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

   1 #include "db.h"
   2 #include "io_uring_engine.h"
   3 #include "parse_trigrams.h"
   4 #include "unique_sort.h"
   5
   6 #include <algorithm>
   7 #include <assert.h>
   8 #include <chrono>
   9 #include <fcntl.h>
  10 #include <fnmatch.h>
  11 #include <functional>
  12 #include <getopt.h>
  13 #include <iosfwd>
  14 #include <iterator>
  15 #include <limits>
  16 #include <memory>
  17 #include <queue>
  18 #include <regex.h>
  19 #include <stdint.h>
  20 #include <stdio.h>
  21 #include <stdlib.h>
  22 #include <string.h>
  23 #include <string>
  24 #include <string_view>
  25 #include <unistd.h>
  26 #include <unordered_map>
  27 #include <unordered_set>
  28 #include <utility>
  29 #include <vector>
  30 #include <zstd.h>
  31
  32 using namespace std;
  33 using namespace std::chrono;
  34
  35 #define dprintf(...)
  36 //#define dprintf(...) fprintf(stderr, __VA_ARGS__);
  37
  38 #include "turbopfor.h"
  39
  40 const char *dbpath = "/var/lib/mlocate/plocate.db";
  41 bool ignore_case = false;
  42 bool only_count = false;
  43 bool print_nul = false;
  44 int64_t limit_matches = numeric_limits<int64_t>::max();
  45
  46 class Serializer {
  47 public:
  48         bool ready_to_print(int seq) { return next_seq == seq; }
  49         void print_delayed(int seq, const vector<string> msg);
  50         void release_current();
  51
  52 private:
  53         int next_seq = 0;
  54         struct Element {
  55                 int seq;
  56                 vector<string> msg;
  57
  58                 bool operator<(const Element &other) const
  59                 {
  60                         return seq > other.seq;
  61                 }
  62         };
  63         priority_queue<Element> pending;
  64 };
  65
  66 void Serializer::print_delayed(int seq, const vector<string> msg)
  67 {
  68         pending.push(Element{ seq, move(msg) });
  69 }
  70
  71 void Serializer::release_current()
  72 {
  73         ++next_seq;
  74
  75         // See if any delayed prints can now be dealt with.
  76         while (!pending.empty() && pending.top().seq == next_seq) {
  77                 if (limit_matches-- <= 0)
  78                         return;
  79                 for (const string &msg : pending.top().msg) {
  80                         if (print_nul) {
  81                                 printf("%s%c", msg.c_str(), 0);
  82                         } else {
  83                                 printf("%s\n", msg.c_str());
  84                         }
  85                 }
  86                 pending.pop();
  87                 ++next_seq;
  88         }
  89 }
  90
  91 struct Needle {
  92         enum { STRSTR,
  93                REGEX,  // Not currently used.
  94                GLOB } type;
  95         string str;  // Filled in no matter what.
  96         regex_t re;  // For REGEX.
