Prioritize modified rows when calculating changes for sorted results

This commit is contained in:
Thomas Goyne 2016-03-08 19:00:31 -08:00
parent a428f813d5
commit 424f4e829f
6 changed files with 116 additions and 34 deletions

View File

@ -357,10 +357,11 @@ void calculate_moves_unsorted(std::vector<RowInfo>& new_rows, CollectionChangeIn
using items = std::vector<std::pair<size_t, size_t>>; using items = std::vector<std::pair<size_t, size_t>>;
struct Match { struct Match {
size_t i, j, size; size_t i, j, size, modified;
}; };
Match find_longest_match(items const& a, items const& b, Match find_longest_match(items const& a, items const& b,
IndexSet const& modified,
size_t begin1, size_t end1, size_t begin2, size_t end2) size_t begin1, size_t end1, size_t begin2, size_t end2)
{ {
Match best = {begin1, begin2, 0}; Match best = {begin1, begin2, 0};
@ -384,11 +385,18 @@ Match find_longest_match(items const& a, items const& b,
size_t off = j - begin2; size_t off = j - begin2;
size_t size = off == 0 ? 1 : len_from_j_prev[off - 1] + 1; size_t size = off == 0 ? 1 : len_from_j_prev[off - 1] + 1;
len_from_j[off] = size; len_from_j[off] = size;
if (size > best.size) { if (size > best.size)
best.i = i - size + 1; best = {i - size + 1, j - size + 1, size, npos};
best.j = j - size + 1; // Given two equal-length matches, prefer the one with fewer modified rows
best.size = size; else if (size == best.size) {
if (best.modified == npos)
best.modified = modified.count(best.j - size + 1, best.j + 1);
auto count = modified.count(j - size + 1, j + 1);
if (count < best.modified)
best = {i - size + 1, j - size + 1, size, count};
} }
REALM_ASSERT(best.i >= begin1 && best.i + best.size <= end1);
REALM_ASSERT(best.j >= begin2 && best.j + best.size <= end2);
} }
len_from_j.swap(len_from_j_prev); len_from_j.swap(len_from_j_prev);
} }
@ -396,51 +404,40 @@ Match find_longest_match(items const& a, items const& b,
} }
void find_longest_matches(items const& a, items const& b_ndx, void find_longest_matches(items const& a, items const& b_ndx,
size_t begin1, size_t end1, size_t begin2, size_t end2, std::vector<Match>& ret) size_t begin1, size_t end1, size_t begin2, size_t end2,
IndexSet const& modified, std::vector<Match>& ret)
{ {
// FIXME: recursion could get too deep here // FIXME: recursion could get too deep here
Match m = find_longest_match(a, b_ndx, begin1, end1, begin2, end2); auto m = find_longest_match(a, b_ndx, modified, begin1, end1, begin2, end2);
if (!m.size) if (!m.size)
return; return;
if (m.i > begin1 && m.j > begin2) if (m.i > begin1 && m.j > begin2)
find_longest_matches(a, b_ndx, begin1, m.i, begin2, m.j, ret); find_longest_matches(a, b_ndx, begin1, m.i, begin2, m.j, modified, ret);
ret.push_back(m); ret.push_back(m);
if (m.i + m.size < end2 && m.j + m.size < end2) if (m.i + m.size < end2 && m.j + m.size < end2)
find_longest_matches(a, b_ndx, m.i + m.size, end1, m.j + m.size, end2, ret); find_longest_matches(a, b_ndx, m.i + m.size, end1, m.j + m.size, end2, modified, ret);
} }
void calculate_moves_sorted(std::vector<RowInfo>& new_rows, CollectionChangeIndices& changeset, void calculate_moves_sorted(std::vector<RowInfo>& new_rows, CollectionChangeIndices& changeset,
