test: add lz_synthetic_generator for more interesting random data

This commit is contained in:
Marcus Holland-Moritz 2025-08-17 18:11:13 +02:00
parent 0c0bfd0cc7
commit 8549136da4
3 changed files with 256 additions and 5 deletions

View File

@ -0,0 +1,216 @@
/* vim:set ts=2 sw=2 sts=2 et: */
/**
* \author Marcus Holland-Moritz (github@mhxnet.de)
* \copyright Copyright (c) Marcus Holland-Moritz
*
* This file is part of dwarfs.
*
* dwarfs is free software: you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation, either version 3 of the License, or
* (at your option) any later version.
*
* dwarfs is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with dwarfs. If not, see <https://www.gnu.org/licenses/>.
*
* SPDX-License-Identifier: GPL-3.0-only
*/
#pragma once
#include <algorithm>
#include <array>
#include <cstdint>
#include <random>
#include <string>
#include <vector>
namespace dwarfs::test {
struct lz_params {
// Probability of choosing a "copy from the past" step vs. emitting a literal
double copy_prob = 0.70;
// Max distance for backreferences (typical LZ77 windows are 3264 KiB)
std::size_t window = 1u << 15; // 32 KiB
// Copy lengths ~ truncated geometric around this mean (controls
// repetitiveness)
std::size_t min_match = 4;
std::size_t max_match = 128;
double target_match_mean = 20.0; // average copy length
// Geometric distribution for distance (smaller distances more likely)
double distance_mean = 128.0;
// Chance each character in a copy mutates into a random literal (adds
// “noise”)
double mutation_rate = 0.005;
// If true, literals look like English-ish text; if false, literals are 0255
// bytes
bool text_mode = true;
// RNG seed for reproducibility
std::uint64_t seed = 0x1234'5678'9abc'def0ULL;
};
class lz_synthetic_generator {
public:
explicit lz_synthetic_generator(lz_params p = {})
: p_{p}
, rng_{p.seed} {
if (p_.text_mode) {
init_text_alphabet();
} else {
init_binary_alphabet();
}
// geometric_distribution parameterization: mean of failures = (1-p)/p
// We want E[min_match + failures] ≈ target_match_mean => E[failures] ≈
// target - min
double mean_fail =
std::max(1.0, p_.target_match_mean - static_cast<double>(p_.min_match));
double p_len = 1.0 / (mean_fail + 1.0);
geo_len_ = std::geometric_distribution<int>(p_len);
double mean_dist_fail = std::max(1.0, p_.distance_mean);
double p_dist = 1.0 / (mean_dist_fail + 1.0);
geo_dist_ = std::geometric_distribution<int>(p_dist);
bern_copy_ = std::bernoulli_distribution(p_.copy_prob);
bern_mut_ = std::bernoulli_distribution(p_.mutation_rate);
}
std::string generate(std::size_t n_bytes) {
std::string out;
out.reserve(n_bytes);
while (out.size() < n_bytes) {
bool const can_copy = out.size() >= p_.min_match;
if (can_copy && bern_copy_(rng_)) {
emit_copy(out, n_bytes);
} else {
out.push_back(sample_literal());
}
}
return out;
}
private:
void init_text_alphabet() {
// Rough English-ish frequencies via "etaoin shrdlu..." ranking.
// Higher rank => higher weight. We include space/newline/punct/digits.
static constexpr std::string_view freq_rank =
" etaoinshrdlucmfwypvbgkqjxz"; // space first (most frequent)
// Map ranks to weights (largest for rank 0).
std::array<int, 256> weights{};
for (int i = 0; i < 256; ++i) {
weights[i] = 1;
}
auto apply_rank = [&](char c, size_t rank_base) {
int r = std::max(1, static_cast<int>(freq_rank.size()) -
static_cast<int>(rank_base));
weights[static_cast<unsigned char>(c)] += r;
};
for (size_t i = 0; i < freq_rank.size(); ++i) {
char c = freq_rank[i];
apply_rank(c, i);
if (c >= 'a' && c <= 'z') {
apply_rank(char(c - 'a' + 'A'), i + 6); // uppercase similar but rarer
}
}
// Common punctuation and digits
std::string const punct = ".,;:-()[]{}!?\"'";
for (char c : punct) {
weights[static_cast<unsigned char>(c)] += 8;
}
for (char c = '0'; c <= '9'; ++c) {
weights[static_cast<unsigned char>(c)] += 4;
}
// Newlines and tabs, for “document” feel
weights['\n'] += 6;
weights['\t'] += 2;
// Build alphabet and weight vector for std::discrete_distribution
for (int i = 0; i < 256; ++i) {
if (weights[i] > 0) {
text_alphabet_.push_back(static_cast<unsigned char>(i));
text_weights_.push_back(weights[i]);
}
}
text_dist_ = std::discrete_distribution<int>(text_weights_.begin(),
text_weights_.end());
}
void init_binary_alphabet() {
binary_dist_ = std::uniform_int_distribution<int>(0, 255);
}
char sample_literal() {
if (p_.text_mode) {
int idx = text_dist_(rng_);
return static_cast<char>(text_alphabet_[static_cast<std::size_t>(idx)]);
}
return static_cast<char>(binary_dist_(rng_));
}
void emit_copy(std::string& out, std::size_t n_bytes) {
// Distance: 1 + geometric, truncated to current size and window
std::size_t max_dist = std::min<std::size_t>(p_.window, out.size());
if (max_dist == 0) {
out.push_back(sample_literal());
return;
}
std::size_t dist = 1u + static_cast<std::size_t>(geo_dist_(rng_));
if (dist > max_dist)
dist = 1u + (dist % max_dist); // ensure in-range
// Length: min_match + geometric, truncated by end and max_match
std::size_t max_len =
std::min<std::size_t>(p_.max_match, n_bytes - out.size());
if (max_len < p_.min_match) {
out.push_back(sample_literal());
return;
}
std::size_t len = p_.min_match + static_cast<std::size_t>(geo_len_(rng_));
if (len > max_len)
len = max_len;
std::size_t start = out.size() - dist;
for (std::size_t i = 0; i < len && out.size() < n_bytes; ++i) {
unsigned char c = static_cast<unsigned char>(out[start + i]);
if (bern_mut_(rng_)) {
c = static_cast<unsigned char>(sample_literal());
}
out.push_back(static_cast<char>(c));
}
}
lz_params p_;
std::mt19937_64 rng_;
std::vector<unsigned char> text_alphabet_;
std::vector<int> text_weights_;
std::discrete_distribution<int> text_dist_;
std::uniform_int_distribution<int> binary_dist_{0, 255};
std::bernoulli_distribution bern_copy_;
std::bernoulli_distribution bern_mut_;
std::geometric_distribution<int> geo_len_;
std::geometric_distribution<int> geo_dist_;
};
} // namespace dwarfs::test