  97 };
  98
  99 bool matches(const Needle &needle, const char *haystack)
 100 {
 101         if (needle.type == Needle::STRSTR) {
 102                 return strstr(haystack, needle.str.c_str()) != nullptr;
 103         } else if (needle.type == Needle::GLOB) {
 104                 int flags = ignore_case ? FNM_CASEFOLD : 0;
 105                 return fnmatch(needle.str.c_str(), haystack, flags) == 0;
 106         } else {
 107                 assert(needle.type == Needle::REGEX);
 108                 return regexec(&needle.re, haystack, /*nmatch=*/0, /*pmatch=*/nullptr, /*flags=*/0) == 0;
 109         }
 110 }
 111
 112 bool has_access(const char *filename,
 113                 unordered_map<string, bool> *access_rx_cache)
 114 {
 115         const char *end = strchr(filename + 1, '/');
 116         while (end != nullptr) {
 117                 string parent_path(filename, end);
 118                 auto it = access_rx_cache->find(parent_path);
 119                 bool ok;
 120                 if (it == access_rx_cache->end()) {
 121                         ok = access(parent_path.c_str(), R_OK | X_OK) == 0;
 122                         access_rx_cache->emplace(move(parent_path), ok);
 123                 } else {
 124                         ok = it->second;
 125                 }
 126                 if (!ok) {
 127                         return false;
 128                 }
 129                 end = strchr(end + 1, '/');
 130         }
 131
 132         return true;
 133 }
 134
 135 class Corpus {
 136 public:
 137         Corpus(int fd, IOUringEngine *engine);
 138         ~Corpus();
 139         void find_trigram(uint32_t trgm, function<void(const Trigram *trgmptr, size_t len)> cb);
 140         void get_compressed_filename_block(uint32_t docid, function<void(string_view)> cb) const;
 141         size_t get_num_filename_blocks() const;
 142         off_t offset_for_block(uint32_t docid) const
 143         {
 144                 return hdr.filename_index_offset_bytes + docid * sizeof(uint64_t);
 145         }
 146
 147 public:
 148         const int fd;
 149         IOUringEngine *const engine;
 150
 151         Header hdr;
 152 };
 153
 154 Corpus::Corpus(int fd, IOUringEngine *engine)
 155         : fd(fd), engine(engine)
 156 {
 157         // Enable to test cold-cache behavior (except for access()).
 158         if (false) {
 159                 off_t len = lseek(fd, 0, SEEK_END);
 160                 if (len == -1) {
 161                         perror("lseek");
 162                         exit(1);
 163                 }
 164                 posix_fadvise(fd, 0, len, POSIX_FADV_DONTNEED);
 165         }
 166
 167         complete_pread(fd, &hdr, sizeof(hdr), /*offset=*/0);
 168         if (memcmp(hdr.magic, "\0plocate", 8) != 0) {
 169                 fprintf(stderr, "plocate.db is corrupt or an old version; please rebuild it.\n");
 170                 exit(1);
 171         }
 172         if (hdr.version != 0) {
 173                 fprintf(stderr, "plocate.db has version %u, expected 0; please rebuild it.\n", hdr.version);
 174                 exit(1);
 175         }
 176 }
 177
 178 Corpus::~Corpus()
 179 {
 180         close(fd);
 181 }
 182
 183 void Corpus::find_trigram(uint32_t trgm, function<void(const Trigram *trgmptr, size_t len)> cb)
 184 {
 185         uint32_t bucket = hash_trigram(trgm, hdr.hashtable_size);
 186         engine->submit_read(fd, sizeof(Trigram) * (hdr.extra_ht_slots + 2), hdr.hash_table_offset_bytes + sizeof(Trigram) * bucket, [this, trgm, cb{ move(cb) }](string_view s) {
 187                 const Trigram *trgmptr = reinterpret_cast<const Trigram *>(s.data());
 188                 for (unsigned i = 0; i < hdr.extra_ht_slots + 1; ++i) {
 189                         if (trgmptr[i].trgm == trgm) {
 190                                 cb(trgmptr + i, trgmptr[i + 1].offset - trgmptr[i].offset);
 191                                 return;
 192                         }
 193                 }
 194
 195                 // Not found.
 196                 cb(nullptr, 0);
 197         });
 198 }
 199
 200 void Corpus::get_compressed_filename_block(uint32_t docid, function<void(string_view)> cb) const
 201 {
 202         // Read the file offset from this docid and the next one.
 203         // This is always allowed, since we have a sentinel block at the end.
