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

   1 #include "db.h"
   2 #include "turbopfor-encode.h"
   3
   4 #include <algorithm>
   5 #include <assert.h>
   6 #include <chrono>
   7 #include <getopt.h>
   8 #include <iosfwd>
   9 #include <math.h>
  10 #include <memory>
  11 #include <random>
  12 #include <stdint.h>
  13 #include <stdio.h>
  14 #include <stdlib.h>
  15 #include <string.h>
  16 #include <string>
  17 #include <string_view>
  18 #include <sys/stat.h>
  19 #include <utility>
  20 #include <vector>
  21 #include <zdict.h>
  22 #include <zstd.h>
  23
  24 #define P4NENC_BOUND(n) ((n + 127) / 128 + (n + 32) * sizeof(uint32_t))
  25 #define dprintf(...)
  26 //#define dprintf(...) fprintf(stderr, __VA_ARGS__);
  27
  28 #define NUM_TRIGRAMS 16777216
  29
  30 using namespace std;
  31 using namespace std::chrono;
  32
  33 string zstd_compress(const string &src, ZSTD_CDict *cdict, string *tempbuf);
  34
  35 constexpr unsigned num_overflow_slots = 16;
  36
  37 static inline uint32_t read_unigram(const string_view s, size_t idx)
  38 {
  39         if (idx < s.size()) {
  40                 return (unsigned char)s[idx];
  41         } else {
  42                 return 0;
  43         }
  44 }
  45
  46 static inline uint32_t read_trigram(const string_view s, size_t start)
  47 {
  48         return read_unigram(s, start) |
  49                 (read_unigram(s, start + 1) << 8) |
  50                 (read_unigram(s, start + 2) << 16);
  51 }
  52
  53 enum {
  54         DBE_NORMAL = 0, /* A non-directory file */
  55         DBE_DIRECTORY = 1, /* A directory */
  56         DBE_END = 2 /* End of directory contents; contains no name */
  57 };
  58
  59 // From mlocate.
  60 struct db_header {
  61         uint8_t magic[8];
  62         uint32_t conf_size;
  63         uint8_t version;
  64         uint8_t check_visibility;
  65         uint8_t pad[2];
  66 };
  67
  68 // From mlocate.
  69 struct db_directory {
  70         uint64_t time_sec;
  71         uint32_t time_nsec;
  72         uint8_t pad[4];
  73 };
  74
  75 class PostingListBuilder {
  76 public:
  77         inline void add_docid(uint32_t docid);
  78         void finish();
  79
  80         string encoded;
  81         size_t num_docids = 0;
  82
  83 private:
  84         void write_header(uint32_t docid);
  85         void append_block();
  86
  87         vector<uint32_t> pending_deltas;
  88
  89         uint32_t last_block_end, last_docid = -1;
  90 };
  91
  92 void PostingListBuilder::add_docid(uint32_t docid)
  93 {
  94         // Deduplicate against the last inserted value, if any.
  95         if (docid == last_docid) {
  96                 return;
  97         }
  98
  99         if (num_docids == 0) {
 100                 // Very first docid.
 101                 write_header(docid);
 102                 ++num_docids;
 103                 last_block_end = last_docid = docid;
 104                 return;
 105         }
 106
 107         pending_deltas.push_back(docid - last_docid - 1);
 108         last_docid = docid;
 109         if (pending_deltas.size() == 128) {
 110                 append_block();
 111                 pending_deltas.clear();
 112                 last_block_end = docid;
 113         }
 114         ++num_docids;
 115 }
 116
 117 void PostingListBuilder::finish()
 118 {
 119         if (pending_deltas.empty()) {
 120                 return;
 121         }
 122
 123         assert(!encoded.empty());  // write_header() should already have run.
 124
 125         // No interleaving for partial blocks.
