Partial implementation of B-tree name split/lookup

Name lookup brings back the O(m') scan-during-fetch approach of the
previous metadata layout. Since our rbyd trees map id+attr pairs and not
actual names, this beats the alternative O(m log(m)) scan of the tree.

Though tree searching does only include the current attributes, where as
scanning during fetch needs to also look at outdated attributes. Which
may make the winner less obvious depending on how we find the rbyd. But
being able to do the search in the same pass as fetch is an extra plus.

---

What turned out to be surprisingly complicated was the propagation of
names during B-tree splits and merges. The on-disk reference,
lfsr_data_t, does most of the heavy lifting here, but there's just a lot
of corner cases to consider.

At the moment this isn't working due to outdated names on the leading
entries of the rbyds, but to fix this bigger changes to the B-tree
layout may be needed.
This commit is contained in:
Christopher Haster
2023-03-20 01:37:39 -05:00
parent 0756c0acf2
commit 7a0842295c
2 changed files with 1299 additions and 137 deletions
+950 -5
View File
@@ -308,7 +308,7 @@ code = '''
# try larger trees, when exactly a tree splits depends on the disk geometry, so
# we don't really have a better way of testing multi-rbyd trees
[cases.test_btree_push]
defines.N = [4, 8, 16, 32, 64, 128, 256, 512, 1024]
defines.N = [1, 2, 4, 8, 16, 32, 64, 128, 256, 512, 1024]
in = 'lfs.c'
code = '''
lfs_t lfs;
@@ -2106,7 +2106,7 @@ code = '''
# test btree splits
[cases.test_btree_split]
defines.N = [4, 8, 16, 32, 64, 128, 256, 512, 1024]
defines.N = [1, 2, 4, 8, 16, 32, 64, 128, 256, 512, 1024]
in = 'lfs.c'
code = '''
lfs_t lfs;
@@ -2125,7 +2125,7 @@ code = '''
lfsr_btree_push(&lfs, &btree, 0, LFSR_TAG_INLINED, 1,
&alphas[0 % 26], 1) => 0;
for (lfs_size_t i = 1; i < N; i++) {
lfsr_btree_split(&lfs, &btree, i-1,
lfsr_btree_split(&lfs, &btree, i-1, NULL, 0,
LFSR_TAG_INLINED, 1, &alphas[(i-1) % 26], 1,
LFSR_TAG_INLINED, 1, &alphas[(i-0) % 26], 1) => 0;
}
@@ -2206,7 +2206,7 @@ code = '''
lfs_size_t id = TEST_PRNG(&prng) % sim_size;
// split btree
lfsr_btree_split(&lfs, &btree, id,
lfsr_btree_split(&lfs, &btree, id, NULL, 0,
LFSR_TAG_INLINED, 1, &alphas[i % 26], 1,
LFSR_TAG_INLINED, 1, &uppers[i % 26], 1) => 0;
@@ -2281,7 +2281,7 @@ code = '''
lfsr_btree_push(&lfs, &btree, 0, LFSR_TAG_INLINED, W,
&alphas[0 % 26], 1) => 0;
for (lfs_size_t i = 1; i < N; i++) {
lfsr_btree_split(&lfs, &btree, (i-1)*W+W-1,
lfsr_btree_split(&lfs, &btree, (i-1)*W+W-1, NULL, 0,
LFSR_TAG_INLINED, W, &alphas[(i-1) % 26], 1,
LFSR_TAG_INLINED, W, &alphas[(i-0) % 26], 1) => 0;
}
@@ -2376,6 +2376,7 @@ code = '''
// split btree
lfsr_btree_split(&lfs, &btree, weighted_id+sim_weights[id]-1,
NULL, 0,
LFSR_TAG_INLINED, weight1, &alphas[i % 26], 1,
LFSR_TAG_INLINED, weight2, &uppers[i % 26], 1) => 0;
@@ -2758,3 +2759,947 @@ code = '''
}
'''
# test key-value btrees
[cases.test_btree_find_one]
in = 'lfs.c'
code = '''
lfs_t lfs;
lfs_init(&lfs, cfg) => 0;
// create free lookahead
memset(lfs.free.buffer, 0, lfs.cfg->lookahead_size);
