emory: compute possible size reduction from deduplication

cruft/util

@@ -1 +1 @@
-Subproject commit e1d56395f6b16cb70f8dbb942a117dc01fd7e74c
+Subproject commit cc9b9b19c1a9a7002248d7f62053d49002a8b297

emory.cpp

@@ -8,33 +8,108 @@
 
 #include <cruft/util/hash/buzhash.hpp>
 #include <cruft/util/io.hpp>
+#include <cruft/util/parse/value.hpp>
 
+#include <cruft/crypto/hash/sha1.hpp>
+
+#include <algorithm>
 #include <iostream>
 
+
+struct chunk {
+    cruft::view<u08 const*> data;
+    cruft::crypto::hash::SHA1::digest_t digest;
+};
+
+struct digest_equality {
+    bool operator() (chunk const &a, chunk const &b)
+    {
+        return a.digest == b.digest;
+    }
+};
+
+
+enum {
+    ARG_SELF,
+    ARG_BITS,
+    ARG_WINDOW,
+    ARGS_INPUT,
+
+    NUM_ARGS,
+};
+
+
 int main (int argc, char const **argv)
 {
-    (void)argc;
-
-    cruft::mapped_file src (argv[1]);
-    cruft::view bytes (src);
-
-    static constexpr std::size_t BITS = 16;
-    std::vector<std::size_t> counts (BITS, 0);
-
-    static constexpr std::size_t WINDOW = 48;
-    cruft::hash::buzhash<u64> h (WINDOW, bytes);
-
-    for (auto const &val: bytes.consume (WINDOW)) {
-        auto const res = h (&val);
-
-        std::size_t mask = ~u64(0) >> (64 - BITS);
-        for (std::size_t i = 0; i < BITS; ++i) {
-            if ((res & mask) == 0)
-                counts[i]++;
-            mask >>= 1;
-        }
+    if (argc != NUM_ARGS) {
+        std::cerr << "usage: " << argv[ARG_SELF] << " <bits> <window> <input>\n";
+        return EXIT_FAILURE;
     }
 
-    for (auto const &i: counts)
-        std::cout << i << '\n';
+    auto const window = cruft::parse::from_string<std::size_t> (argv[ARG_WINDOW]);
+    auto const bits   = cruft::parse::from_string<std::size_t> (argv[ARG_BITS  ]);
+
+    cruft::mapped_file src (argv[ARGS_INPUT]);
+    cruft::view bytes (src);
+
+    using hash_type = cruft::hash::buzhash<u64>;
+    hash_type h (window, bytes);
+    bytes = bytes.consume (window);
+
+    using digest_type = hash_type::digest_type ;
+    digest_type const mask = ~digest_type (0) >> (sizeof (digest_type) * 8 - bits);
+
+    std::vector<chunk> chunks;
+
+    for (u08 const *cursor = bytes.begin (), *start = src.data (); cursor != bytes.end (); cursor++) {
+        auto const res = h (cursor);
+
+        if (res & mask)
+            continue;
+
+        cruft::view<u08 const*> const region { start, cursor };
+        start = cursor + 1;
+
+        chunks.push_back ({
+            .data = region,
+            .digest = cruft::crypto::hash::SHA1 {} (region),
+        });
+    }
+
+    std::sort (
+        std::begin (chunks),
+        std::end   (chunks),
+        [] (auto const &a, auto const &b)
+    {
+        return a.digest < b.digest;
+    });
+
+    std::size_t reduction = 0;
+    for (auto cursor = chunks.begin (); cursor != chunks.end (); ++cursor) {
+        auto first_match = std::adjacent_find (
+            cursor, chunks.end (),
+            digest_equality {}
+        );
+
+        if (first_match == chunks.end ())
+            break;
+
+        auto last_match = std::find_if_not (
+            first_match,
+            chunks.end (),
+            [&first_match] (auto const &i)
+        {
+            return i.digest == first_match->digest;
+        });
+
+        auto const count = std::distance (first_match, last_match);
+        auto const size  = first_match->data.size ();
+        std::cout << count << " duplicates of " << size << " bytes\n";
+        reduction += (count - 1) * size;
+
+        cursor = last_match;
+    }
+
+    std::cout << chunks.size () << " chunks found\n";
+    std::cout << float (reduction) / src.size () << " reduction\n";
 }