 126         unsigned char buf[P4NENC_BOUND(128)];
 127         unsigned char *end = encode_pfor_single_block<128>(pending_deltas.data(), pending_deltas.size(), /*interleaved=*/false, buf);
 128         encoded.append(reinterpret_cast<char *>(buf), reinterpret_cast<char *>(end));
 129 }
 130
 131 void PostingListBuilder::append_block()
 132 {
 133         unsigned char buf[P4NENC_BOUND(128)];
 134         assert(pending_deltas.size() == 128);
 135         unsigned char *end = encode_pfor_single_block<128>(pending_deltas.data(), 128, /*interleaved=*/true, buf);
 136         encoded.append(reinterpret_cast<char *>(buf), reinterpret_cast<char *>(end));
 137 }
 138
 139 void PostingListBuilder::write_header(uint32_t docid)
 140 {
 141         unsigned char buf[P4NENC_BOUND(1)];
 142         unsigned char *end = write_baseval(docid, buf);
 143         encoded.append(reinterpret_cast<char *>(buf), end - buf);
 144 }
 145
 146 class DatabaseReceiver {
 147 public:
 148         virtual ~DatabaseReceiver() = default;
 149         virtual void add_file(string filename) = 0;
 150         virtual void flush_block() = 0;
 151 };
 152
 153 class DictionaryBuilder : public DatabaseReceiver {
 154 public:
 155         DictionaryBuilder(size_t blocks_to_keep, size_t block_size)
 156                 : blocks_to_keep(blocks_to_keep), block_size(block_size) {}
 157         void add_file(string filename) override;
 158         void flush_block() override;
 159         string train(size_t buf_size);
 160
 161 private:
 162         const size_t blocks_to_keep, block_size;
 163         string current_block;
 164         uint64_t block_num = 0;
 165         size_t num_files_in_block = 0;
 166
 167         std::mt19937 reservoir_rand{ 1234 };  // Fixed seed for reproducibility.
 168         bool keep_current_block = true;
 169         int64_t slot_for_current_block = -1;
 170
 171         vector<string> sampled_blocks;
 172         vector<size_t> lengths;
 173 };
 174
 175 void DictionaryBuilder::add_file(string filename)
 176 {
 177         if (keep_current_block) {  // Only bother saving the filenames if we're actually keeping the block.
 178                 if (!current_block.empty()) {
 179                         current_block.push_back('\0');
 180                 }
 181                 current_block += filename;
 182         }
 183         if (++num_files_in_block == block_size) {
 184                 flush_block();
 185         }
 186 }
 187
 188 void DictionaryBuilder::flush_block()
 189 {
 190         if (keep_current_block) {
 191                 if (slot_for_current_block == -1) {
 192                         lengths.push_back(current_block.size());
 193                         sampled_blocks.push_back(move(current_block));
 194                 } else {
 195                         lengths[slot_for_current_block] = current_block.size();
 196                         sampled_blocks[slot_for_current_block] = move(current_block);
 197                 }
 198         }
 199         current_block.clear();
 200         num_files_in_block = 0;
 201         ++block_num;
 202
 203         if (block_num < blocks_to_keep) {
 204                 keep_current_block = true;
 205                 slot_for_current_block = -1;
 206         } else {
 207                 // Keep every block with equal probability (reservoir sampling).
 208                 uint64_t idx = uniform_int_distribution<uint64_t>(0, block_num)(reservoir_rand);
 209                 keep_current_block = (idx < blocks_to_keep);
 210                 slot_for_current_block = idx;
 211         }
 212 }
 213
 214 string DictionaryBuilder::train(size_t buf_size)
 215 {
 216         string dictionary_buf;
 217         sort(sampled_blocks.begin(), sampled_blocks.end());  // Seemingly important for decompression speed.
 218         for (const string &block : sampled_blocks) {
 219                 dictionary_buf += block;
 220         }
 221
 222         string buf;
 223         buf.resize(buf_size);
 224         size_t ret = ZDICT_trainFromBuffer(&buf[0], buf_size, dictionary_buf.data(), lengths.data(), lengths.size());
 225         dprintf(stderr, "Sampled %zu bytes in %zu blocks, built a dictionary of size %zu\n", dictionary_buf.size(), lengths.size(), ret);
 226         buf.resize(ret);
 227
 228         sampled_blocks.clear();
 229         lengths.clear();
 230
 231         return buf;
 232 }
 233
 234 class Corpus : public DatabaseReceiver {
 235 public:
 236         Corpus(FILE *outfp, size_t block_size, ZSTD_CDict *cdict)
 237                 : invindex(new PostingListBuilder *[NUM_TRIGRAMS]), outfp(outfp), block_size(block_size), cdict(cdict)
 238         {
 239                 fill(invindex.get(), invindex.get() + NUM_TRIGRAMS, nullptr);
 240         }
 241         ~Corpus() override
 242         {
 243                 for (unsigned i = 0; i < NUM_TRIGRAMS; ++i) {
 244                         delete invindex[i];
 245                 }
 246         }
 247
 248         void add_file(string filename) override;
 249         void flush_block() override;
 250
 251         vector<uint64_t> filename_blocks;
 252         size_t num_files = 0, num_files_in_block = 0, num_blocks = 0;
 253         bool seen_trigram(uint32_t trgm)
 254         {
 255                 return invindex[trgm] != nullptr;
 256         }
 257         PostingListBuilder &get_pl_builder(uint32_t trgm)
 258         {
 259                 if (invindex[trgm] == nullptr) {
 260                         invindex[trgm] = new PostingListBuilder;
 261                 }
 262                 return *invindex[trgm];
 263         }
 264
 265 private:
 266         unique_ptr<PostingListBuilder *[]> invindex;
 267         FILE *outfp;
 268         string current_block;
 269         string tempbuf;
 270         const size_t block_size;
 271         ZSTD_CDict *cdict;
 272 };
 273
 274 void Corpus::add_file(string filename)
 275 {
 276         ++num_files;
 277         if (!current_block.empty()) {
 278                 current_block.push_back('\0');
 279         }
 280         current_block += filename;
 281         if (++num_files_in_block == block_size) {
 282                 flush_block();
 283         }
 284 }
 285
 286 void Corpus::flush_block()
 287 {
 288         if (current_block.empty()) {
 289                 return;
 290         }
 291
 292         uint32_t docid = num_blocks;
 293
 294         // Create trigrams.