lfs.free.off = 0;
lfs.free.size = lfs_min(8*lfs.cfg->lookahead_size,
lfs.cfg->block_count);
lfs.free.i = 0;
lfs_alloc_ack(&lfs);
// create a single-entry tree
lfsr_btree_t btree = LFSR_BTREE_NULL;
lfsr_btree_push(&lfs, &btree, 0, LFSR_TAG_INLINED, 1, "0", 1) => 0;
printf("btree: 0x%x.%x 0x%x w%d\n",
btree.u.trunk.block,
btree.u.trunk.limit,
btree.tag,
btree.weight);
assert(btree.weight == 1);
// try to find tags
uint8_t buffer[4];
lfsr_tag_t tag_;
lfs_size_t id_;
lfs_size_t weight_;
lfsr_btree_find(&lfs, &btree, "aaa", 3,
&tag_, &id_, &weight_,
buffer, 4) => 1;
assert(tag_ == LFSR_TAG_INLINED);
assert(id_ == 0);
assert(weight_ == 1);
assert(memcmp(buffer, "0", 1) == 0);
lfsr_btree_find(&lfs, &btree, "aab", 3,
&tag_, &id_, &weight_,
buffer, 4) => 1;
assert(tag_ == LFSR_TAG_INLINED);
assert(id_ == 0);
assert(weight_ == 1);
assert(memcmp(buffer, "0", 1) == 0);
'''
[cases.test_btree_find_two]
in = 'lfs.c'
code = '''
lfs_t lfs;
lfs_init(&lfs, cfg) => 0;
// create free lookahead
memset(lfs.free.buffer, 0, lfs.cfg->lookahead_size);
lfs.free.off = 0;
lfs.free.size = lfs_min(8*lfs.cfg->lookahead_size,
lfs.cfg->block_count);
lfs.free.i = 0;
lfs_alloc_ack(&lfs);
// create a two-entry tree
lfsr_btree_t btree = LFSR_BTREE_NULL;
lfsr_btree_push(&lfs, &btree, 0, LFSR_TAG_INLINED, 1, "0", 1) => 0;
lfsr_btree_split(&lfs, &btree, 0, "aab", 3,
LFSR_TAG_INLINED, 1, "0", 1,
LFSR_TAG_INLINED, 1, "1", 1) => 0;
printf("btree: 0x%x.%x 0x%x w%d\n",
btree.u.trunk.block,
btree.u.trunk.limit,
btree.tag,
btree.weight);
assert(btree.weight == 2);
// try to find tags
uint8_t buffer[4];
lfsr_tag_t tag_;
lfs_size_t id_;
lfs_size_t weight_;
lfsr_btree_find(&lfs, &btree, "aaa", 3,
&tag_, &id_, &weight_,
buffer, 4) => 1;
assert(tag_ == LFSR_TAG_INLINED);
assert(id_ == 0);
assert(weight_ == 1);
assert(memcmp(buffer, "0", 1) == 0);
lfsr_btree_find(&lfs, &btree, "aab", 3,
&tag_, &id_, &weight_,
buffer, 4) => 1;
assert(tag_ == LFSR_TAG_INLINED);
assert(id_ == 1);
assert(weight_ == 1);
assert(memcmp(buffer, "1", 1) == 0);
lfsr_btree_find(&lfs, &btree, "aac", 3,
&tag_, &id_, &weight_,
buffer, 4) => 1;
assert(tag_ == LFSR_TAG_INLINED);
assert(id_ == 1);
assert(weight_ == 1);
assert(memcmp(buffer, "1", 1) == 0);
'''
[cases.test_btree_find_three]
in = 'lfs.c'
code = '''
lfs_t lfs;
lfs_init(&lfs, cfg) => 0;
// create free lookahead
memset(lfs.free.buffer, 0, lfs.cfg->lookahead_size);
lfs.free.off = 0;
lfs.free.size = lfs_min(8*lfs.cfg->lookahead_size,
lfs.cfg->block_count);
lfs.free.i = 0;
lfs_alloc_ack(&lfs);
// create a two-entry tree
lfsr_btree_t btree = LFSR_BTREE_NULL;
lfsr_btree_push(&lfs, &btree, 0, LFSR_TAG_INLINED, 1, "0", 1) => 0;
lfsr_btree_split(&lfs, &btree, 0, "aab", 3,
LFSR_TAG_INLINED, 1, "0", 1,
LFSR_TAG_INLINED, 1, "1", 1) => 0;
lfsr_btree_split(&lfs, &btree, 1, "aac", 3,
LFSR_TAG_INLINED, 1, "1", 1,
LFSR_TAG_INLINED, 1, "2", 1) => 0;
printf("btree: 0x%x.%x 0x%x w%d\n",
btree.u.trunk.block,
btree.u.trunk.limit,
btree.tag,
btree.weight);
assert(btree.weight == 3);
// try to find tags
uint8_t buffer[4];
lfsr_tag_t tag_;
lfs_size_t id_;
lfs_size_t weight_;
