From f0be855e7c535a02219b77f98dc3d2f14f834f4c Mon Sep 17 00:00:00 2001
From: Joe Jevnik <joe@quantopian.com>
Date: Mon, 9 Mar 2020 21:42:42 -0400
Subject: [PATCH] ENH: don't vendor the hash implementation

---
 include/libpy/hash.h | 52 ++++----------------------------------------
 tests/test_hash.cc   |  6 -----
 2 files changed, 4 insertions(+), 54 deletions(-)

diff --git a/include/libpy/hash.h b/include/libpy/hash.h
index b98ab7f7..a24a9d7c 100644
--- a/include/libpy/hash.h
+++ b/include/libpy/hash.h
@@ -1,6 +1,7 @@
 #pragma once
 
 #include <array>
+#include <string_view>
 #include <tuple>
 
 namespace py {
@@ -35,59 +36,14 @@ auto hash_tuple(const std::tuple<Ts...>& t) {
     return std::apply<decltype(hash_many<Ts...>)>(hash_many, t);
 }
 
-namespace detail {
-inline std::size_t unaligned_load(const char* p) {
-    std::size_t result;
-    __builtin_memcpy(&result, p, sizeof(result));
-    return result;
-}
-
-// Loads n bytes, where 1 <= n < 8.
-inline std::size_t load_bytes(const char* p, int n) {
-    std::size_t result = 0;
-    --n;
-    do {
-        result = (result << 8) + static_cast<unsigned char>(p[n]);
-    } while (--n >= 0);
-    return result;
-}
-
-inline std::size_t shift_mix(std::size_t v) {
-    return v ^ (v >> 47);
-}
-
-constexpr std::size_t hash_seed = static_cast<std::size_t>(0xc70f6907UL);
-constexpr std::size_t hash_mul = (static_cast<std::size_t>(0xc6a4a793UL) << 32UL) +
-                                 static_cast<std::size_t>(0x5bd1e995UL);
-}  // namespace detail
-
-/** Hash a buffer of characters using the same algorithm a libstdc++
-    `std::hash<std::string>`.
+/** Hash a buffer of characters using the same algorithm as
+    `std::hash<std::string_view>`
 
     @param buf The buffer to hash.
     @param len The length of the buffer.
     @return The hash of the string.
  */
 inline std::size_t hash_buffer(const char* buf, std::size_t len) {
-    // Remove the bytes not divisible by the sizeof(size_t).  This
-    // allows the main loop to process the data as 64-bit integers.
-    const int len_aligned = len & ~0x7;
-    const char* const end = buf + len_aligned;
-    std::size_t hash = detail::hash_seed ^ (len * detail::hash_mul);
-    for (const char* p = buf; p != end; p += 8) {
-        const std::size_t data = detail::shift_mix(detail::unaligned_load(p) *
-                                                   detail::hash_mul) *
-                                 detail::hash_mul;
-        hash ^= data;
-        hash *= detail::hash_mul;
-    }
-    if ((len & 0x7) != 0) {
-        const std::size_t data = detail::load_bytes(end, len & 0x7);
-        hash ^= data;
-        hash *= detail::hash_mul;
-    }
-    hash = detail::shift_mix(hash) * detail::hash_mul;
-    hash = detail::shift_mix(hash);
-    return hash;
+    return std::hash<std::string_view>{}(std::string_view{buf, len});
 }
 }  // namespace py
diff --git a/tests/test_hash.cc b/tests/test_hash.cc
index c8f4e96f..5c586a89 100644
--- a/tests/test_hash.cc
+++ b/tests/test_hash.cc
@@ -18,17 +18,11 @@ std::string random_string(RandomEngine& g) {
 }
 
 TEST(hash, buffer) {
-#if !defined(_LIBCPP_VERSION)
-    // XXX: This asserts that `py::hash_buffer` does the same thing as
-    // `std::hash<std::string>`; however, it is a copy of the libstdc++
-    // algorithm. libc++ uses a different hash algorithm which doesn't produce
-    // the same results.
     std::mt19937 g(1868655980);
 
     for (std::size_t n = 0; n < 1000; ++n) {
         std::string s = random_string(g);
         EXPECT_EQ(std::hash<std::string>{}(s), py::hash_buffer(s.data(), s.size()));
     }
-#endif
 }
 }  // namespace test_hash