View File

@ -38,6 +38,7 @@
#include <dwarfs/util.h>
#include "loremipsum.h"
#include "lz_synthetic_generator.h"
#include "mmap_mock.h"
#include "test_helpers.h"
@ -271,9 +272,10 @@ void os_access_mock::add_file(fs::path const& path, size_t size, bool random) {
if (random) {
thread_local std::mt19937_64 rng{42};
std::uniform_int_distribution<> choice_dist{0, 3};
std::uniform_int_distribution<> choice_dist{0, 4};
auto choice = choice_dist(rng);
switch (choice_dist(rng)) {
switch (choice) {
default:
break;
@ -281,6 +283,18 @@ void os_access_mock::add_file(fs::path const& path, size_t size, bool random) {
add(path, st,
[size, seed = rng()] { return create_random_string(size, seed); });
return;
case 1:
case 2: {
add(path, st, [size, seed = rng(), text_mode = choice == 1] {
lz_params lzp{};
lzp.text_mode = text_mode;
lzp.seed = seed;
lz_synthetic_generator gen{lzp};
return gen.generate(size);
});
return;
}
}
}

View File

@ -67,6 +67,7 @@
#include "filter_test_data.h"
#include "loremipsum.h"
#include "lz_synthetic_generator.h"
#include "mmap_mock.h"
#include "test_helpers.h"
#include "test_logger.h"
@ -258,6 +259,15 @@ class mkdwarfs_tester : public tester_common {
return test::create_random_string(size, 'A', 'Z', rng);
};
test::lz_params text_lzp{};
test::lz_params binary_lzp{};
text_lzp.text_mode = true;
binary_lzp.text_mode = false;
text_lzp.seed = rng();
binary_lzp.seed = rng();
test::lz_synthetic_generator text_gen{text_lzp};
test::lz_synthetic_generator binary_gen{binary_lzp};
for (int x = 0; x < opt.dimension; ++x) {
fs::path d1{random_path_component() + std::to_string(x)};
os->add_dir(d1);
@ -268,13 +278,24 @@ class mkdwarfs_tester : public tester_common {
for (int z = 0; z < opt.dimension; ++z) {
fs::path f{d2 / (random_path_component() + std::to_string(z))};
auto size = std::min(max_size, static_cast<size_t>(size_dist(rng)));
auto const size =
std::min(max_size, static_cast<size_t>(size_dist(rng)));
std::string data;
if (size < 1024 * 1024 && rng() % 2 == 0) {
auto const choice = rng() % 4;
switch (choice) {
case 0:
data = test::create_random_string(size, rng);
} else {
break;
case 1:
data = test::loremipsum(size);
break;
case 3:
data = text_gen.generate(size);
break;
case 4:
data = binary_gen.generate(size);
break;
}
os->add_file(f, data);