std::function<bool (size_t)> row_did_change) std::function<bool (size_t)> row_did_change)
{ {
// Remove new insertions, as they're already reported in changeset
new_rows.erase(std::remove_if(begin(new_rows), end(new_rows),
[](auto& row) { return row.prev_tv_index == npos; }),
end(new_rows));
std::sort(begin(new_rows), end(new_rows),
[](auto& lft, auto& rgt) { return lft.tv_index < rgt.tv_index; });
std::vector<std::pair<size_t, size_t>> old_candidates; std::vector<std::pair<size_t, size_t>> old_candidates;
std::vector<std::pair<size_t, size_t>> new_candidates; std::vector<std::pair<size_t, size_t>> new_candidates;
std::sort(begin(new_rows), end(new_rows), [](auto& lft, auto& rgt) {
return lft.tv_index < rgt.tv_index;
});
IndexSet::IndexInterator ins = changeset.insertions.begin(), del = changeset.deletions.begin();
int shift = 0;
for (auto& row : new_rows) { for (auto& row : new_rows) {
while (del != changeset.deletions.end() && *del <= row.tv_index) {
++del;
++shift;
}
while (ins != changeset.insertions.end() && *ins <= row.tv_index) {
++ins;
--shift;
}
if (row.prev_tv_index == npos)
continue;
if (row_did_change(row.shifted_row_index)) { if (row_did_change(row.shifted_row_index)) {
// FIXME: needlessly quadratic
if (!changeset.insertions.contains(row.tv_index))
changeset.modifications.add(row.tv_index); changeset.modifications.add(row.tv_index);
} }
old_candidates.push_back({row.shifted_row_index, row.prev_tv_index}); old_candidates.push_back({row.shifted_row_index, row.prev_tv_index});
new_candidates.push_back({row.shifted_row_index, row.tv_index}); new_candidates.push_back({row.shifted_row_index, row.tv_index});
// }
} }
std::sort(begin(old_candidates), end(old_candidates), [](auto a, auto b) { std::sort(begin(old_candidates), end(old_candidates), [](auto a, auto b) {
@ -477,7 +474,7 @@ void calculate_moves_sorted(std::vector<RowInfo>& new_rows, CollectionChangeIndi
find_longest_matches(old_candidates, b_ndx, find_longest_matches(old_candidates, b_ndx,
first_difference, old_candidates.size(), first_difference, old_candidates.size(),
first_difference, new_candidates.size(), first_difference, new_candidates.size(),
longest_matches); changeset.modifications, longest_matches);
longest_matches.push_back({old_candidates.size(), new_candidates.size(), 0}); longest_matches.push_back({old_candidates.size(), new_candidates.size(), 0});
size_t i = first_difference, j = first_difference; size_t i = first_difference, j = first_difference;
@ -553,5 +550,22 @@ CollectionChangeIndices CollectionChangeIndices::calculate(std::vector<size_t> c
} }
ret.verify(); ret.verify();
#ifdef REALM_DEBUG
{ // Verify that applying the calculated change to prev_rows actually produces next_rows
auto rows = prev_rows;
auto it = std::make_reverse_iterator(ret.deletions.end());
auto end = std::make_reverse_iterator(ret.deletions.begin());
for (; it != end; ++it) {
rows.erase(rows.begin() + it->first, rows.begin() + it->second);
}
for (auto i : ret.insertions.as_indexes()) {
rows.insert(rows.begin() + i, next_rows[i]);
}
REALM_ASSERT(rows == next_rows);
}
#endif
return ret; return ret;
} }