lfsr_btree_find(&lfs, &btree, "aaa", 3,
&tag_, &id_, &weight_,
buffer, 4) => 1;
assert(tag_ == LFSR_TAG_INLINED);
assert(id_ == 0);
assert(weight_ == 1);
assert(memcmp(buffer, "0", 1) == 0);
lfsr_btree_find(&lfs, &btree, "aab", 3,
&tag_, &id_, &weight_,
buffer, 4) => 1;
assert(tag_ == LFSR_TAG_INLINED);
assert(id_ == 1);
assert(weight_ == 1);
assert(memcmp(buffer, "1", 1) == 0);
lfsr_btree_find(&lfs, &btree, "aac", 3,
&tag_, &id_, &weight_,
buffer, 4) => 1;
assert(tag_ == LFSR_TAG_INLINED);
assert(id_ == 2);
assert(weight_ == 1);
assert(memcmp(buffer, "2", 1) == 0);
lfsr_btree_find(&lfs, &btree, "aad", 3,
&tag_, &id_, &weight_,
buffer, 4) => 1;
assert(tag_ == LFSR_TAG_INLINED);
assert(id_ == 2);
assert(weight_ == 1);
assert(memcmp(buffer, "2", 1) == 0);
'''
[cases.test_btree_find_three_backwards]
in = 'lfs.c'
code = '''
lfs_t lfs;
lfs_init(&lfs, cfg) => 0;
// create free lookahead
memset(lfs.free.buffer, 0, lfs.cfg->lookahead_size);
lfs.free.off = 0;
lfs.free.size = lfs_min(8*lfs.cfg->lookahead_size,
lfs.cfg->block_count);
lfs.free.i = 0;
lfs_alloc_ack(&lfs);
// create a two-entry tree
lfsr_btree_t btree = LFSR_BTREE_NULL;
lfsr_btree_push(&lfs, &btree, 0, LFSR_TAG_INLINED, 1, "0", 1) => 0;
lfsr_btree_split(&lfs, &btree, 0, "aac", 3,
LFSR_TAG_INLINED, 1, "1", 1,
LFSR_TAG_INLINED, 1, "2", 1) => 0;
lfsr_btree_split(&lfs, &btree, 0, "aab", 3,
LFSR_TAG_INLINED, 1, "0", 1,
LFSR_TAG_INLINED, 1, "1", 1) => 0;
printf("btree: 0x%x.%x 0x%x w%d\n",
btree.u.trunk.block,
btree.u.trunk.limit,
btree.tag,
btree.weight);
assert(btree.weight == 3);
// try to find tags
uint8_t buffer[4];
lfsr_tag_t tag_;
lfs_size_t id_;
lfs_size_t weight_;
lfsr_btree_find(&lfs, &btree, "aaa", 3,
&tag_, &id_, &weight_,
buffer, 4) => 1;
assert(tag_ == LFSR_TAG_INLINED);
assert(id_ == 0);
assert(weight_ == 1);
assert(memcmp(buffer, "0", 1) == 0);
lfsr_btree_find(&lfs, &btree, "aab", 3,
&tag_, &id_, &weight_,
buffer, 4) => 1;
assert(tag_ == LFSR_TAG_INLINED);
assert(id_ == 1);
assert(weight_ == 1);
assert(memcmp(buffer, "1", 1) == 0);
lfsr_btree_find(&lfs, &btree, "aac", 3,
&tag_, &id_, &weight_,
buffer, 4) => 1;
assert(tag_ == LFSR_TAG_INLINED);
assert(id_ == 2);
assert(weight_ == 1);
assert(memcmp(buffer, "2", 1) == 0);
lfsr_btree_find(&lfs, &btree, "aad", 3,
&tag_, &id_, &weight_,
buffer, 4) => 1;
assert(tag_ == LFSR_TAG_INLINED);
assert(id_ == 2);
assert(weight_ == 1);
assert(memcmp(buffer, "2", 1) == 0);
'''
[cases.test_btree_find]
defines.N = [1, 2, 4, 8, 16, 32, 64, 128, 256, 512, 1024]
in = 'lfs.c'
code = '''
lfs_t lfs;
lfs_init(&lfs, cfg) => 0;
// create free lookahead
memset(lfs.free.buffer, 0, lfs.cfg->lookahead_size);
lfs.free.off = 0;
lfs.free.size = lfs_min(8*lfs.cfg->lookahead_size,
lfs.cfg->block_count);
lfs.free.i = 0;
lfs_alloc_ack(&lfs);
// create a tree with N elements
lfsr_btree_t btree = LFSR_BTREE_NULL;
const char *alphas = "abcdefghijklmnopqrstuvwxyz";
const char *nums = "0123456789";
lfsr_btree_push(&lfs, &btree, 0, LFSR_TAG_INLINED, 1,
&nums[0 % 10], 1) => 0;
for (lfs_size_t i = 1; i < N; i++) {
char name[3] = {
alphas[(i/26/26) % 26], alphas[(i/26) % 26], alphas[i % 26]