 295         const char *ptr = current_block.c_str();
 296         while (ptr < current_block.c_str() + current_block.size()) {
 297                 string_view s(ptr);
 298                 if (s.size() >= 3) {
 299                         for (size_t j = 0; j < s.size() - 2; ++j) {
 300                                 uint32_t trgm = read_trigram(s, j);
 301                                 get_pl_builder(trgm).add_docid(docid);
 302                         }
 303                 }
 304                 ptr += s.size() + 1;
 305         }
 306
 307         // Compress and add the filename block.
 308         filename_blocks.push_back(ftell(outfp));
 309         string compressed = zstd_compress(current_block, cdict, &tempbuf);
 310         if (fwrite(compressed.data(), compressed.size(), 1, outfp) != 1) {
 311                 perror("fwrite()");
 312                 exit(1);
 313         }
 314
 315         current_block.clear();
 316         num_files_in_block = 0;
 317         ++num_blocks;
 318 }
 319
 320 string read_cstr(FILE *fp)
 321 {
 322         string ret;
 323         for (;;) {
 324                 int ch = getc(fp);
 325                 if (ch == -1) {
 326                         perror("getc");
 327                         exit(1);
 328                 }
 329                 if (ch == 0) {
 330                         return ret;
 331                 }
 332                 ret.push_back(ch);
 333         }
 334 }
 335
 336 void handle_directory(FILE *fp, DatabaseReceiver *receiver)
 337 {
 338         db_directory dummy;
 339         if (fread(&dummy, sizeof(dummy), 1, fp) != 1) {
 340                 if (feof(fp)) {
 341                         return;
 342                 } else {
 343                         perror("fread");
 344                 }
 345         }
 346
 347         string dir_path = read_cstr(fp);
 348         if (dir_path == "/") {
 349                 dir_path = "";
 350         }
 351
 352         for (;;) {
 353                 int type = getc(fp);
 354                 if (type == DBE_NORMAL) {
 355                         string filename = read_cstr(fp);
 356                         receiver->add_file(dir_path + "/" + filename);
 357                 } else if (type == DBE_DIRECTORY) {
 358                         string dirname = read_cstr(fp);
 359                         receiver->add_file(dir_path + "/" + dirname);
 360                 } else {
 361                         return;  // Probably end.
 362                 }
 363         }
 364 }
 365
 366 void read_mlocate(const char *filename, DatabaseReceiver *receiver)
 367 {
 368         FILE *fp = fopen(filename, "rb");
 369         if (fp == nullptr) {
 370                 perror(filename);
 371                 exit(1);
 372         }
 373
 374         db_header hdr;
 375         if (fread(&hdr, sizeof(hdr), 1, fp) != 1) {
 376                 perror("short read");
 377                 exit(1);
 378         }
 379
 380         // TODO: Care about the base path.