 204         engine->submit_read(fd, sizeof(uint64_t) * 2, offset_for_block(docid), [this, cb{ move(cb) }](string_view s) {
 205                 const uint64_t *ptr = reinterpret_cast<const uint64_t *>(s.data());
 206                 off_t offset = ptr[0];
 207                 size_t len = ptr[1] - ptr[0];
 208                 engine->submit_read(fd, len, offset, cb);
 209         });
 210 }
 211
 212 size_t Corpus::get_num_filename_blocks() const
 213 {
 214         return hdr.num_docids;
 215 }
 216
 217 uint64_t scan_file_block(const vector<Needle> &needles, string_view compressed,
 218                          unordered_map<string, bool> *access_rx_cache, int seq,
 219                          Serializer *serializer)
 220 {
 221         uint64_t matched = 0;
 222
 223         unsigned long long uncompressed_len = ZSTD_getFrameContentSize(compressed.data(), compressed.size());
 224         if (uncompressed_len == ZSTD_CONTENTSIZE_UNKNOWN || uncompressed_len == ZSTD_CONTENTSIZE_ERROR) {
 225                 fprintf(stderr, "ZSTD_getFrameContentSize() failed\n");
 226                 exit(1);
 227         }
 228
 229         string block;
 230         block.resize(uncompressed_len + 1);
 231
 232         size_t err = ZSTD_decompress(&block[0], block.size(), compressed.data(),
 233                                      compressed.size());
 234         if (ZSTD_isError(err)) {
 235                 fprintf(stderr, "ZSTD_decompress(): %s\n", ZSTD_getErrorName(err));
 236                 exit(1);
 237         }
 238         block[block.size() - 1] = '\0';
 239
 240         bool immediate_print = (serializer == nullptr || serializer->ready_to_print(seq));
 241         vector<string> delayed;
 242
 243         for (const char *filename = block.data();
 244              filename != block.data() + block.size();
 245              filename += strlen(filename) + 1) {
 246                 bool found = true;
 247                 for (const Needle &needle : needles) {
 248                         if (!matches(needle, filename)) {
 249                                 found = false;
 250                                 break;
 251                         }
 252                 }
 253                 if (found && has_access(filename, access_rx_cache)) {
 254                         if (limit_matches-- <= 0)
 255                                 break;
 256                         ++matched;
 257                         if (only_count)
 258                                 continue;
 259                         if (immediate_print) {
 260                                 if (print_nul) {
 261                                         printf("%s%c", filename, 0);
 262                                 } else {
 263                                         printf("%s\n", filename);
 264                                 }
 265                         } else {
 266                                 delayed.push_back(filename);
 267                         }
 268                 }
 269         }
 270         if (serializer != nullptr && !only_count) {
 271                 if (immediate_print) {
 272                         serializer->release_current();
 273                 } else {
 274                         serializer->print_delayed(seq, move(delayed));
 275                 }
 276         }
 277         return matched;
 278 }
 279
 280 size_t scan_docids(const vector<Needle> &needles, const vector<uint32_t> &docids, const Corpus &corpus, IOUringEngine *engine)
 281 {
 282         Serializer docids_in_order;
 283         unordered_map<string, bool> access_rx_cache;
 284         uint64_t matched = 0;
 285         for (size_t i = 0; i < docids.size(); ++i) {
 286                 uint32_t docid = docids[i];
 287                 corpus.get_compressed_filename_block(docid, [i, &matched, &needles, &access_rx_cache, &docids_in_order](string_view compressed) {
 288                         matched += scan_file_block(needles, compressed, &access_rx_cache, i, &docids_in_order);
 289                 });
 290         }
 291         engine->finish();
 292         return matched;
 293 }
 294
 295 // We do this sequentially, as it's faster than scattering
 296 // a lot of I/O through io_uring and hoping the kernel will
 297 // coalesce it plus readahead for us.
