AK: Add CircularBuffer::find_copy_in_seekback()

This is useful for compressors, which quite frequently need to find a
matching span of data within the seekback.

AK/CircularBuffer.cpp

@@ -249,4 +249,107 @@ ErrorOr<size_t> CircularBuffer::copy_from_seekback(size_t distance, size_t lengt
     return length - remaining_length;
 }
 
+ErrorOr<Vector<CircularBuffer::Match>> CircularBuffer::find_copy_in_seekback(size_t maximum_length, size_t minimum_length, Optional<Vector<size_t> const&> distance_hints) const
+{
+    VERIFY(minimum_length > 0);
+
+    // Clip the maximum length to the amount of data that we actually store.
+    if (maximum_length > m_used_space)
+        maximum_length = m_used_space;
+
+    if (maximum_length < minimum_length)
+        return Vector<Match> {};
+
+    Vector<Match> matches;
+
+    if (distance_hints.has_value()) {
+        // If we have any hints, verify and use those.
+        for (auto const& distance : distance_hints.value()) {
+            // TODO: This does not yet support looping repetitions.
+            if (distance < minimum_length)
+                continue;
+
+            auto needle_offset = (capacity() + m_reading_head) % capacity();
+            auto haystack_offset = (capacity() + m_reading_head - distance) % capacity();
+
+            for (size_t i = 0; i < minimum_length; i++) {
+                if (m_buffer[needle_offset] != m_buffer[haystack_offset])
+                    break;
+
+                needle_offset = (needle_offset + 1) % capacity();
+                haystack_offset = (haystack_offset + 1) % capacity();
+
+                if (i + 1 == minimum_length)
+                    TRY(matches.try_empend(distance, minimum_length));
+            }
+        }
+    } else {
+        // Otherwise, use memmem to find the initial matches.
+        // Note: We have the read head as our reference point, but `next_read_span_with_seekback` isn't aware of that and continues to use the write head.
+        //       Therefore, we need to make sure to slice off the extraneous bytes from the end of the span and shift the returned distances by the correct amount.
+        size_t haystack_offset_from_start = 0;
+        Vector<ReadonlyBytes, 2> haystack;
+        haystack.append(next_read_span_with_seekback(m_seekback_limit));
+        if (haystack[0].size() < m_seekback_limit - used_space())
+            haystack.append(next_read_span_with_seekback(m_seekback_limit - haystack[0].size()));
+
+        haystack.last() = haystack.last().trim(haystack.last().size() - used_space());
+
+        auto needle = next_read_span().trim(minimum_length);
+
+        auto memmem_match = AK::memmem(haystack.begin(), haystack.end(), needle);
+        while (memmem_match.has_value()) {
+            auto match_offset = memmem_match.release_value();
+
+            // Add the match to the list of matches to work with.
+            TRY(matches.try_empend(m_seekback_limit - used_space() - haystack_offset_from_start - match_offset, minimum_length));
+
+            auto size_to_discard = match_offset + 1;
+
+            // Trim away the already processed bytes from the haystack.
+            haystack_offset_from_start += size_to_discard;
+            while (size_to_discard > 0) {
+                if (haystack[0].size() < size_to_discard) {
+                    size_to_discard -= haystack[0].size();
+                    haystack.remove(0);
+                } else {
+                    haystack[0] = haystack[0].slice(size_to_discard);
+                    break;
+                }
+            }
+
+            if (haystack.size() == 0)
+                break;
+
+            // Try and find the next match.
+            memmem_match = AK::memmem(haystack.begin(), haystack.end(), needle);
+        }
+    }
+
+    // From now on, all matches that we have stored have at least a length of `minimum_length` and they all refer to the same value.
+    // For the remaining part, we will keep checking the next byte incrementally and keep eliminating matches until we eliminated all of them.
+    Vector<Match> next_matches;
+
+    for (size_t offset = minimum_length; offset < maximum_length; offset++) {
+        auto needle_data = m_buffer[(capacity() + m_reading_head + offset) % capacity()];
+
+        for (auto const& match : matches) {
+            auto haystack_data = m_buffer[(capacity() + m_reading_head - match.distance + offset) % capacity()];
+
+            if (haystack_data != needle_data)
+                continue;
+
+            TRY(next_matches.try_empend(match.distance, match.length + 1));
+        }
+
+        if (next_matches.size() == 0)
+            return matches;
+
+        swap(matches, next_matches);
+        next_matches.clear_with_capacity();
+    }
+
+    return matches;
+}
+
 }

AK/CircularBuffer.h

@@ -9,6 +9,7 @@
 #include <AK/ByteBuffer.h>
 #include <AK/Error.h>
 #include <AK/Noncopyable.h>
+#include <AK/Vector.h>
 
 namespace AK {
 
@@ -36,6 +37,15 @@ public:
 
     ErrorOr<size_t> copy_from_seekback(size_t distance, size_t length);
 
+    struct Match {
+        size_t distance;
+        size_t length;
+    };
+    /// This searches the seekback buffer (between read head and limit) for occurrences where it matches the next `length` bytes from the read buffer.
+    /// Supplying any hints will only consider those distances, in case existing offsets need to be validated.
+    /// Note that, since we only start searching at the read head, the length between read head and write head is excluded from the distance.
+    ErrorOr<Vector<Match>> find_copy_in_seekback(size_t maximum_length, size_t minimum_length = 2, Optional<Vector<size_t> const&> distance_hints = {}) const;
+
     [[nodiscard]] size_t empty_space() const;
     [[nodiscard]] size_t used_space() const;
     [[nodiscard]] size_t capacity() const;

Tests/AK/TestCircularBuffer.cpp

@@ -329,6 +329,116 @@ TEST_CASE(offset_of_with_until_and_after_wrapping_around)
     EXPECT_EQ(result.value_or(42), 14ul);
 }
 