 381         string path = read_cstr(fp);
 382         while (!feof(fp)) {
 383                 handle_directory(fp, receiver);
 384         }
 385         fclose(fp);
 386 }
 387
 388 string zstd_compress(const string &src, ZSTD_CDict *cdict, string *tempbuf)
 389 {
 390         static ZSTD_CCtx *ctx = nullptr;
 391         if (ctx == nullptr) {
 392                 ctx = ZSTD_createCCtx();
 393         }
 394
 395         size_t max_size = ZSTD_compressBound(src.size());
 396         if (tempbuf->size() < max_size) {
 397                 tempbuf->resize(max_size);
 398         }
 399         size_t size;
 400         if (cdict == nullptr) {
 401                 size = ZSTD_compressCCtx(ctx, &(*tempbuf)[0], max_size, src.data(), src.size(), /*level=*/6);
 402         } else {
 403                 size = ZSTD_compress_usingCDict(ctx, &(*tempbuf)[0], max_size, src.data(), src.size(), cdict);
 404         }
 405         return string(tempbuf->data(), size);
 406 }
 407
 408 bool is_prime(uint32_t x)
 409 {
 410         if ((x % 2) == 0 || (x % 3) == 0) {
 411                 return false;
 412         }
 413         uint32_t limit = ceil(sqrt(x));
 414         for (uint32_t factor = 5; factor <= limit; ++factor) {
 415                 if ((x % factor) == 0) {
 416                         return false;
 417                 }
 418         }
 419         return true;
 420 }
 421
 422 uint32_t next_prime(uint32_t x)
 423 {
 424         if ((x % 2) == 0) {
 425                 ++x;
 426         }
 427         while (!is_prime(x)) {
 428                 x += 2;
 429         }
 430         return x;
 431 }
 432
 433 unique_ptr<Trigram[]> create_hashtable(Corpus &corpus, const vector<uint32_t> &all_trigrams, uint32_t ht_size, uint32_t num_overflow_slots)
 434 {
 435         unique_ptr<Trigram[]> ht(new Trigram[ht_size + num_overflow_slots + 1]);  // 1 for the sentinel element at the end.
 436         for (unsigned i = 0; i < ht_size + num_overflow_slots + 1; ++i) {
 437                 ht[i].trgm = uint32_t(-1);
 438                 ht[i].num_docids = 0;
 439                 ht[i].offset = 0;
 440         }
 441         for (uint32_t trgm : all_trigrams) {
 442                 // We don't know offset yet, so set it to zero.
 443                 Trigram to_insert{ trgm, uint32_t(corpus.get_pl_builder(trgm).num_docids), 0 };
 444
 445                 uint32_t bucket = hash_trigram(trgm, ht_size);
 446                 unsigned distance = 0;
 447                 while (ht[bucket].num_docids != 0) {
 448                         // Robin Hood hashing; reduces the longest distance by a lot.
 449                         unsigned other_distance = bucket - hash_trigram(ht[bucket].trgm, ht_size);
 450                         if (distance > other_distance) {
 451                                 swap(to_insert, ht[bucket]);
 452                                 distance = other_distance;
 453                         }
 454
 455                         ++bucket, ++distance;
 456                         if (distance > num_overflow_slots) {
 457                                 return nullptr;
 458                         }
 459                 }
 460                 ht[bucket] = to_insert;
 461         }
 462         return ht;
 463 }
 464
 465 void do_build(const char *infile, const char *outfile, int block_size)
 466 {
 467         steady_clock::time_point start __attribute__((unused)) = steady_clock::now();
 468
 469         umask(0027);
 470         FILE *outfp = fopen(outfile, "wb");
 471
 472         // Write the header.
 473         Header hdr;
 474         memcpy(hdr.magic, "\0plocate", 8);
 475         hdr.version = -1;  // Mark as broken.
 476         hdr.hashtable_size = 0;  // Not known yet.
 477         hdr.extra_ht_slots = num_overflow_slots;
 478         hdr.num_docids = 0;
 479         hdr.hash_table_offset_bytes = -1;  // We don't know these offsets yet.
 480         hdr.max_version = 1;
 481         hdr.filename_index_offset_bytes = -1;
 482         hdr.zstd_dictionary_length_bytes = -1;
 483         fwrite(&hdr, sizeof(hdr), 1, outfp);
 484
 485         // Train the dictionary by sampling real blocks.
 486         // The documentation for ZDICT_trainFromBuffer() claims that a reasonable
 487         // dictionary size is ~100 kB, but 1 kB seems to actually compress better for us,
 488         // and decompress just as fast.