};
lfsr_btree_split(&lfs, &btree, i-1, name, 3,
LFSR_TAG_INLINED, 1, &nums[(i-1) % 10], 1,
LFSR_TAG_INLINED, 1, &nums[(i-0) % 10], 1) => 0;
}
printf("btree: 0x%x.%x 0x%x w%d\n",
btree.u.trunk.block,
btree.u.trunk.limit,
btree.tag,
btree.weight);
assert(btree.weight == N);
// try to find tags
uint8_t buffer[4];
lfsr_tag_t tag_;
lfs_size_t id_;
lfs_size_t weight_;
for (lfs_size_t i = 0; i < N; i++) {
char name[3] = {
alphas[(i/26/26) % 26], alphas[(i/26) % 26], alphas[i % 26]
};
lfsr_btree_find(&lfs, &btree, name, 3,
&tag_, &id_, &weight_,
buffer, 4) => 1;
assert(tag_ == LFSR_TAG_INLINED);
assert(id_ == i);
assert(weight_ == 1);
assert(memcmp(buffer, &nums[i % 10], 1) == 0);
}
'''
[cases.test_btree_find_fuzz]
defines.N = [1, 2, 4, 8, 16, 32, 64, 128, 256, 512]
defines.SAMPLES = 10
# -1 => all pseudo-random seeds
# n => reproduce a specific seed
defines.SEED = -1
in = 'lfs.c'
code = '''
const char *alphas = "abcdefghijklmnopqrstuvwxyz";
const char *nums = "0123456789";
// iterate through severals seeds that we can reproduce easily
for (uint32_t seed = (SEED == -1 ? 1 : SEED);
(SEED == -1 ? seed < SAMPLES+1 : seed == SEED);
seed++) {
printf("--- seed: %d ---\n", seed);
// create lfs here since we need to reset each iteration, we're
// space constrained and we can't expect gc to work at this point
lfs_t lfs;
lfs_init(&lfs, cfg) => 0;
// create free lookahead
memset(lfs.free.buffer, 0, lfs.cfg->lookahead_size);
lfs.free.off = 0;
lfs.free.size = lfs_min(8*lfs.cfg->lookahead_size,
lfs.cfg->block_count);
lfs.free.i = 0;
lfs_alloc_ack(&lfs);
// create a btree
lfsr_btree_t btree = LFSR_BTREE_NULL;
lfsr_btree_push(&lfs, &btree, 0, LFSR_TAG_INLINED, 1,
&alphas[0 % 26], 1) => 0;
// set up a simulation to compare against
//
// fun fact this is slower than our actual tree! unfun fact this is
// starting to be a problem...
char *sim = malloc(N);
char (*sim_names)[3] = malloc(N*3);
lfs_size_t sim_size = 1;
memset(sim, 0, N);
memset(sim_names, 0, N*3);
sim[0] = alphas[0 % 26];
memcpy(&sim_names[0], "___", 3);
uint32_t prng = seed;
for (lfs_size_t i = 1; i < N; i++) {
// choose a pseudo-random name
lfs_size_t x = TEST_PRNG(&prng) % (26*26*26);
char name[3] = {
alphas[(x/26/26) % 26], alphas[(x/26) % 26], alphas[x % 26]
};
// find where to split
lfs_size_t id = 0;
while (id+1 < sim_size && memcmp(sim_names[id+1], name, 3) <= 0) {
id += 1;
}
// just skip exact matches for now
if (memcmp(sim_names[id], name, 3) == 0) {
continue;
}
// split btree
lfsr_btree_split(&lfs, &btree, id, name, 3,
LFSR_TAG_INLINED, 1, &nums[i % 10], 1,
LFSR_TAG_INLINED, 1, &nums[i % 10], 1) => 0;
// split sim
memmove(&sim[id+1], &sim[id], sim_size-id);
memmove(&sim_names[id+1], &sim_names[id], (sim_size-id)*3);
sim[id+0] = nums[i % 10];
sim[id+1] = nums[i % 10];
memcpy(&sim_names[id+1], name, 3);
sim_size += 1;
}
// check that btree matches sim
printf("expd: [");
bool first = true;
for (lfs_size_t i = 0; i < sim_size; i++) {
if (!first) {
printf(", ");
}
first = false;
printf("%.3s=%c", sim_names[i], sim[i]);
}
printf("]\n");
printf("btree: 0x%x.%x 0x%x w%d\n",
