#include "db.h"
#include "io_uring_engine.h"
-#include "vp4.h"
#include <algorithm>
#include <arpa/inet.h>
+#include <assert.h>
#include <chrono>
#include <endian.h>
#include <fcntl.h>
#include <functional>
#include <getopt.h>
+#include <limits.h>
#include <memory>
#include <stdio.h>
#include <string.h>
#define dprintf(...)
//#define dprintf(...) fprintf(stderr, __VA_ARGS__);
+#include "turbopfor.h"
+
const char *dbpath = "/var/lib/mlocate/plocate.db";
bool print_nul = false;
class Serializer {
public:
- void do_or_wait(int seq, function<void()> cb);
+ bool ready_to_print(int seq) { return next_seq == seq; }
+ void print_delayed(int seq, const vector<string> msg);
+ void release_current();
private:
int next_seq = 0;
struct Element {
int seq;
- function<void()> cb;
+ vector<string> msg;
bool operator<(const Element &other) const
{
priority_queue<Element> pending;
};
-void Serializer::do_or_wait(int seq, function<void()> cb)
+void Serializer::print_delayed(int seq, const vector<string> msg)
{
- if (seq != next_seq) {
- pending.emplace(Element{ seq, move(cb) });
- return;
- }
+ pending.push(Element{ seq, move(msg) });
+}
- cb();
+void Serializer::release_current()
+{
++next_seq;
+ // See if any delayed prints can now be dealt with.
while (!pending.empty() && pending.top().seq == next_seq) {
- pending.top().cb();
+ for (const string &msg : pending.top().msg) {
+ if (print_nul) {
+ printf("%s%c", msg.c_str(), 0);
+ } else {
+ printf("%s\n", msg.c_str());
+ }
+ }
pending.pop();
++next_seq;
}
void Corpus::find_trigram(uint32_t trgm, function<void(const Trigram *trgmptr, size_t len)> cb)
{
uint32_t bucket = hash_trigram(trgm, hdr.hashtable_size);
- engine->submit_read(fd, sizeof(Trigram) * (hdr.extra_ht_slots + 2), hdr.hash_table_offset_bytes + sizeof(Trigram) * bucket, [this, trgm, bucket, cb{ move(cb) }](string s) {
+ 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 s) {
const Trigram *trgmptr = reinterpret_cast<const Trigram *>(s.data());
for (unsigned i = 0; i < hdr.extra_ht_slots + 1; ++i) {
if (trgmptr[i].trgm == trgm) {
size_t Corpus::get_num_filename_blocks() const
{
- // The beginning of the filename blocks is the end of the filename index blocks.
- uint64_t end;
- complete_pread(fd, &end, sizeof(end), hdr.filename_index_offset_bytes);
-
- // Subtract the sentinel block.
- return (end - hdr.filename_index_offset_bytes) / sizeof(uint64_t) - 1;
+ return hdr.num_docids;
}
size_t scan_file_block(const vector<string> &needles, string_view compressed,
- unordered_map<string, bool> *access_rx_cache)
+ unordered_map<string, bool> *access_rx_cache, int seq,
+ Serializer *serializer)
{
size_t matched = 0;
}
block[block.size() - 1] = '\0';
+ bool immediate_print = (serializer == nullptr || serializer->ready_to_print(seq));
+ vector<string> delayed;
+
for (const char *filename = block.data();
filename != block.data() + block.size();
filename += strlen(filename) + 1) {
}
if (found && has_access(filename, access_rx_cache)) {
++matched;
- if (print_nul) {
- printf("%s%c", filename, 0);
+ if (immediate_print) {
+ if (print_nul) {
+ printf("%s%c", filename, 0);
+ } else {
+ printf("%s\n", filename);
+ }
} else {
- printf("%s\n", filename);
+ delayed.push_back(filename);
}
}
}
+ if (serializer != nullptr) {
+ if (immediate_print) {
+ serializer->release_current();
+ } else {
+ serializer->print_delayed(seq, move(delayed));
+ }
+ }
return matched;
}
for (size_t i = 0; i < docids.size(); ++i) {
uint32_t docid = docids[i];