 298 uint64_t scan_all_docids(const vector<Needle> &needles, int fd, const Corpus &corpus, IOUringEngine *engine)
 299 {
 300         unordered_map<string, bool> access_rx_cache;
 301         uint32_t num_blocks = corpus.get_num_filename_blocks();
 302         unique_ptr<uint64_t[]> offsets(new uint64_t[num_blocks + 1]);
 303         complete_pread(fd, offsets.get(), (num_blocks + 1) * sizeof(uint64_t), corpus.offset_for_block(0));
 304         string compressed;
 305         uint64_t matched = 0;
 306         for (uint32_t io_docid = 0; io_docid < num_blocks; io_docid += 32) {
 307                 uint32_t last_docid = std::min(io_docid + 32, num_blocks);
 308                 size_t io_len = offsets[last_docid] - offsets[io_docid];
 309                 if (compressed.size() < io_len) {
 310                         compressed.resize(io_len);
 311                 }
 312                 complete_pread(fd, &compressed[0], io_len, offsets[io_docid]);
 313
 314                 for (uint32_t docid = io_docid; docid < last_docid; ++docid) {
 315                         size_t relative_offset = offsets[docid] - offsets[io_docid];
 316                         size_t len = offsets[docid + 1] - offsets[docid];
 317                         matched += scan_file_block(needles, { &compressed[relative_offset], len }, &access_rx_cache, 0, nullptr);
 318                         if (limit_matches <= 0)
 319                                 return matched;
 320                 }
 321         }
 322         return matched;
 323 }
 324
 325 // Takes the given posting list, unions it into the parts of the trigram disjunction
 326 // already read; if the list is complete, intersects with “cur_candidates”.
 327 //
 328 // Returns true if the search should be aborted (we are done).
 329 bool new_posting_list_read(TrigramDisjunction *td, vector<uint32_t> decoded, vector<uint32_t> *cur_candidates, vector<uint32_t> *tmp)
 330 {
 331         if (td->docids.empty()) {
 332                 td->docids = move(decoded);
 333         } else {
 334                 tmp->clear();
 335                 set_union(decoded.begin(), decoded.end(), td->docids.begin(), td->docids.end(), back_inserter(*tmp));
 336                 swap(*tmp, td->docids);
 337         }
 338         if (--td->remaining_trigrams_to_read > 0) {
 339                 // Need to wait for more.
 340                 if (ignore_case) {
 341                         dprintf("  ... %u reads left in OR group %u (%zu docids in list)\n",
 342                                 td->remaining_trigrams_to_read, td->index, td->docids.size());
 343                 }
 344                 return false;
 345         }
 346         if (cur_candidates->empty()) {
 347                 if (ignore_case) {
 348                         dprintf("  ... all reads done for OR group %u (%zu docids)\n",
 349                                 td->index, td->docids.size());
 350                 }
 351                 *cur_candidates = move(td->docids);
 352         } else {
 353                 tmp->clear();
 354                 set_intersection(cur_candidates->begin(), cur_candidates->end(),
 355                                  td->docids.begin(), td->docids.end(),
 356                                  back_inserter(*tmp));
 357                 swap(*cur_candidates, *tmp);
 358                 if (ignore_case) {
 359                         if (cur_candidates->empty()) {
 360                                 dprintf("  ... all reads done for OR group %u (%zu docids), intersected (none left, search is done)\n",
 361                                         td->index, td->docids.size());
 362                                 return true;
 363                         } else {
 364                                 dprintf("  ... all reads done for OR group %u (%zu docids), intersected (%zu left)\n",
 365                                         td->index, td->docids.size(), cur_candidates->size());
 366                         }
 367                 }
 368         }
 369         return false;
 370 }
 371
 372 void do_search_file(const vector<Needle> &needles, const char *filename)
 373 {
 374         int fd = open(filename, O_RDONLY);
 375         if (fd == -1) {
 376                 perror(filename);
 377                 exit(1);
 378         }
 379
 380         // Drop privileges.
 381         if (setgid(getgid()) != 0) {
 382                 perror("setgid");
 383                 exit(EXIT_FAILURE);
 384         }
 385
 386         steady_clock::time_point start __attribute__((unused)) = steady_clock::now();
 387         if (access("/", R_OK | X_OK)) {
 388                 // We can't find anything, no need to bother...