btree.u.trunk.block,
btree.u.trunk.limit,
btree.tag,
btree.weight);
assert(btree.weight == sim_size);
uint8_t buffer[4];
lfsr_tag_t tag_;
lfs_size_t id_;
lfs_size_t weight_;
for (lfs_size_t i = 0; i < sim_size; i++) {
lfsr_btree_find(&lfs, &btree, sim_names[i], 3,
&tag_, &id_, &weight_,
buffer, 4) => 1;
assert(tag_ == LFSR_TAG_INLINED);
assert(id_ == i);
assert(weight_ == 1);
assert(memcmp(buffer, &sim[i], 1) == 0);
}
// clean up sim
free(sim);
free(sim_names);
lfs_deinit(&lfs) => 0;
}
'''
[cases.test_btree_find_sparse]
defines.N = [1, 2, 4, 8, 16, 32, 64, 128, 256, 512, 1024]
defines.W = 5
in = 'lfs.c'
code = '''
lfs_t lfs;
lfs_init(&lfs, cfg) => 0;
// create free lookahead
memset(lfs.free.buffer, 0, lfs.cfg->lookahead_size);
lfs.free.off = 0;
lfs.free.size = lfs_min(8*lfs.cfg->lookahead_size,
lfs.cfg->block_count);
lfs.free.i = 0;
lfs_alloc_ack(&lfs);
// create a tree with N elements
lfsr_btree_t btree = LFSR_BTREE_NULL;
const char *alphas = "abcdefghijklmnopqrstuvwxyz";
const char *nums = "0123456789";
lfsr_btree_push(&lfs, &btree, 0, LFSR_TAG_INLINED, W,
&nums[0 % 10], 1) => 0;
for (lfs_size_t i = 1; i < N; i++) {
char name[3] = {
alphas[(i/26/26) % 26], alphas[(i/26) % 26], alphas[i % 26]
};
lfsr_btree_split(&lfs, &btree, (i-1)*W+W-1, name, 3,
LFSR_TAG_INLINED, W, &nums[(i-1) % 10], 1,
LFSR_TAG_INLINED, W, &nums[(i-0) % 10], 1) => 0;
}
printf("btree: 0x%x.%x 0x%x w%d\n",
btree.u.trunk.block,
btree.u.trunk.limit,
btree.tag,
btree.weight);
assert(btree.weight == N*W);
// try to find tags
uint8_t buffer[4];
lfsr_tag_t tag_;
lfs_size_t id_;
lfs_size_t weight_;
for (lfs_size_t i = 0; i < N; i++) {
char name[3] = {
alphas[(i/26/26) % 26], alphas[(i/26) % 26], alphas[i % 26]
};
lfsr_btree_find(&lfs, &btree, name, 3,
&tag_, &id_, &weight_,
buffer, 4) => 1;
assert(tag_ == LFSR_TAG_INLINED);
assert(id_ == i*W+W-1);
assert(weight_ == W);
assert(memcmp(buffer, &nums[i % 10], 1) == 0);
}
'''
[cases.test_btree_find_sparse_fuzz]
defines.N = [1, 2, 4, 8, 16, 32, 64, 128, 256, 512]
defines.W = 5
defines.SAMPLES = 10
# -1 => all pseudo-random seeds
# n => reproduce a specific seed
defines.SEED = -1
in = 'lfs.c'
code = '''
const char *alphas = "abcdefghijklmnopqrstuvwxyz";
const char *nums = "0123456789";
// iterate through severals seeds that we can reproduce easily
for (uint32_t seed = (SEED == -1 ? 1 : SEED);
(SEED == -1 ? seed < SAMPLES+1 : seed == SEED);
seed++) {
printf("--- seed: %d ---\n", seed);
// create lfs here since we need to reset each iteration, we're
// space constrained and we can't expect gc to work at this point
lfs_t lfs;
lfs_init(&lfs, cfg) => 0;
// create free lookahead
memset(lfs.free.buffer, 0, lfs.cfg->lookahead_size);
lfs.free.off = 0;
lfs.free.size = lfs_min(8*lfs.cfg->lookahead_size,
lfs.cfg->block_count);
lfs.free.i = 0;
lfs_alloc_ack(&lfs);
// create a btree
lfsr_btree_t btree = LFSR_BTREE_NULL;
lfsr_btree_push(&lfs, &btree, 0, LFSR_TAG_INLINED, W,
&alphas[0 % 26], 1) => 0;
// set up a simulation to compare against
//
// fun fact this is slower than our actual tree! unfun fact this is
// starting to be a problem...
