about summary refs log tree commit diff
diff options
context:
space:
mode:
authorAbseil Team <absl-team@google.com>2020-01-16T21·38-0800
committerXiaoyi Zhang <zhangxy@google.com>2020-01-21T16·47-0500
commit159bf2bf6d1cc8087e02468d071e94d1177d1bae (patch)
tree5ba7def08ad7032867f629eaf431c496efbd24b3
parenta2e6adecc294dc4cd98cc285a9134ce58e0f2ad0 (diff)
Export of internal Abseil changes
--
c42a234e2c186bf697ce8d77e85628601fa514a6 by Abseil Team <absl-team@google.com>:

Enable the assertion in the iterator's operator++

PiperOrigin-RevId: 290134813

--
f8c53ba8e9c5bb16bbcc1e412a5c2519c912c83e by Abseil Team <absl-team@google.com>:

Define operator== and operator!= for absl::{weak,strong}_equality and
absl::{partial,weak,strong}_ordering types themselves.

PiperOrigin-RevId: 290111564

--
36bc574090cefad74a451719ce2761982647a51d by Tom Manshreck <shreck@google.com>:

Specify Time library flag formats

PiperOrigin-RevId: 289928010

--
26dd40281add260baab2b60fec05dfb9c5304aaa by Mark Barolak <mbar@google.com>:

Delete an extraneous forward declaration of absl::Cord.

PiperOrigin-RevId: 289708481

--
e60aea7f33554ff66d7699bb70e7af1d26323f1d by Abseil Team <absl-team@google.com>:

Release b-tree benchmarks.

PiperOrigin-RevId: 289654429

--
660aa83fa000d4bae072b2d1c790f81d0939bc7e by Greg Falcon <gfalcon@google.com>:

Use https links.

Import of https://github.com/abseil/abseil-cpp/pull/586

PiperOrigin-RevId: 289479559

--
0611ea4482dcf23d6b0a0389fe041eeb9052449a by Derek Mauro <dmauro@google.com>:

Removes the static initializer for LookupTables<absl::uint128>::kVmaxOverBase

Uses template specialization to hard code the resulting array.

Static initializers are problematic for a number of reasons. Not only
are they responsible for the static initialization order fiasco, but
they are in the critical path during program startup. For these
reasons, the Google C++ style guide strongly discourages them (and
forbids them when they are not trivially destructible), and Chromium
even has a test forbidding them.

https://google.github.io/styleguide/cppguide.html#Static_and_Global_Variables
https://chromium.googlesource.com/chromium/src.git/+/master/docs/static_initializers.md
http://neugierig.org/software/chromium/notes/2011/08/static-initializers.html

PiperOrigin-RevId: 289458677

--
c869362f6bb7a872314f74750d38d81bdaa73f95 by Greg Falcon <gfalcon@google.com>:

Step 2 of 2 to fix our CCTZ fork to respect inline namespaces.

Re-import of CCTZ from GitHub, applying new changes to honor Abseil's optional inline namespace in MSVC.

PiperOrigin-RevId: 289454407

--
fdb3474d76c2ee0371ccdf7593a78137c03a3f58 by Greg Falcon <gfalcon@google.com>:

Step 1 of 2 to fix our CCTZ fork to respect inline namespaces.

CCTZ uses a linker flag to simulate weak symbol support in MSVC.  This takes the form of a #pragma that includes the mangled names of two types: the symbol to treat as weak, and the symbol to use as its default value if no override is provided.

When Abseil is configured to use inline namespaces, the mangled names of these symbols change, and the pragma should change to reflect that.  Fortunately for us, MSVC name mangling is simple enough that we can generate the needed string literals in the preprocessor.

This CL introduces the new macros; the uses will be introduced in a follow-up CL.