corpus.get_compressed_filename_block(docid, [i, &matched, &needles, &access_rx_cache, &docids_in_order](string compressed) {
- docids_in_order.do_or_wait(i, [&matched, &needles, compressed{ move(compressed) }, &access_rx_cache] {
- matched += scan_file_block(needles, compressed, &access_rx_cache);
- });
+ matched += scan_file_block(needles, compressed, &access_rx_cache, i, &docids_in_order);
});
}
engine->finish();
for (uint32_t docid = io_docid; docid < last_docid; ++docid) {
size_t relative_offset = offsets[docid] - offsets[io_docid];
size_t len = offsets[docid + 1] - offsets[docid];
- scan_file_block(needles, { &compressed[relative_offset], len }, &access_rx_cache);
+ scan_file_block(needles, { &compressed[relative_offset], len }, &access_rx_cache, 0, nullptr);
}
}
}
dprintf("Corpus init done after %.1f ms.\n", 1e3 * duration<float>(steady_clock::now() - start).count());
vector<pair<Trigram, size_t>> trigrams;
+ uint64_t shortest_so_far = numeric_limits<uint32_t>::max();
for (const string &needle : needles) {
- if (needle.size() < 3) continue;
+ if (needle.size() < 3)
+ continue;
for (size_t i = 0; i < needle.size() - 2; ++i) {
uint32_t trgm = read_trigram(needle, i);
- corpus.find_trigram(trgm, [trgm, &trigrams](const Trigram *trgmptr, size_t len) {
+ corpus.find_trigram(trgm, [trgm, &trigrams, &shortest_so_far](const Trigram *trgmptr, size_t len) {
if (trgmptr == nullptr) {
- dprintf("trigram %06x isn't found, we abort the search\n", trgm);
- return;
+ dprintf("trigram '%c%c%c' isn't found, we abort the search\n",
+ trgm & 0xff, (trgm >> 8) & 0xff, (trgm >> 16) & 0xff);
+ exit(0);
+ }
+ if (trgmptr->num_docids > shortest_so_far * 100) {
+ dprintf("not loading trigram '%c%c%c' with %u docids, it would be ignored later anyway\n",
+ trgm & 0xff, (trgm >> 8) & 0xff, (trgm >> 16) & 0xff,
+ trgmptr->num_docids);
+ } else {
+ trigrams.emplace_back(*trgmptr, len);
+ shortest_so_far = std::min<uint64_t>(shortest_so_far, trgmptr->num_docids);
}
- trigrams.emplace_back(*trgmptr, len);
});
}
}
if (done)
break;
}
- engine.submit_read(fd, len, trgmptr.offset, [trgmptr, len, &done, &in1, &in2, &out](string s) {
+ engine.submit_read(fd, len, trgmptr.offset, [trgmptr{trgmptr}, len{len}, &done, &in1, &in2, &out](string s) {
if (done)
return;
uint32_t trgm __attribute__((unused)) = trgmptr.trgm;
unsigned char *pldata = reinterpret_cast<unsigned char *>(s.data());
if (in1.empty()) {
in1.resize(num + 128);
- p4nd1dec128v32(pldata, num, &in1[0]);
+ decode_pfor_delta1<128>(pldata, num, /*interleaved=*/true, &in1[0]);
in1.resize(num);
dprintf("trigram '%c%c%c' (%zu bytes) decoded to %zu entries\n", trgm & 0xff,
(trgm >> 8) & 0xff, (trgm >> 16) & 0xff, len, num);
if (in2.size() < num + 128) {
in2.resize(num + 128);
}
- p4nd1dec128v32(pldata, num, &in2[0]);
+ decode_pfor_delta1<128>(pldata, num, /*interleaved=*/true, &in2[0]);
out.clear();
set_intersection(in1.begin(), in1.end(), in2.begin(), in2.begin() + num,
void usage()
{
- printf("Usage: slocate [OPTION]... PATTERN...\n");
+ // The help text comes from mlocate.
+ printf("Usage: plocate [OPTION]... PATTERN...\n");
+ printf("\n");
printf(" -d, --database DBPATH use DBPATH instead of default database (which is\n");
printf(" %s)\n", dbpath);
printf(" -h, --help print this help\n");
case '0':
print_nul = true;
break;
+ default:
+ exit(1);
}
}
needles.push_back(argv[i]);
}
if (needles.empty()) {
- fprintf(stderr, "slocate: no pattern to search for specified\n");
+ fprintf(stderr, "plocate: no pattern to search for specified\n");
exit(0);
}
do_search_file(needles, dbpath);
projects / plocate / blobdiff