char *sim = malloc(N);
char (*sim_names)[3] = malloc(N*3);
lfs_size_t *sim_weights = malloc(N*sizeof(lfs_size_t));
lfs_size_t sim_size = 1;
memset(sim, 0, N);
memset(sim_names, 0, N*3);
memset(sim_weights, 0, N*sizeof(lfs_size_t));
sim[0] = alphas[0 % 26];
memcpy(&sim_names[0], "___", 3);
sim_weights[0] = W;
uint32_t prng = seed;
for (lfs_size_t i = 1; i < N; i++) {
// choose a pseudo-random name
lfs_size_t x = TEST_PRNG(&prng) % (26*26*26);
char name[3] = {
alphas[(x/26/26) % 26], alphas[(x/26) % 26], alphas[x % 26]
};
// choose pseudo-random weights
lfs_size_t weight1 = 1 + (TEST_PRNG(&prng) % W);
lfs_size_t weight2 = 1 + (TEST_PRNG(&prng) % W);
// find where to split
lfs_size_t id = 0;
while (id+1 < sim_size && memcmp(sim_names[id+1], name, 3) <= 0) {
id += 1;
}
// just skip exact matches for now
if (memcmp(sim_names[id], name, 3) == 0) {
continue;
}
// calculate actual id in btree space
lfs_size_t weighted_id = 0;
for (lfs_size_t j = 0; j < id; j++) {
weighted_id += sim_weights[j];
}
// split btree
lfsr_btree_split(&lfs, &btree, weighted_id+sim_weights[id]-1,
name, 3,
LFSR_TAG_INLINED, weight1, &nums[i % 10], 1,
LFSR_TAG_INLINED, weight2, &nums[i % 10], 1) => 0;
// split sim
memmove(&sim[id+1], &sim[id], sim_size-id);
memmove(&sim_names[id+1], &sim_names[id], (sim_size-id)*3);
memmove(&sim_weights[id+1], &sim_weights[id],
(sim_size-id)*sizeof(lfs_size_t));
sim[id+0] = nums[i % 10];
sim[id+1] = nums[i % 10];
memcpy(&sim_names[id+1], name, 3);
sim_weights[id+0] = weight1;
sim_weights[id+1] = weight2;
sim_size += 1;
}
// check that btree matches sim
printf("expd: [");
bool first = true;
for (lfs_size_t i = 0; i < sim_size; i++) {
// calculate actual id in btree space
lfs_size_t weighted_id = 0;
for (lfs_size_t j = 0; j < i; j++) {
weighted_id += sim_weights[j];
}
if (!first) {
printf(", ");
}
first = false;
printf("%.3sid%dw%d=%c",
sim_names[i],
weighted_id+sim_weights[i]-1,
sim_weights[i],
sim[i]);
}
printf("]\n");
printf("btree: 0x%x.%x 0x%x w%d\n",
btree.u.trunk.block,
btree.u.trunk.limit,
btree.tag,
btree.weight);
lfs_size_t total_weight = 0;
for (lfs_size_t j = 0; j < N; j++) {
total_weight += sim_weights[j];
}
assert(btree.weight == total_weight);
uint8_t buffer[4];
lfsr_tag_t tag_;
lfs_size_t id_;
lfs_size_t weight_;
for (lfs_size_t i = 0; i < sim_size; i++) {
// calculate actual id in btree space
lfs_size_t weighted_id = 0;
for (lfs_size_t j = 0; j < i; j++) {
weighted_id += sim_weights[j];
}
lfsr_btree_find(&lfs, &btree, sim_names[i], 3,
&tag_, &id_, &weight_,
buffer, 4) => 1;
assert(tag_ == LFSR_TAG_INLINED);
assert(id_ == weighted_id+sim_weights[i]-1);
assert(weight_ == sim_weights[i]);
assert(memcmp(buffer, &sim[i], 1) == 0);
}
// clean up sim
free(sim);
free(sim_names);
free(sim_weights);
lfs_deinit(&lfs) => 0;
}
'''
# make sure we test finds with other operations
[cases.test_btree_find_general_fuzz]
defines.N = [1, 2, 4, 8, 16, 32, 64, 128, 256, 512, 1024]
defines.SAMPLES = 100
# -1 => all pseudo-random seeds
# n => reproduce a specific seed
defines.SEED = -1
in = 'lfs.c'
code = '''
const char *alphas = "abcdefghijklmnopqrstuvwxyz";
const char *nums = "0123456789";
// iterate through severals seeds that we can reproduce easily
for (uint32_t seed = (SEED == -1 ? 1 : SEED);
(SEED == -1 ? seed < SAMPLES+1 : seed == SEED);
seed++) {
printf("--- seed: %d ---\n", seed);
// create lfs here since we need to reset each iteration, we're
// space constrained and we can't expect gc to work at this point
lfs_t lfs;
lfs_init(&lfs, cfg) => 0;
// create free lookahead
memset(lfs.free.buffer, 0, lfs.cfg->lookahead_size);
lfs.free.off = 0;
lfs.free.size = lfs_min(8*lfs.cfg->lookahead_size,
lfs.cfg->block_count);
lfs.free.i = 0;
lfs_alloc_ack(&lfs);
// create a btree
lfsr_btree_t btree = LFSR_BTREE_NULL;
lfsr_btree_push(&lfs, &btree, 0, LFSR_TAG_INLINED, 1,
&alphas[0 % 26], 1) => 0;
// set up a simulation to compare against
//
// fun fact this is slower than our actual tree! unfun fact this is
// starting to be a problem...