 489         DictionaryBuilder builder(/*blocks_to_keep=*/1000, block_size);
 490         read_mlocate(infile, &builder);
 491         string dictionary = builder.train(1024);
 492         ZSTD_CDict *cdict = ZSTD_createCDict(dictionary.data(), dictionary.size(), /*level=*/6);
 493
 494         hdr.zstd_dictionary_offset_bytes = ftell(outfp);
 495         fwrite(dictionary.data(), dictionary.size(), 1, outfp);
 496         hdr.zstd_dictionary_length_bytes = dictionary.size();
 497
 498         Corpus corpus(outfp, block_size, cdict);
 499         read_mlocate(infile, &corpus);
 500         if (false) {  // To read a plain text file.
 501                 FILE *fp = fopen(infile, "r");
 502                 while (!feof(fp)) {
 503                         char buf[1024];
 504                         if (fgets(buf, 1024, fp) == nullptr || feof(fp)) {
 505                                 break;
 506                         }
 507                         string s(buf);
 508                         if (s.back() == '\n')
 509                                 s.pop_back();
 510                         corpus.add_file(move(s));
 511                 }
 512                 fclose(fp);
 513         }
 514         corpus.flush_block();
 515         dprintf("Read %zu files from %s\n", corpus.num_files, infile);
 516         hdr.num_docids = corpus.filename_blocks.size();
 517
 518         // Stick an empty block at the end as sentinel.
 519         corpus.filename_blocks.push_back(ftell(outfp));
 520         const size_t bytes_for_filenames = corpus.filename_blocks.back() - corpus.filename_blocks.front();
 521
 522         // Write the offsets to the filenames.
 523         hdr.filename_index_offset_bytes = ftell(outfp);
 524         const size_t bytes_for_filename_index = corpus.filename_blocks.size() * sizeof(uint64_t);
 525         fwrite(corpus.filename_blocks.data(), corpus.filename_blocks.size(), sizeof(uint64_t), outfp);
 526         corpus.filename_blocks.clear();
 527         corpus.filename_blocks.shrink_to_fit();
 528
 529         // Finish up encoding the posting lists.
 530         size_t trigrams = 0, longest_posting_list = 0;
 531         size_t bytes_for_posting_lists = 0;
 532         for (unsigned trgm = 0; trgm < NUM_TRIGRAMS; ++trgm) {
 533                 if (!corpus.seen_trigram(trgm))
 534                         continue;
 535                 PostingListBuilder &pl_builder = corpus.get_pl_builder(trgm);
 536                 pl_builder.finish();
 537                 longest_posting_list = max(longest_posting_list, pl_builder.num_docids);
 538                 trigrams += pl_builder.num_docids;
 539                 bytes_for_posting_lists += pl_builder.encoded.size();
 540         }
 541         dprintf("%zu files, %zu different trigrams, %zu entries, avg len %.2f, longest %zu\n",
 542                 corpus.num_files, corpus.invindex.size(), trigrams, double(trigrams) / corpus.invindex.size(), longest_posting_list);
 543         dprintf("%zu bytes used for posting lists (%.2f bits/entry)\n", bytes_for_posting_lists, 8 * bytes_for_posting_lists / double(trigrams));
 544
 545         dprintf("Building posting lists took %.1f ms.\n\n", 1e3 * duration<float>(steady_clock::now() - start).count());
 546
 547         // Find the used trigrams.
 548         vector<uint32_t> all_trigrams;
 549         for (unsigned trgm = 0; trgm < NUM_TRIGRAMS; ++trgm) {
 550                 if (corpus.seen_trigram(trgm)) {
 551                         all_trigrams.push_back(trgm);
 552                 }
 553         }
 554
 555         // Create the hash table.
 556         unique_ptr<Trigram[]> hashtable;
 557         uint32_t ht_size = next_prime(all_trigrams.size());
 558         for (;;) {
 559                 hashtable = create_hashtable(corpus, all_trigrams, ht_size, num_overflow_slots);
 560                 if (hashtable == nullptr) {
 561                         dprintf("Failed creating hash table of size %u, increasing by 5%% and trying again.\n", ht_size);
 562                         ht_size = next_prime(ht_size * 1.05);
 563                 } else {
 564                         dprintf("Created hash table of size %u.\n\n", ht_size);
 565                         break;
 566                 }
 567         }
 568
 569         // Find the offsets for each posting list.
 570         size_t bytes_for_hashtable = (ht_size + num_overflow_slots + 1) * sizeof(Trigram);
 571         uint64_t offset = ftell(outfp) + bytes_for_hashtable;
 572         for (unsigned i = 0; i < ht_size + num_overflow_slots + 1; ++i) {
 573                 hashtable[i].offset = offset;  // Needs to be there even for empty slots.