View File

@ -133,6 +133,7 @@ void AsyncQuery::run()
m_changes = CollectionChangeIndices::calculate(m_previous_rows, next_rows, m_changes = CollectionChangeIndices::calculate(m_previous_rows, next_rows,
[&](size_t row) { return m_info->row_did_change(*m_query->get_table(), row); }, [&](size_t row) { return m_info->row_did_change(*m_query->get_table(), row); },
!!m_sort); !!m_sort);
m_previous_rows = std::move(next_rows); m_previous_rows = std::move(next_rows);
if (m_changes.empty()) { if (m_changes.empty()) {
m_tv = {}; m_tv = {};

View File

@ -36,6 +36,17 @@ bool IndexSet::contains(size_t index) const
return it != m_ranges.end() && it->first <= index; return it != m_ranges.end() && it->first <= index;
} }
size_t IndexSet::count(size_t start_index, size_t end_index) const
{
auto it = const_cast<IndexSet*>(this)->find(start_index);
size_t ret = 0;
for (; end_index > start_index && it != m_ranges.end() && it->first < end_index; ++it) {
ret += std::min(it->second, end_index) - std::max(it->first, start_index);
start_index = it->second;
}
return ret;
}
IndexSet::iterator IndexSet::find(size_t index) IndexSet::iterator IndexSet::find(size_t index)
{ {
return find(index, m_ranges.begin()); return find(index, m_ranges.begin());

View File

@ -44,6 +44,9 @@ public:
// Check if the index set contains the given index // Check if the index set contains the given index
bool contains(size_t index) const; bool contains(size_t index) const;
// Counts the number of indices in the set in the given range
size_t count(size_t start_index, size_t end_index) const;
// Add an index to the set, doing nothing if it's already present // Add an index to the set, doing nothing if it's already present
void add(size_t index); void add(size_t index);
void add(IndexSet const& is); void add(IndexSet const& is);

View File

@ -242,6 +242,29 @@ TEST_CASE("collection change indices") {
REQUIRE_MOVES(calc({3, 1, 2}), {2, 0}); REQUIRE_MOVES(calc({3, 1, 2}), {2, 0});
REQUIRE_MOVES(calc({3, 2, 1}), {2, 0}, {1, 1}); REQUIRE_MOVES(calc({3, 2, 1}), {2, 0}, {1, 1});
} }
SECTION("merge can collapse insert -> move -> delete to no-op") {
auto four_modified = [](size_t ndx) { return ndx == 4; };
for (int insert_pos = 0; insert_pos < 4; ++insert_pos) {
for (int move_to_pos = 0; move_to_pos < 4; ++move_to_pos) {
if (insert_pos == move_to_pos)
continue;
CAPTURE(insert_pos);
CAPTURE(move_to_pos);
std::vector<size_t> after_insert = {1, 2, 3};
after_insert.insert(after_insert.begin() + insert_pos, 4);
c = CollectionChangeIndices::calculate({1, 2, 3}, after_insert, four_modified, true);
std::vector<size_t> after_move = {1, 2, 3};
after_move.insert(after_move.begin() + move_to_pos, 4);
c.merge(CollectionChangeIndices::calculate(after_insert, after_move, four_modified, true));
c.merge(CollectionChangeIndices::calculate(after_move, {1, 2, 3}, four_modified, true));
REQUIRE(c.empty());
}
}
}
} }
SECTION("merge") { SECTION("merge") {

View File

@ -7,6 +7,36 @@
TEST_CASE("index set") { TEST_CASE("index set") {
realm::IndexSet set; realm::IndexSet set;
SECTION("contains() returns if the index is in the set") {
set = {1, 2, 3, 5};
REQUIRE_FALSE(set.contains(0));
REQUIRE(set.contains(1));
REQUIRE(set.contains(2));
REQUIRE(set.contains(3));
REQUIRE_FALSE(set.contains(4));
REQUIRE(set.contains(5));
REQUIRE_FALSE(set.contains(6));
}
SECTION("count() returns the number of indices int he range in the set") {
set = {1, 2, 3, 5};
REQUIRE(set.count(0, 6) == 4);
REQUIRE(set.count(0, 5) == 3);
REQUIRE(set.count(0, 4) == 3);
REQUIRE(set.count(0, 3) == 2);
REQUIRE(set.count(0, 2) == 1);
REQUIRE(set.count(0, 1) == 0);
REQUIRE(set.count(0, 0) == 0);
REQUIRE(set.count(0, 6) == 4);
REQUIRE(set.count(1, 6) == 4);
REQUIRE(set.count(2, 6) == 3);
REQUIRE(set.count(3, 6) == 2);
REQUIRE(set.count(4, 6) == 1);
REQUIRE(set.count(5, 6) == 1);
REQUIRE(set.count(6, 6) == 0);
}
SECTION("add() extends existing ranges") { SECTION("add() extends existing ranges") {
set.add(1); set.add(1);
REQUIRE_INDICES(set, 1); REQUIRE_INDICES(set, 1);