PiperOrigin-RevId: 289435599

--
5f152cc36f008acb9ab78f30b5efa40ebaf2754b by Matt Kulukundis <kfm@google.com>:

Improve documentation for lazy_emplace

PiperOrigin-RevId: 289333112
GitOrigin-RevId: c42a234e2c186bf697ce8d77e85628601fa514a6
Change-Id: I139ce6c7044a70d083af53e428bcb987f0fd88c6
-rw-r--r--absl/base/config.h28
-rw-r--r--absl/container/BUILD.bazel26
-rw-r--r--absl/container/btree_benchmark.cc707
-rw-r--r--absl/container/internal/raw_hash_set.h15
-rw-r--r--absl/flags/marshalling.h6
-rw-r--r--absl/strings/internal/str_format/extension.h3
-rw-r--r--absl/strings/numbers.cc52
-rw-r--r--absl/time/internal/cctz/src/zone_info_source.cc40
-rw-r--r--absl/time/time.h76
-rw-r--r--absl/types/compare.h40
-rw-r--r--absl/types/compare_test.cc92
11 files changed, 1025 insertions, 60 deletions
diff --git a/absl/base/config.h b/absl/base/config.h
index edbf22468ec5..c4e8dce403bb 100644
--- a/absl/base/config.h
+++ b/absl/base/config.h
@@ -66,6 +66,10 @@
 #include "absl/base/options.h"
 #include "absl/base/policy_checks.h"
 
+// Helper macro to convert a CPP variable to a string literal.
+#define ABSL_INTERNAL_DO_TOKEN_STR(x) #x
+#define ABSL_INTERNAL_TOKEN_STR(x) ABSL_INTERNAL_DO_TOKEN_STR(x)
+
 // -----------------------------------------------------------------------------
 // Abseil namespace annotations
 // -----------------------------------------------------------------------------
@@ -98,8 +102,6 @@
 // Check that ABSL_OPTION_INLINE_NAMESPACE_NAME is neither "head" nor ""
 #if defined(__cplusplus) && ABSL_OPTION_USE_INLINE_NAMESPACE == 1
 
-#define ABSL_INTERNAL_DO_TOKEN_STR(x) #x
-#define ABSL_INTERNAL_TOKEN_STR(x) ABSL_INTERNAL_DO_TOKEN_STR(x)
 #define ABSL_INTERNAL_INLINE_NAMESPACE_STR \
   ABSL_INTERNAL_TOKEN_STR(ABSL_OPTION_INLINE_NAMESPACE_NAME)
 
@@ -617,6 +619,28 @@ static_assert(ABSL_INTERNAL_INLINE_NAMESPACE_STR[0] != 'h' ||
 #define ABSL_INTERNAL_MSVC_2017_DBG_MODE
 #endif
 
+// ABSL_INTERNAL_MANGLED_NS
+// ABSL_INTERNAL_MANGLED_BACKREFERENCE
+//
+// Internal macros for building up mangled names in our internal fork of CCTZ.
+// This implementation detail is only needed and provided for the MSVC build.
+//
+// These macros both expand to string literals.  ABSL_INTERNAL_MANGLED_NS is
+// the mangled spelling of the `absl` namespace, and
+// ABSL_INTERNAL_MANGLED_BACKREFERENCE is a back-reference integer representing
+// the proper count to skip past the CCTZ fork namespace names.  (This number
+// is one larger when there is an inline namespace name to skip.)
+#if defined(_MSC_VER)
+#if ABSL_OPTION_USE_INLINE_NAMESPACE == 0
+#define ABSL_INTERNAL_MANGLED_NS "absl"
+#define ABSL_INTERNAL_MANGLED_BACKREFERENCE "5"
+#else
+#define ABSL_INTERNAL_MANGLED_NS \
+  ABSL_INTERNAL_TOKEN_STR(ABSL_OPTION_INLINE_NAMESPACE_NAME) "@absl"
+#define ABSL_INTERNAL_MANGLED_BACKREFERENCE "6"
+#endif
+#endif
+
 #undef ABSL_INTERNAL_HAS_KEYWORD
 
