[Courgette] Refactor BSDiff namespaces and bsdiff::search() interface.

courgette/third_party/bsdiff/bsdiff_search.h

+/*
+  bsdiff_search.h -- Suffix array search.
+
+  Copyright 2003 Colin Percival
+
+  For the terms under which this work may be distributed, please see
+  the adjoining file "LICENSE".
+
+  ChangeLog:
+  2005-05-05 - Use the modified header struct from bspatch.h; use 32-bit
+               values throughout.
+                 --Benjamin Smedberg <benjamin@smedbergs.us>
+  2015-08-03 - Change search() to template to allow PagedArray usage.
+                 --Samuel Huang <huangs@chromium.org>
+  2015-08-19 - Optimized search().
+                 --Samuel Huang <huangs@chromium.org>
+  2016-06-02 - Move matchlen() and search() to a separate file; comment; format.
+                 --Samuel Huang <huangs@chromium.org>
+*/
+
+#include <algorithm>
+#include <cstring>
+
+namespace bsdiff {
+
+// Similar to ::memcmp(), but returns match length instead.
+inline int matchlen(const unsigned char *srcbuf, const unsigned char *keybuf,
+                    int size) {
+  for (int i = 0; i < size; ++i)
+    if (srcbuf[i] != keybuf[i])
+      return i;
+  return size;
+}
+
+// Finds a suffix in |srcbuf| that has the longest common prefix with |keybuf|,
+// aided by suffix array |I| of |srcbuf|. Returns the length of the match, and
+// writes the position within |srcbuf| to |pos|.
+template <class T>
+int search(T I, const unsigned char *srcbuf, int srcize,
+           const unsigned char *keybuf, int keysize, int *pos) {
+  // We use binary search here, but the desired position is not the usual
+  // position from std::lower_bound()-like binary search.
+  // For example, given sorted suffixes of |srcbuf|:
+  //   SAAA...
+  //   SNNN...
+  //   TAAA...
+  // Applying std::lower_bound() binary search results in:
+  //   |keybuf| = "SNBB" => "SNNN...".
+  //   |keybuf| = "SNZZ" => "TAAA...".
+  // But the desired result for both cases is "SNNN...", with "SN" as the
+  // longest common prefix. To solve this:
+  // Case 1: |keysize| == 0: We'd return 0, although |pos| is arbitrary.
+  // Case 2: |srcsize| == 0: We'd return 0, and |pos| = 0.
+  // Case 3: |srcsize| > 0 and |keysize| > 0: We'd assert rank-|lo| suffix
+  //   of |srcbuf| is always strictly lexicographically less than |keybuf|. This
+  //   holds for |lo| = 0, which corresponds to "". The search terminates when
+  //   hi == lo + 1. At this point, suffixes at rank |lo| and rank |hi| both
+  //   may be the desired match, so we check both. In case of tie, there's NO
+  //   guarantee that we take the solution with minimum rank.
+  // Finding the minimum rank solution is not needed, and it would create more
+  // complexity. For example, with suffixes "MA...", "MB...", "MC...", "ME...",
+  // searching look for "MD" would yield |lo| -> "MC...", and the minimum rank
+  // solution "MA...", would be rather far away.
+  //
+  // TODO(huangs): Optimize this: match length at |t| is >= match length at
+  //   |lo| and at |hi|. If we track this then we can skip comparisons!
+  int lo = 0;
+  int hi = srcize;
+  while (hi - lo > 1) {
+    int t = (lo + hi) / 2;
+    if (::memcmp(srcbuf + I[t], keybuf, std::min(srcize - I[t], keysize)) < 0)
+      lo = t;
+    else
+      hi = t;
+  }
+
+  int x = matchlen(srcbuf + I[lo], keybuf, std::min(srcize - I[lo], keysize));
+  int y = matchlen(srcbuf + I[hi], keybuf, std::min(srcize - I[hi], keysize));
+  if (x > y) {
+    *pos = I[lo];
+    return x;
+  }
+  *pos = I[hi];
+  return y;
+}
+
+}  // namespace bsdiff