 389                 return;
 390         }
 391
 392         IOUringEngine engine(/*slop_bytes=*/16);  // 16 slop bytes as described in turbopfor.h.
 393         Corpus corpus(fd, &engine);
 394         dprintf("Corpus init done after %.1f ms.\n", 1e3 * duration<float>(steady_clock::now() - start).count());
 395
 396         vector<TrigramDisjunction> trigram_groups;
 397         for (const Needle &needle : needles) {
 398                 if (needle.str.size() < 3)
 399                         continue;
 400                 parse_trigrams(needle.str, ignore_case, &trigram_groups);
 401         }
 402         unique_sort(
 403                 &trigram_groups,
 404                 [](const TrigramDisjunction &a, const TrigramDisjunction &b) { return a.trigram_alternatives < b.trigram_alternatives; },
 405                 [](const TrigramDisjunction &a, const TrigramDisjunction &b) { return a.trigram_alternatives == b.trigram_alternatives; });
 406
 407         // Give them names for debugging.
 408         unsigned td_index = 0;
 409         for (TrigramDisjunction &td : trigram_groups) {
 410                 td.index = td_index++;
 411         }
 412
 413         // Collect which trigrams we need to look up in the hash table.
 414         unordered_map<uint32_t, vector<TrigramDisjunction *>> trigrams_to_lookup;
 415         for (TrigramDisjunction &td : trigram_groups) {
 416                 for (uint32_t trgm : td.trigram_alternatives) {
 417                         trigrams_to_lookup[trgm].push_back(&td);
 418                 }
 419         }
 420         if (trigrams_to_lookup.empty()) {
 421                 // Too short for trigram matching. Apply brute force.
 422                 // (We could have searched through all trigrams that matched
 423                 // the pattern and done a union of them, but that's a lot of
 424                 // work for fairly unclear gain.)
 425                 uint64_t matched = scan_all_docids(needles, fd, corpus, &engine);
 426                 if (only_count) {
 427                         printf("%zu\n", matched);
 428                 }
 429                 return;
 430         }
 431
 432         // Look them all up on disk.
 433         for (auto &[trgm, trigram_groups] : trigrams_to_lookup) {
 434                 corpus.find_trigram(trgm, [trgm{ trgm }, trigram_groups{ &trigram_groups }](const Trigram *trgmptr, size_t len) {
 435                         if (trgmptr == nullptr) {
 436                                 dprintf("trigram %s isn't found\n", print_trigram(trgm).c_str());
 437                                 for (TrigramDisjunction *td : *trigram_groups) {
 438                                         --td->remaining_trigrams_to_read;
 439                                         if (td->remaining_trigrams_to_read == 0 && td->read_trigrams.empty()) {
 440                                                 dprintf("zero matches in %s, so we are done\n", print_td(*td).c_str());
 441                                                 if (only_count) {
 442                                                         printf("0\n");
 443                                                 }
 444                                                 exit(0);
 445                                         }
 446                                 }
 447                                 return;
 448                         }
 449                         for (TrigramDisjunction *td : *trigram_groups) {
 450                                 --td->remaining_trigrams_to_read;
 451                                 td->max_num_docids += trgmptr->num_docids;
 452                                 td->read_trigrams.emplace_back(*trgmptr, len);
 453                         }
 454                 });
 455         }
 456         engine.finish();
 457         dprintf("Hashtable lookups done after %.1f ms.\n", 1e3 * duration<float>(steady_clock::now() - start).count());
 458
 459         for (TrigramDisjunction &td : trigram_groups) {
 460                 // Reset for reads.
 461                 td.remaining_trigrams_to_read = td.read_trigrams.size();
 462
 463                 if (ignore_case) {  // If case-sensitive, they'll all be pretty obvious single-entry groups.
 464                         dprintf("OR group %u (max_num_docids=%u): %s\n", td.index, td.max_num_docids, print_td(td).c_str());
 465                 }
 466         }
 467
 468         // TODO: For case-insensitive (ie. more than one alternative in each),
 469         // prioritize the ones with fewer seeks?