 #endif  // ABSL_BASE_CONFIG_H_
diff --git a/absl/container/BUILD.bazel b/absl/container/BUILD.bazel
index 1f7abe07ff07..f221714027ac 100644
--- a/absl/container/BUILD.bazel
+++ b/absl/container/BUILD.bazel
@@ -874,3 +874,29 @@ cc_test(
         "@com_google_googletest//:gtest_main",
     ],
 )
+
+cc_binary(
+    name = "btree_benchmark",
+    testonly = 1,
+    srcs = [
+        "btree_benchmark.cc",
+    ],
+    copts = ABSL_TEST_COPTS,
+    linkopts = ABSL_DEFAULT_LINKOPTS,
+    tags = ["benchmark"],
+    visibility = ["//visibility:private"],
+    deps = [
+        ":btree",
+        ":btree_test_common",
+        ":flat_hash_map",
+        ":flat_hash_set",
+        ":hashtable_debug",
+        "//absl/base:raw_logging_internal",
+        "//absl/flags:flag",
+        "//absl/hash",
+        "//absl/memory",
+        "//absl/strings:str_format",
+        "//absl/time",
+        "@com_github_google_benchmark//:benchmark_main",
+    ],
+)
diff --git a/absl/container/btree_benchmark.cc b/absl/container/btree_benchmark.cc
new file mode 100644
index 000000000000..4af92f9fd888
--- /dev/null
+++ b/absl/container/btree_benchmark.cc
@@ -0,0 +1,707 @@
+// Copyright 2018 The Abseil Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      https://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include <stdint.h>
+
+#include <algorithm>
+#include <functional>
+#include <map>
+#include <numeric>
+#include <random>
+#include <set>
+#include <string>
+#include <type_traits>
+#include <unordered_map>
+#include <unordered_set>
+#include <vector>
+
+#include "absl/base/internal/raw_logging.h"
+#include "absl/container/btree_map.h"
+#include "absl/container/btree_set.h"
+#include "absl/container/btree_test.h"
+#include "absl/container/flat_hash_map.h"
+#include "absl/container/flat_hash_set.h"
+#include "absl/container/internal/hashtable_debug.h"
+#include "absl/flags/flag.h"
+#include "absl/hash/hash.h"
+#include "absl/memory/memory.h"
+#include "absl/strings/str_format.h"
+#include "absl/time/time.h"
+#include "benchmark/benchmark.h"
+
+namespace absl {
+ABSL_NAMESPACE_BEGIN
+namespace container_internal {
+namespace {
+
+constexpr size_t kBenchmarkValues = 1 << 20;
+
+// How many times we add and remove sub-batches in one batch of *AddRem
+// benchmarks.
+constexpr size_t kAddRemBatchSize = 1 << 2;
+
+// Generates n values in the range [0, 4 * n].
+template <typename V>
+std::vector<V> GenerateValues(int n) {
+  constexpr int kSeed = 23;
+  return GenerateValuesWithSeed<V>(n, 4 * n, kSeed);
+}
+
+// Benchmark insertion of values into a container.
+template <typename T>
+void BM_InsertImpl(benchmark::State& state, bool sorted) {
+  using V = typename remove_pair_const<typename T::value_type>::type;
+  typename KeyOfValue<typename T::key_type, V>::type key_of_value;
+
+  std::vector<V> values = GenerateValues<V>(kBenchmarkValues);
+  if (sorted) {
+    std::sort(values.begin(), values.end());
+  }
+  T container(values.begin(), values.end());
+
+  // Remove and re-insert 10% of the keys per batch.
+  const int batch_size = (kBenchmarkValues + 9) / 10;
+  while (state.KeepRunningBatch(batch_size)) {
+    state.PauseTiming();
+    const auto i = static_cast<int>(state.iterations());
+
+    for (int j = i; j < i + batch_size; j++) {
+      int x = j % kBenchmarkValues;
+      container.erase(key_of_value(values[x]));
+    }
+
+    state.ResumeTiming();
+
+    for (int j = i; j < i + batch_size; j++) {
+      int x = j % kBenchmarkValues;
+      container.insert(values[x]);
+    }
+  }
+}
+
+template <typename T>
+void BM_Insert(benchmark::State& state) {
+  BM_InsertImpl<T>(state, false);
+}
+
+template <typename T>