char *sim = malloc(N);
char (*sim_names)[3] = malloc(N*3);
lfs_size_t sim_size = 1;
memset(sim, 0, N);
memset(sim_names, 0, N*3);
sim[0] = alphas[0 % 26];
memcpy(&sim_names[0], "___", 3);
uint32_t prng = seed;
for (lfs_size_t i = 0; i < N; i++) {
// choose a pseudo-random op
uint8_t op = TEST_PRNG(&prng) % 3;
// choose a pseudo-random id
lfs_size_t id = TEST_PRNG(&prng) % (sim_size == 0 ? 1 : sim_size);
// choose a pseudo-random name
lfs_size_t x = TEST_PRNG(&prng) % (26*26*26);
char name[3] = {
alphas[(x/26/26) % 26], alphas[(x/26) % 26], alphas[x % 26]
};
// don't let sim drop below one element
if (op == 0 || sim_size <= 1) {
// find where to split
printf("- split(\"%.3s\", \"%c\")\n", name, nums[i % 10]);
lfs_size_t id = 0;
while (id+1 < sim_size
&& memcmp(sim_names[id+1], name, 3) <= 0) {
id += 1;
}
// just skip exact matches for now
if (memcmp(sim_names[id], name, 3) == 0) {
continue;
}
// split btree
lfsr_btree_split(&lfs, &btree, id, name, 3,
LFSR_TAG_INLINED, 1, &nums[i % 10], 1,
LFSR_TAG_INLINED, 1, &nums[i % 10], 1) => 0;
// split sim
memmove(&sim[id+1], &sim[id], sim_size-id);
memmove(&sim_names[id+1], &sim_names[id], (sim_size-id)*3);
sim[id+0] = nums[i % 10];
sim[id+1] = nums[i % 10];
memcpy(&sim_names[id+1], name, 3);
sim_size += 1;
} else if (op == 1) {
// update btree
printf("- update(%d, \"%c\")\n", id, nums[i % 10]);
lfsr_btree_update(&lfs, &btree, id,
LFSR_TAG_INLINED, 1,
&nums[i % 10], 1) => 0;
// update sim
sim[id] = nums[i % 10];
} else {
// pop from btree
printf("- pop(%d)\n", id);
lfsr_btree_pop(&lfs, &btree, id) => 0;
// pop from sim
memmove(&sim[id], &sim[id+1], sim_size-(id+1));
memmove(&sim_names[id], &sim_names[id+1], (sim_size-(id+1))*3);
sim_size -= 1;
// our B-tree doesn't actually track the name of id0, so we need
// mirror this in our sim
if (id == 0) {
memcpy(&sim_names[0], "___", 3);
}
}
}
// check that btree matches sim
printf("expd: [");
bool first = true;
for (lfs_size_t i = 0; i < sim_size; i++) {
if (!first) {
printf(", ");
}
first = false;
printf("%.3s=%c", sim_names[i], sim[i]);
}
printf("]\n");
printf("btree: 0x%x.%x 0x%x w%d\n",
btree.u.trunk.block,
btree.u.trunk.limit,
btree.tag,
btree.weight);
assert(btree.weight == sim_size);
uint8_t buffer[4];
lfsr_tag_t tag_;
lfs_size_t id_;
lfs_size_t weight_;
for (lfs_size_t i = 0; i < sim_size; i++) {
lfsr_btree_find(&lfs, &btree, sim_names[i], 3,
&tag_, &id_, &weight_,
buffer, 4) => 1;
assert(tag_ == LFSR_TAG_INLINED);
assert(id_ == i);
assert(weight_ == 1);
assert(memcmp(buffer, &sim[i], 1) == 0);
}
// clean up sim
free(sim);
free(sim_names);
}
'''
[cases.test_btree_find_general_sparse_fuzz]
defines.N = [1, 2, 4, 8, 16, 32, 64, 128, 256, 512, 1024]
defines.W = 5
defines.SAMPLES = 100
# -1 => all pseudo-random seeds
# n => reproduce a specific seed
defines.SEED = -1
in = 'lfs.c'
code = '''
const char *alphas = "abcdefghijklmnopqrstuvwxyz";
// iterate through severals seeds that we can reproduce easily
for (uint32_t seed = (SEED == -1 ? 1 : SEED);
(SEED == -1 ? seed < SAMPLES+1 : seed == SEED);
seed++) {
printf("--- seed: %d ---\n", seed);
// create lfs here since we need to reset each iteration, we're
// space constrained and we can't expect gc to work at this point
lfs_t lfs;
lfs_init(&lfs, cfg) => 0;
// create free lookahead
memset(lfs.free.buffer, 0, lfs.cfg->lookahead_size);
lfs.free.off = 0;
lfs.free.size = lfs_min(8*lfs.cfg->lookahead_size,
lfs.cfg->block_count);
lfs.free.i = 0;
lfs_alloc_ack(&lfs);
// create a btree
lfsr_btree_t btree = LFSR_BTREE_NULL;
// set up a simulation to compare against
//
// fun fact this is slower than our actual tree! unfun fact this is
// starting to be a problem...