 470         sort(trigram_groups.begin(), trigram_groups.end(),
 471              [&](const TrigramDisjunction &a, const TrigramDisjunction &b) {
 472                      return a.max_num_docids < b.max_num_docids;
 473              });
 474
 475         unordered_map<uint32_t, vector<TrigramDisjunction *>> uses_trigram;
 476         for (TrigramDisjunction &td : trigram_groups) {
 477                 for (uint32_t trgm : td.trigram_alternatives) {
 478                         uses_trigram[trgm].push_back(&td);
 479                 }
 480         }
 481
 482         unordered_set<uint32_t> trigrams_submitted_read;
 483         vector<uint32_t> cur_candidates, tmp, decoded;
 484         bool done = false;
 485         for (TrigramDisjunction &td : trigram_groups) {
 486                 if (!cur_candidates.empty() && td.max_num_docids > cur_candidates.size() * 100) {
 487                         dprintf("%s has up to %u entries, ignoring the rest (will "
 488                                 "weed out false positives later)\n",
 489                                 print_td(td).c_str(), td.max_num_docids);
 490                         break;
 491                 }
 492
 493                 for (auto &[trgmptr, len] : td.read_trigrams) {
 494                         if (trigrams_submitted_read.count(trgmptr.trgm) != 0) {
 495                                 continue;
 496                         }
 497                         trigrams_submitted_read.insert(trgmptr.trgm);
 498                         // Only stay a certain amount ahead, so that we don't spend I/O
 499                         // on reading the latter, large posting lists. We are unlikely
 500                         // to need them anyway, even if they should come in first.
 501                         if (engine.get_waiting_reads() >= 5) {
 502                                 engine.finish();
 503                                 if (done)
 504                                         break;
 505                         }
 506                         engine.submit_read(fd, len, trgmptr.offset, [trgmptr{ trgmptr }, len{ len }, &done, &cur_candidates, &tmp, &decoded, &uses_trigram](string_view s) {
 507                                 if (done)
 508                                         return;
 509
 510                                 uint32_t trgm __attribute__((unused)) = trgmptr.trgm;
 511                                 const unsigned char *pldata = reinterpret_cast<const unsigned char *>(s.data());
 512                                 size_t num = trgmptr.num_docids;
 513                                 decoded.resize(num);
 514                                 decode_pfor_delta1_128(pldata, num, /*interleaved=*/true, &decoded[0]);
 515
 516                                 assert(uses_trigram.count(trgm) != 0);
 517                                 bool was_empty = cur_candidates.empty();
 518                                 if (ignore_case) {
 519                                         dprintf("trigram %s (%zu bytes) decoded to %zu entries\n", print_trigram(trgm).c_str(), len, num);
 520                                 }
 521
 522                                 for (TrigramDisjunction *td : uses_trigram[trgm]) {
 523                                         done |= new_posting_list_read(td, decoded, &cur_candidates, &tmp);
 524                                         if (done)
 525                                                 break;
 526                                 }
 527                                 if (!ignore_case) {
 528                                         if (was_empty) {
 529                                                 dprintf("trigram %s (%zu bytes) decoded to %zu entries\n", print_trigram(trgm).c_str(), len, num);
 530                                         } else if (cur_candidates.empty()) {
 531                                                 dprintf("trigram %s (%zu bytes) decoded to %zu entries (none left, search is done)\n", print_trigram(trgm).c_str(), len, num);
 532                                         } else {
 533                                                 dprintf("trigram %s (%zu bytes) decoded to %zu entries (%zu left)\n", print_trigram(trgm).c_str(), len, num, cur_candidates.size());
 534                                         }
 535                                 }
 536                         });
 537                 }
 538         }
 539         engine.finish();
 540         if (done) {
 541                 return;
 542         }
 543         dprintf("Intersection done after %.1f ms. Doing final verification and printing:\n",
 544                 1e3 * duration<float>(steady_clock::now() - start).count());
 545
 546         uint64_t matched = scan_docids(needles, cur_candidates, corpus, &engine);
 547         dprintf("Done in %.1f ms, found %zu matches.\n",
 548                 1e3 * duration<float>(steady_clock::now() - start).count(), matched);
 549
 550         if (only_count) {
 551                 printf("%zu\n", matched);
 552         }
 553 }
 554
 555 string unescape_glob_to_plain_string(const string &needle)
 556 {
 557         string unescaped;
 558         for (size_t i = 0; i < needle.size(); i += read_unigram(needle, i).second) {
 559                 uint32_t ch = read_unigram(needle, i).first;
 560                 assert(ch != WILDCARD_UNIGRAM);
 561                 if (ch == PREMATURE_END_UNIGRAM) {
 562                         fprintf(stderr, "Pattern '%s' ended prematurely\n", needle.c_str());
 563                         exit(1);
 564                 }
 565                 unescaped.push_back(ch);
 566         }
 567         return unescaped;
 568 }
 569
 570 void usage()
 571 {
 572         // The help text comes from mlocate.