+void BM_InsertSorted(benchmark::State& state) {
+  BM_InsertImpl<T>(state, true);
+}
+
+// container::insert sometimes returns a pair<iterator, bool> and sometimes
+// returns an iterator (for multi- containers).
+template <typename Iter>
+Iter GetIterFromInsert(const std::pair<Iter, bool>& pair) {
+  return pair.first;
+}
+template <typename Iter>
+Iter GetIterFromInsert(const Iter iter) {
+  return iter;
+}
+
+// Benchmark insertion of values into a container at the end.
+template <typename T>
+void BM_InsertEnd(benchmark::State& state) {
+  using V = typename remove_pair_const<typename T::value_type>::type;
+  typename KeyOfValue<typename T::key_type, V>::type key_of_value;
+
+  T container;
+  const int kSize = 10000;
+  for (int i = 0; i < kSize; ++i) {
+    container.insert(Generator<V>(kSize)(i));
+  }
+  V v = Generator<V>(kSize)(kSize - 1);
+  typename T::key_type k = key_of_value(v);
+
+  auto it = container.find(k);
+  while (state.KeepRunning()) {
+    // Repeatedly removing then adding v.
+    container.erase(it);
+    it = GetIterFromInsert(container.insert(v));
+  }
+}
+
+template <typename T>
+void BM_LookupImpl(benchmark::State& state, bool sorted) {
+  using V = typename remove_pair_const<typename T::value_type>::type;
+  typename KeyOfValue<typename T::key_type, V>::type key_of_value;
+
+  std::vector<V> values = GenerateValues<V>(kBenchmarkValues);
+  if (sorted) {
+    std::sort(values.begin(), values.end());
+  }
+  T container(values.begin(), values.end());
+
+  while (state.KeepRunning()) {
+    int idx = state.iterations() % kBenchmarkValues;
+    benchmark::DoNotOptimize(container.find(key_of_value(values[idx])));
+  }
+}
+
+// Benchmark lookup of values in a container.
+template <typename T>
+void BM_Lookup(benchmark::State& state) {
+  BM_LookupImpl<T>(state, false);
+}
+
+// Benchmark lookup of values in a full container, meaning that values
+// are inserted in-order to take advantage of biased insertion, which
+// yields a full tree.
+template <typename T>
+void BM_FullLookup(benchmark::State& state) {
+  BM_LookupImpl<T>(state, true);
+}
+
+// Benchmark deletion of values from a container.
+template <typename T>
+void BM_Delete(benchmark::State& state) {
+  using V = typename remove_pair_const<typename T::value_type>::type;
+  typename KeyOfValue<typename T::key_type, V>::type key_of_value;
+  std::vector<V> values = GenerateValues<V>(kBenchmarkValues);
+  T container(values.begin(), values.end());
+
+  // Remove and re-insert 10% of the keys per batch.
+  const int batch_size = (kBenchmarkValues + 9) / 10;
+  while (state.KeepRunningBatch(batch_size)) {
+    const int i = state.iterations();
+
+    for (int j = i; j < i + batch_size; j++) {
+      int x = j % kBenchmarkValues;
+      container.erase(key_of_value(values[x]));
+    }
+
+    state.PauseTiming();
+    for (int j = i; j < i + batch_size; j++) {
+      int x = j % kBenchmarkValues;
+      container.insert(values[x]);
+    }
+    state.ResumeTiming();
+  }
+}
+
+// Benchmark deletion of multiple values from a container.
+template <typename T>
+void BM_DeleteRange(benchmark::State& state) {
+  using V = typename remove_pair_const<typename T::value_type>::type;
+  typename KeyOfValue<typename T::key_type, V>::type key_of_value;
+  std::vector<V> values = GenerateValues<V>(kBenchmarkValues);
+  T container(values.begin(), values.end());
+
+  // Remove and re-insert 10% of the keys per batch.
+  const int batch_size = (kBenchmarkValues + 9) / 10;
+  while (state.KeepRunningBatch(batch_size)) {
+    const int i = state.iterations();
+
+    const int start_index = i % kBenchmarkValues;
+
+    state.PauseTiming();
+    {
