflow_slots);
 418                 if (hashtable == nullptr) {
 419                         dprintf("Failed creating hash table of size %u, increasing by 5%% and trying again.\n", ht_size);
 420                         ht_size = next_prime(ht_size * 1.05);
 421                 } else {
 422                         dprintf("Created hash table of size %u.\n\n", ht_size);
 423                         break;
 424                 }
 425         }
 426
 427         // Find the offsets for each posting list.
 428         size_t bytes_for_hashtable = (ht_size + num_overflow_slots + 1) * sizeof(Trigram);
 429         uint64_t offset = ftell(outfp) + bytes_for_hashtable;
 430         for (unsigned i = 0; i < ht_size + num_overflow_slots + 1; ++i) {
 431                 hashtable[i].offset = offset;  // Needs to be there even for empty slots.
 432                 if (hashtable[i].num_docids == 0) {
 433                         continue;
 434                 }
 435
 436                 const string &encoded = corpus.invindex[hashtable[i].trgm].encoded;
 437                 offset += encoded.size();
 438         }
 439
 440         // Write the hash table.
 441         hdr.hash_table_offset_bytes = ftell(outfp);
 442         hdr.hashtable_size = ht_size;
 443         fwrite(hashtable.get(), ht_size + num_overflow_slots + 1, sizeof(Trigram), outfp);
 444
 445         // Write the actual posting lists.
 446         for (unsigned i = 0; i < ht_size + num_overflow_slots + 1; ++i) {
 447                 if (hashtable[i].num_docids == 0) {
 448                         continue;
 449                 }
 450                 const string &encoded = corpus.invindex[hashtable[i].trgm].encoded;
 451                 fwrite(encoded.data(), encoded.size(), 1, outfp);
 452         }
 453
 454         // Rewind, and write the updated header.
 455         hdr.version = 0;
 456         fseek(outfp, 0, SEEK_SET);
 457         fwrite(&hdr, sizeof(hdr), 1, outfp);
 458         fclose(outfp);
 459
 460         size_t total_bytes __attribute__((unused)) = (bytes_for_hashtable + bytes_for_posting_lists + bytes_for_filename_index + bytes_for_filenames);
 461
 462         dprintf("Block size:     %7d files\n", block_size);
 463         dprintf("Hash table:     %'7.1f MB\n", bytes_for_hashtable / 1048576.0);
 464         dprintf("Posting lists:  %'7.1f MB\n", bytes_for_posting_lists / 1048576.0);
 465         dprintf("Filename index: %'7.1f MB\n", bytes_for_filename_index / 1048576.0);
 466         dprintf("Filenames:      %'7.1f MB\n", bytes_for_filenames / 1048576.0);
 467         dprintf("Total:          %'7.1f MB\n", total_bytes / 1048576.0);
 468         dprintf("\n");
 469 }
 470
 471 int main(int argc, char **argv)
 472 {
 473         do_build(argv[1], argv[2], 32);
 474         exit(EXIT_SUCCESS);
 475 }