 574                 if (hashtable[i].num_docids == 0) {
 575                         continue;
 576                 }
 577
 578                 const string &encoded = corpus.get_pl_builder(hashtable[i].trgm).encoded;
 579                 offset += encoded.size();
 580         }
 581
 582         // Write the hash table.
 583         hdr.hash_table_offset_bytes = ftell(outfp);
 584         hdr.hashtable_size = ht_size;
 585         fwrite(hashtable.get(), ht_size + num_overflow_slots + 1, sizeof(Trigram), outfp);
 586
 587         // Write the actual posting lists.
 588         for (unsigned i = 0; i < ht_size + num_overflow_slots + 1; ++i) {
 589                 if (hashtable[i].num_docids == 0) {
 590                         continue;
 591                 }
 592                 const string &encoded = corpus.get_pl_builder(hashtable[i].trgm).encoded;
 593                 fwrite(encoded.data(), encoded.size(), 1, outfp);
 594         }
 595
 596         // Rewind, and write the updated header.
 597         hdr.version = 1;
 598         fseek(outfp, 0, SEEK_SET);
 599         fwrite(&hdr, sizeof(hdr), 1, outfp);
 600         fclose(outfp);
 601
 602         size_t total_bytes __attribute__((unused)) = (bytes_for_hashtable + bytes_for_posting_lists + bytes_for_filename_index + bytes_for_filenames);
 603
 604         dprintf("Block size:     %7d files\n", block_size);
 605         dprintf("Dictionary:     %'7.1f MB\n", dictionary.size() / 1048576.0);
 606         dprintf("Hash table:     %'7.1f MB\n", bytes_for_hashtable / 1048576.0);
 607         dprintf("Posting lists:  %'7.1f MB\n", bytes_for_posting_lists / 1048576.0);
 608         dprintf("Filename index: %'7.1f MB\n", bytes_for_filename_index / 1048576.0);
 609         dprintf("Filenames:      %'7.1f MB\n", bytes_for_filenames / 1048576.0);
 610         dprintf("Total:          %'7.1f MB\n", total_bytes / 1048576.0);
 611         dprintf("\n");
 612 }
 613
 614 void usage()
 615 {
 616         printf(
 617                 "Usage: plocate-build MLOCATE_DB PLOCATE_DB\n"
 618                 "\n"
 619                 "Generate plocate index from mlocate.db, typically /var/lib/mlocate/mlocate.db.\n"
 620                 "Normally, the destination should be /var/lib/mlocate/plocate.db.\n"
 621                 "\n"
 622                 "  -b, --block-size SIZE  number of filenames to store in each block (default 32)\n"
 623                 "      --help             print this help\n"
 624                 "      --version          print version information\n");
 625 }
 626
 627 void version()
 628 {
 629         printf("plocate-build %s\n", PLOCATE_VERSION);
 630         printf("Copyright 2020 Steinar H. Gunderson\n");
 631         printf("License GPLv2+: GNU GPL version 2 or later <https://gnu.org/licenses/gpl.html>.\n");
 632         printf("This is free software: you are free to change and redistribute it.\n");
 633         printf("There is NO WARRANTY, to the extent permitted by law.\n");
 634 }
 635
 636 int main(int argc, char **argv)
 637 {
 638         static const struct option long_options[] = {
 639                 { "block-size", required_argument, 0, 'b' },
 640                 { "help", no_argument, 0, 'h' },
 641                 { "version", no_argument, 0, 'V' },
 642                 { 0, 0, 0, 0 }
 643         };
 644
 645         int block_size = 32;
 646
 647         setlocale(LC_ALL, "");
 648         for (;;) {
 649                 int option_index = 0;
 650                 int c = getopt_long(argc, argv, "b:hV", long_options, &option_index);
 651                 if (c == -1) {
 652                         break;
 653                 }
 654                 switch (c) {
 655                 case 'b':
 656                         block_size = atoi(optarg);
 657                         break;
 658                 case 'h':
 659                         usage();
 660                         exit(0);
 661                 case 'v':
 662                         version();
 663                         exit(0);
 664                 default:
 665                         exit(1);
 666                 }
 667         }
 668
 669         if (argc - optind != 2) {
 670                 usage();
 671                 exit(1);
 672         }
 673
 674         do_build(argv[optind], argv[optind + 1], block_size);
 675         exit(EXIT_SUCCESS);
 676 }