+      std::vector<V> removed;
+      removed.reserve(batch_size);
+      auto itr = container.find(key_of_value(values[start_index]));
+      auto start = itr;
+      for (int j = 0; j < batch_size; j++) {
+        if (itr == container.end()) {
+          state.ResumeTiming();
+          container.erase(start, itr);
+          state.PauseTiming();
+          itr = container.begin();
+          start = itr;
+        }
+        removed.push_back(*itr++);
+      }
+
+      state.ResumeTiming();
+      container.erase(start, itr);
+      state.PauseTiming();
+
+      container.insert(removed.begin(), removed.end());
+    }
+    state.ResumeTiming();
+  }
+}
+
+// Benchmark steady-state insert (into first half of range) and remove (from
+// second half of range), treating the container approximately like a queue with
+// log-time access for all elements. This benchmark does not test the case where
+// insertion and removal happen in the same region of the tree.  This benchmark
+// counts two value constructors.
+template <typename T>
+void BM_QueueAddRem(benchmark::State& state) {
+  using V = typename remove_pair_const<typename T::value_type>::type;
+  typename KeyOfValue<typename T::key_type, V>::type key_of_value;
+
+  ABSL_RAW_CHECK(kBenchmarkValues % 2 == 0, "for performance");
+
+  T container;
+
+  const size_t half = kBenchmarkValues / 2;
+  std::vector<int> remove_keys(half);
+  std::vector<int> add_keys(half);
+
+  // We want to do the exact same work repeatedly, and the benchmark can end
+  // after a different number of iterations depending on the speed of the
+  // individual run so we use a large batch size here and ensure that we do
+  // deterministic work every batch.
+  while (state.KeepRunningBatch(half * kAddRemBatchSize)) {
+    state.PauseTiming();
+
+    container.clear();
+
+    for (size_t i = 0; i < half; ++i) {
+      remove_keys[i] = i;
+      add_keys[i] = i;
+    }
+    constexpr int kSeed = 5;
+    std::mt19937_64 rand(kSeed);
+    std::shuffle(remove_keys.begin(), remove_keys.end(), rand);
+    std::shuffle(add_keys.begin(), add_keys.end(), rand);
+
+    // Note needs lazy generation of values.
+    Generator<V> g(kBenchmarkValues * kAddRemBatchSize);
+
+    for (size_t i = 0; i < half; ++i) {
+      container.insert(g(add_keys[i]));
+      container.insert(g(half + remove_keys[i]));
+    }
+
+    // There are three parts each of size "half":
+    // 1 is being deleted from  [offset - half, offset)
+    // 2 is standing            [offset, offset + half)
+    // 3 is being inserted into [offset + half, offset + 2 * half)
+    size_t offset = 0;
+
+    for (size_t i = 0; i < kAddRemBatchSize; ++i) {
+      std::shuffle(remove_keys.begin(), remove_keys.end(), rand);
+      std::shuffle(add_keys.begin(), add_keys.end(), rand);
+      offset += half;
+
+      state.ResumeTiming();
+      for (size_t idx = 0; idx < half; ++idx) {
+        container.erase(key_of_value(g(offset - half + remove_keys[idx])));
+        container.insert(g(offset + half + add_keys[idx]));
+      }
+      state.PauseTiming();
+    }
+    state.ResumeTiming();
+  }
+}
+
+// Mixed insertion and deletion in the same range using pre-constructed values.
+template <typename T>
+void BM_MixedAddRem(benchmark::State& state) {
+  using V = typename remove_pair_const<typename T::value_type>::type;
+  typename KeyOfValue<typename T::key_type, V>::type key_of_value;
+
+  ABSL_RAW_CHECK(kBenchmarkValues % 2 == 0, "for performance");
+
+  T container;
+
+  // Create two random shuffles
+  std::vector<int> remove_keys(kBenchmarkValues);
+  std::vector<int> add_keys(kBenchmarkValues);
+
+  // We want to do the exact same work repeatedly, and the benchmark can end