+TEST_CASE(find_copy_in_seekback)
+{
+    auto haystack = "ABABCABCDAB"sv.bytes();
+    auto needle = "ABCD"sv.bytes();
+
+    // Set up the buffer for testing.
+    auto buffer = MUST(CircularBuffer::create_empty(haystack.size() + needle.size()));
+    auto written_haystack_bytes = buffer.write(haystack);
+    VERIFY(written_haystack_bytes == haystack.size());
+    MUST(buffer.discard(haystack.size()));
+    auto written_needle_bytes = buffer.write(needle);
+    VERIFY(written_needle_bytes == needle.size());
+
+    {
+        // Find the largest matches with a length between 1 and 1 (all "A").
+        auto matches = MUST(buffer.find_copy_in_seekback(1, 1));
+        EXPECT_EQ(matches.size(), 4ul);
+        EXPECT_EQ(matches[0].distance, 11ul);
+        EXPECT_EQ(matches[0].length, 1ul);
+        EXPECT_EQ(matches[1].distance, 9ul);
+        EXPECT_EQ(matches[1].length, 1ul);
+        EXPECT_EQ(matches[2].distance, 6ul);
+        EXPECT_EQ(matches[2].length, 1ul);
+        EXPECT_EQ(matches[3].distance, 2ul);
+        EXPECT_EQ(matches[3].length, 1ul);
+    }
+
+    {
+        // Find the largest matches with a length between 1 and 2 (all "AB", everything smaller gets eliminated).
+        auto matches = MUST(buffer.find_copy_in_seekback(2, 1));
+        EXPECT_EQ(matches.size(), 4ul);
+        EXPECT_EQ(matches[0].distance, 11ul);
+        EXPECT_EQ(matches[0].length, 2ul);
+        EXPECT_EQ(matches[1].distance, 9ul);
+        EXPECT_EQ(matches[1].length, 2ul);
+        EXPECT_EQ(matches[2].distance, 6ul);
+        EXPECT_EQ(matches[2].length, 2ul);
+        EXPECT_EQ(matches[3].distance, 2ul);
+        EXPECT_EQ(matches[3].length, 2ul);
+    }
+
+    {
+        // Find the largest matches with a length between 1 and 3 (all "ABC", everything smaller gets eliminated).
+        auto matches = MUST(buffer.find_copy_in_seekback(3, 1));
+        EXPECT_EQ(matches.size(), 2ul);
+        EXPECT_EQ(matches[0].distance, 9ul);
+        EXPECT_EQ(matches[0].length, 3ul);
+        EXPECT_EQ(matches[1].distance, 6ul);
+        EXPECT_EQ(matches[1].length, 3ul);
+    }
+
+    {
+        // Find the largest matches with a length between 1 and 4 (all "ABCD", everything smaller gets eliminated).
+        auto matches = MUST(buffer.find_copy_in_seekback(4, 1));
+        EXPECT_EQ(matches.size(), 1ul);
+        EXPECT_EQ(matches[0].distance, 6ul);
+        EXPECT_EQ(matches[0].length, 4ul);
+    }
+
+    {
+        // Find the largest matches with a length between 1 and 5 (all "ABCD", everything smaller gets eliminated, and nothing larger exists).
+        auto matches = MUST(buffer.find_copy_in_seekback(5, 1));
+        EXPECT_EQ(matches.size(), 1ul);
+        EXPECT_EQ(matches[0].distance, 6ul);
+        EXPECT_EQ(matches[0].length, 4ul);
+    }
+
+    {
+        // Find the largest matches with a length between 4 and 5 (all "ABCD", everything smaller never gets found, nothing larger exists).
+        auto matches = MUST(buffer.find_copy_in_seekback(5, 4));
+        EXPECT_EQ(matches.size(), 1ul);
+        EXPECT_EQ(matches[0].distance, 6ul);
+        EXPECT_EQ(matches[0].length, 4ul);
+    }
+
+    {
+        // Find the largest matches with a length between 5 and 5 (nothing is found).
+        auto matches = MUST(buffer.find_copy_in_seekback(5, 5));
+        EXPECT_EQ(matches.size(), 0ul);
+    }
+
+    {
+        // Find the largest matches with a length between 1 and 2 (selected "AB", everything smaller gets eliminated).
+        auto matches = MUST(buffer.find_copy_in_seekback(2, 1, Vector<size_t> { 6ul, 9ul }));
+        EXPECT_EQ(matches.size(), 2ul);
+        EXPECT_EQ(matches[0].distance, 6ul);
+        EXPECT_EQ(matches[0].length, 2ul);
+        EXPECT_EQ(matches[1].distance, 9ul);
+        EXPECT_EQ(matches[1].length, 2ul);
+    }
+
+    {
+        // Check that we don't find anything for hints before the valid range.
+        auto matches = MUST(buffer.find_copy_in_seekback(2, 1, Vector<size_t> { 0ul }));
+        EXPECT_EQ(matches.size(), 0ul);
+    }
+
+    {
+        // Check that we don't find anything for hints after the valid range.
+        auto matches = MUST(buffer.find_copy_in_seekback(2, 1, Vector<size_t> { 12ul }));
+        EXPECT_EQ(matches.size(), 0ul);
+    }
+
+    {
+        // Check that we don't find anything for a minimum length beyond the whole buffer size.
+        auto matches = MUST(buffer.find_copy_in_seekback(12, 13));
+        EXPECT_EQ(matches.size(), 0ul);
+    }
+}
+
 BENCHMARK_CASE(looping_copy_from_seekback)
 {
     auto circular_buffer = MUST(CircularBuffer::create_empty(16 * MiB));