 573         printf("Usage: plocate [OPTION]... PATTERN...\n");
 574         printf("\n");
 575         printf("  -c, --count            only print number of found entries\n");
 576         printf("  -d, --database DBPATH  use DBPATH instead of default database (which is\n");
 577         printf("                         %s)\n", dbpath);
 578         printf("  -h, --help             print this help\n");
 579         printf("  -i, --ignore-case      ignore case distinctions when matching patterns\n");
 580         printf("  -l, --limit, -n LIMIT  limit output (or counting) to LIMIT entries\n");
 581         printf("  -0, --null             separate entries with NUL on output\n");
 582 }
 583
 584 int main(int argc, char **argv)
 585 {
 586         static const struct option long_options[] = {
 587                 { "help", no_argument, 0, 'h' },
 588                 { "count", no_argument, 0, 'c' },
 589                 { "database", required_argument, 0, 'd' },
 590                 { "ignore-case", no_argument, 0, 'i' },
 591                 { "limit", required_argument, 0, 'l' },
 592                 { nullptr, required_argument, 0, 'n' },
 593                 { "null", no_argument, 0, '0' },
 594                 { 0, 0, 0, 0 }
 595         };
 596
 597         setlocale(LC_ALL, "");
 598         for (;;) {
 599                 int option_index = 0;
 600                 int c = getopt_long(argc, argv, "cd:hil:n:0", long_options, &option_index);
 601                 if (c == -1) {
 602                         break;
 603                 }
 604                 switch (c) {
 605                 case 'c':
 606                         only_count = true;
 607                         break;
 608                 case 'd':
 609                         dbpath = strdup(optarg);
 610                         break;
 611                 case 'h':
 612                         usage();
 613                         exit(0);
 614                 case 'i':
 615                         ignore_case = true;
 616                         break;
 617                 case 'l':
 618                 case 'n':
 619                         limit_matches = atoll(optarg);
 620                         break;
 621                 case '0':
 622                         print_nul = true;
 623                         break;
 624                 default:
 625                         exit(1);
 626                 }
 627         }
 628
 629         vector<Needle> needles;
 630         for (int i = optind; i < argc; ++i) {
 631                 Needle needle;
 632                 needle.str = argv[i];
 633
 634                 // See if there are any wildcard characters, which indicates we should treat it
 635                 // as an (anchored) glob.
 636                 bool any_wildcard = false;
 637                 for (size_t i = 0; i < needle.str.size(); i += read_unigram(needle.str, i).second) {
 638                         if (read_unigram(needle.str, i).first == WILDCARD_UNIGRAM) {
 639                                 any_wildcard = true;
 640                                 break;
 641                         }
 642                 }
 643
 644                 if (any_wildcard) {
 645                         needle.type = Needle::GLOB;
 646                 } else if (ignore_case) {
 647                         // strcasestr() doesn't handle locales correctly (even though LSB
 648                         // claims it should), but somehow, fnmatch() does, and it's about
 649                         // the same speed as using a regex.
 650                         needle.type = Needle::GLOB;
 651                         needle.str = "*" + needle.str + "*";
 652                 } else {
 653                         needle.type = Needle::STRSTR;
 654                         needle.str = unescape_glob_to_plain_string(needle.str);
 655                 }
 656                 needles.push_back(move(needle));
 657         }
 658         if (needles.empty()) {
 659                 fprintf(stderr, "plocate: no pattern to search for specified\n");
 660                 exit(0);
 661         }
 662         do_search_file(needles, dbpath);
 663 }