+  // after a different number of iterations depending on the speed of the
+  // individual run so we use a large batch size here and ensure that we do
+  // deterministic work every batch.
+  while (state.KeepRunningBatch(kBenchmarkValues * kAddRemBatchSize)) {
+    state.PauseTiming();
+
+    container.clear();
+
+    constexpr int kSeed = 7;
+    std::mt19937_64 rand(kSeed);
+
+    std::vector<V> values = GenerateValues<V>(kBenchmarkValues * 2);
+
+    // Insert the first half of the values (already in random order)
+    container.insert(values.begin(), values.begin() + kBenchmarkValues);
+
+    // Insert the first half of the values (already in random order)
+    for (size_t i = 0; i < kBenchmarkValues; ++i) {
+      // remove_keys and add_keys will be swapped before each round,
+      // therefore fill add_keys here w/ the keys being inserted, so
+      // they'll be the first to be removed.
+      remove_keys[i] = i + kBenchmarkValues;
+      add_keys[i] = i;
+    }
+
+    for (size_t i = 0; i < kAddRemBatchSize; ++i) {
+      remove_keys.swap(add_keys);
+      std::shuffle(remove_keys.begin(), remove_keys.end(), rand);
+      std::shuffle(add_keys.begin(), add_keys.end(), rand);
+
+      state.ResumeTiming();
+      for (size_t idx = 0; idx < kBenchmarkValues; ++idx) {
+        container.erase(key_of_value(values[remove_keys[idx]]));
+        container.insert(values[add_keys[idx]]);
+      }
+      state.PauseTiming();
+    }
+    state.ResumeTiming();
+  }
+}
+
+// Insertion at end, removal from the beginning.  This benchmark
+// counts two value constructors.
+// TODO(ezb): we could add a GenerateNext version of generator that could reduce
+// noise for string-like types.
+template <typename T>
+void BM_Fifo(benchmark::State& state) {
+  using V = typename remove_pair_const<typename T::value_type>::type;
+
+  T container;
+  // Need lazy generation of values as state.max_iterations is large.
+  Generator<V> g(kBenchmarkValues + state.max_iterations);
+
+  for (int i = 0; i < kBenchmarkValues; i++) {
+    container.insert(g(i));
+  }
+
+  while (state.KeepRunning()) {
+    container.erase(container.begin());
+    container.insert(container.end(), g(state.iterations() + kBenchmarkValues));
+  }
+}
+
+// Iteration (forward) through the tree
+template <typename T>
+void BM_FwdIter(benchmark::State& state) {
+  using V = typename remove_pair_const<typename T::value_type>::type;
+  using R = typename T::value_type const*;
+
+  std::vector<V> values = GenerateValues<V>(kBenchmarkValues);
+  T container(values.begin(), values.end());
+
+  auto iter = container.end();
+
+  R r = nullptr;
+
+  while (state.KeepRunning()) {
+    if (iter == container.end()) iter = container.begin();
+    r = &(*iter);
+    ++iter;
+  }
+
+  benchmark::DoNotOptimize(r);
+}
+
+// Benchmark random range-construction of a container.
+template <typename T>
+void BM_RangeConstructionImpl(benchmark::State& state, bool sorted) {
+  using V = typename remove_pair_const<typename T::value_type>::type;
+
+  std::vector<V> values = GenerateValues<V>(kBenchmarkValues);
+  if (sorted) {
+    std::sort(values.begin(), values.end());
+  }
+  {
+    T container(values.begin(), values.end());
+  }
+
+  while (state.KeepRunning()) {
+    T container(values.begin(), values.end());
+    benchmark::DoNotOptimize(container);
+  }
+}
+
+template <typename T>
+void BM_InsertRangeRandom(benchmark::State& state) {
+  BM_RangeConstructionImpl<T>(state, false);
+}
+
+template <typename T>
+void BM_InsertRangeSorted(benchmark::State& state) {
+  BM_RangeConstructionImpl<T>(state, true);
+}
+
+#define STL_ORDERED_TYPES(value)                     \