char *sim = malloc(N);
lfs_size_t *sim_weights = malloc(N*sizeof(lfs_size_t));
lfs_size_t sim_size = 0;
memset(sim, 0, N);
memset(sim_weights, 0, N*sizeof(lfs_size_t));
uint32_t prng = seed;
for (lfs_size_t i = 0; i < N; i++) {
// choose a pseudo-random op
uint8_t op = TEST_PRNG(&prng) % 3;
// choose a pseudo-random id
lfs_size_t id = TEST_PRNG(&prng) % (sim_size+1);
// choose a pseudo-random weight
lfs_size_t weight = 1 + (TEST_PRNG(&prng) % W);
// calculate actual id in btree space
lfs_size_t weighted_id = 0;
for (lfs_size_t j = 0; j < id; j++) {
weighted_id += sim_weights[j];
}
if (op == 0 || id == sim_size) {
// push to btree
lfsr_btree_push(&lfs, &btree, weighted_id,
LFSR_TAG_INLINED, weight,
&alphas[i % 26], 1) => 0;
// push to sim
memmove(&sim[id+1], &sim[id], sim_size-id);
memmove(&sim_weights[id+1], &sim_weights[id],
(sim_size-id)*sizeof(lfs_size_t));
sim[id] = alphas[i % 26];
sim_weights[id] = weight;
sim_size += 1;
} else if (op == 1) {
// update btree
lfsr_btree_update(&lfs, &btree,
weighted_id+sim_weights[id]-1, LFSR_TAG_INLINED, weight,
&alphas[i % 26], 1) => 0;
// update sim
sim[id] = alphas[i % 26];
sim_weights[id] = weight;
} else {
// remove from btree
lfsr_btree_pop(&lfs, &btree,
weighted_id+sim_weights[id]-1) => 0;
// remove from sim
memmove(&sim[id], &sim[id+1], sim_size-(id+1));
memmove(&sim_weights[id], &sim_weights[id+1],
(sim_size-(id+1))*sizeof(lfs_size_t));
sim_size -= 1;
}
}
// check that btree matches sim
printf("expd: [");
bool first = true;
for (lfs_size_t i = 0; i < sim_size; i++) {
if (!first) {
printf(", ");
}
first = false;
printf("%c", sim[i]);
}
printf("]\n");
printf("btree: 0x%x.%x 0x%x w%d\n",
btree.u.trunk.block,
btree.u.trunk.limit,
btree.tag,
btree.weight);
lfs_size_t total_weight = 0;
for (lfs_size_t j = 0; j < sim_size; j++) {
total_weight += sim_weights[j];
}
assert(btree.weight == total_weight);
uint8_t buffer[4];
lfsr_tag_t tag_;
lfs_size_t id_;
lfs_size_t weight_;
for (lfs_size_t i = 0; i < sim_size; i++) {
// calculate actual id in btree space
lfs_size_t weighted_id = 0;
for (lfs_size_t j = 0; j < i; j++) {
weighted_id += sim_weights[j];
}
lfsr_btree_get(&lfs, &btree, weighted_id+sim_weights[i]-1,
&tag_, &id_, &weight_,
buffer, 4) => 1;
assert(tag_ == LFSR_TAG_INLINED);
assert(id_ == weighted_id+sim_weights[i]-1);
assert(weight_ == sim_weights[i]);
assert(memcmp(buffer, &sim[i], 1) == 0);
}
// and no extra elements
lfsr_btree_get(&lfs, &btree, total_weight,
&tag_, &id_, &weight_,
buffer, 4) => LFS_ERR_NOENT;
// also test that we can traverse the tree without prior knowledge
id_ = -1;
for (lfs_size_t i = 0; i < sim_size; i++) {
// calculate actual id in btree space
lfs_size_t weighted_id = 0;
for (lfs_size_t j = 0; j < i; j++) {
weighted_id += sim_weights[j];
}
lfsr_btree_get(&lfs, &btree, id_+1,
&tag_, &id_, &weight_,
buffer, 4) => 1;
assert(tag_ == LFSR_TAG_INLINED);
assert(id_ == weighted_id+sim_weights[i]-1);
assert(weight_ == sim_weights[i]);
assert(memcmp(buffer, &sim[i], 1) == 0);
}
lfsr_btree_get(&lfs, &btree, id_+1,
&tag_, &id_, &weight_,
buffer, 4) => LFS_ERR_NOENT;
// clean up sim
free(sim);
}
'''