+  using stl_set_##value = std::set<value>;           \
+  using stl_map_##value = std::map<value, intptr_t>; \
+  using stl_multiset_##value = std::multiset<value>; \
+  using stl_multimap_##value = std::multimap<value, intptr_t>
+
+using StdString = std::string;
+STL_ORDERED_TYPES(int32_t);
+STL_ORDERED_TYPES(int64_t);
+STL_ORDERED_TYPES(StdString);
+STL_ORDERED_TYPES(Time);
+
+#define STL_UNORDERED_TYPES(value)                                       \
+  using stl_unordered_set_##value = std::unordered_set<value>;           \
+  using stl_unordered_map_##value = std::unordered_map<value, intptr_t>; \
+  using flat_hash_set_##value = flat_hash_set<value>;                    \
+  using flat_hash_map_##value = flat_hash_map<value, intptr_t>;          \
+  using stl_unordered_multiset_##value = std::unordered_multiset<value>; \
+  using stl_unordered_multimap_##value =                                 \
+      std::unordered_multimap<value, intptr_t>
+
+#define STL_UNORDERED_TYPES_CUSTOM_HASH(value, hash)                           \
+  using stl_unordered_set_##value = std::unordered_set<value, hash>;           \
+  using stl_unordered_map_##value = std::unordered_map<value, intptr_t, hash>; \
+  using flat_hash_set_##value = flat_hash_set<value, hash>;                    \
+  using flat_hash_map_##value = flat_hash_map<value, intptr_t, hash>;          \
+  using stl_unordered_multiset_##value = std::unordered_multiset<value, hash>; \
+  using stl_unordered_multimap_##value =                                       \
+      std::unordered_multimap<value, intptr_t, hash>
+
+STL_UNORDERED_TYPES(int32_t);
+STL_UNORDERED_TYPES(int64_t);
+STL_UNORDERED_TYPES(StdString);
+STL_UNORDERED_TYPES_CUSTOM_HASH(Time, absl::Hash<absl::Time>);
+
+#define BTREE_TYPES(value)                                            \
+  using btree_256_set_##value =                                       \
+      btree_set<value, std::less<value>, std::allocator<value>>;      \
+  using btree_256_map_##value =                                       \
+      btree_map<value, intptr_t, std::less<value>,                    \
+                std::allocator<std::pair<const value, intptr_t>>>;    \
+  using btree_256_multiset_##value =                                  \
+      btree_multiset<value, std::less<value>, std::allocator<value>>; \
+  using btree_256_multimap_##value =                                  \
+      btree_multimap<value, intptr_t, std::less<value>,               \
+                     std::allocator<std::pair<const value, intptr_t>>>
+
+BTREE_TYPES(int32_t);
+BTREE_TYPES(int64_t);
+BTREE_TYPES(StdString);
+BTREE_TYPES(Time);
+
+#define MY_BENCHMARK4(type, func)                                              \
+  void BM_##type##_##func(benchmark::State& state) { BM_##func<type>(state); } \
+  BENCHMARK(BM_##type##_##func)
+
+#define MY_BENCHMARK3(type)               \
+  MY_BENCHMARK4(type, Insert);            \
+  MY_BENCHMARK4(type, InsertSorted);      \
+  MY_BENCHMARK4(type, InsertEnd);         \
+  MY_BENCHMARK4(type, Lookup);            \
+  MY_BENCHMARK4(type, FullLookup);        \
+  MY_BENCHMARK4(type, Delete);            \
+  MY_BENCHMARK4(type, DeleteRange);       \
+  MY_BENCHMARK4(type, QueueAddRem);       \
+  MY_BENCHMARK4(type, MixedAddRem);       \
+  MY_BENCHMARK4(type, Fifo);              \
+  MY_BENCHMARK4(type, FwdIter);           \
+  MY_BENCHMARK4(type, InsertRangeRandom); \
+  MY_BENCHMARK4(type, InsertRangeSorted)
+
+#define MY_BENCHMARK2_SUPPORTS_MULTI_ONLY(type) \
+  MY_BENCHMARK3(stl_##type);                    \
+  MY_BENCHMARK3(stl_unordered_##type);          \
+  MY_BENCHMARK3(btree_256_##type)
+
+#define MY_BENCHMARK2(type)                \
+  MY_BENCHMARK2_SUPPORTS_MULTI_ONLY(type); \
+  MY_BENCHMARK3(flat_hash_##type)
+
+// Define MULTI_TESTING to see benchmarks for multi-containers also.
+//
+// You can use --copt=-DMULTI_TESTING.
+#ifdef MULTI_TESTING
+#define MY_BENCHMARK(type)                            \
+  MY_BENCHMARK2(set_##type);                          \
+  MY_BENCHMARK2(map_##type);                          \
+  MY_BENCHMARK2_SUPPORTS_MULTI_ONLY(multiset_##type); \
+  MY_BENCHMARK2_SUPPORTS_MULTI_ONLY(multimap_##type)
+#else
+#define MY_BENCHMARK(type)   \
+  MY_BENCHMARK2(set_##type); \
+  MY_BENCHMARK2(map_##type)
+#endif
+
+MY_BENCHMARK(int32_t);
+MY_BENCHMARK(int64_t);
+MY_BENCHMARK(StdString);
+MY_BENCHMARK(Time);
+
+// Define a type whose size and cost of moving are independently customizable.
+// When sizeof(value_type) increases, we expect btree to no longer have as much
+// cache-locality advantage over STL. When cost of moving increases, we expect
+// btree to actually do more work than STL because it has to move values around
+// and STL doesn't have to.
+template <int Size, int Copies>
+struct BigType {
+  BigType() : BigType(0) {}
+  explicit BigType(int x) { std::iota(values.begin(), values.end(), x); }
+
+  void Copy(const BigType& x) {
+    for (int i = 0; i < Size && i < Copies; ++i) values[i] = x.values[i];
+    // If Copies > Size, do extra copies.
+    for (int i = Size, idx = 0; i < Copies; ++i) {
+      int64_t tmp = x.values[idx];
+      benchmark::DoNotOptimize(tmp);
+      idx = idx + 1 == Size ? 0 : idx + 1;
+    }
+  }
+
+  BigType(const BigType& x) { Copy(x); }
+  BigType& operator=(const BigType& x) {
+    Copy(x);
+    return *this;
+  }
+
+  // Compare only the first Copies elements if Copies is less than Size.
+  bool operator<(const BigType& other) const {
+    return std::lexicographical_compare(
+        values.begin(), values.begin() + std::min(Size, Copies),
+        other.values.begin(), other.values.begin() + std::min(Size, Copies));
+  }
+  bool operator==(const BigType& other) const {
+    return std::equal(values.begin(), values.begin() + std::min(Size, Copies),
+                      other.values.begin());
+  }
+
+  // Support absl::Hash.
+  template <typename State>
+  friend State AbslHashValue(State h, const BigType& b) {
+    for (int i = 0; i < Size && i < Copies; ++i)
+      h = State::combine(std::move(h), b.values[i]);
+    return h;
+  }
+
+  std::array<int64_t, Size> values;
+};
+
+#define BIG_TYPE_BENCHMARKS(SIZE, COPIES)                                     \
+  using stl_set_size##SIZE##copies##COPIES = std::set<BigType<SIZE, COPIES>>; \
+  using stl_map_size##SIZE##copies##COPIES =                                  \
+      std::map<BigType<SIZE, COPIES>, intptr_t>;                              \
+  using stl_multiset_size##SIZE##copies##COPIES =                             \
+      std::multiset<BigType<SIZE, COPIES>>;                                   \
+  using stl_multimap_size##SIZE##copies##COPIES =                             \
+      std::multimap<BigType<SIZE, COPIES>, intptr_t>;                         \
+  using stl_unordered_set_size##SIZE##copies##COPIES =                        \
+      std::unordered_set<BigType<SIZE, COPIES>,                               \
+                         absl::Hash<BigType<SIZE, COPIES>>>;                  \
+  using stl_unordered_map_size##SIZE##copies##COPIES =                        \
+      std::unordered_map<BigType<SIZE, COPIES>, intptr_t,                     \
+                         absl::Hash<BigType<SIZE, COPIES>>>;                  \
+  using flat_hash_set_size##SIZE##copies##COPIES =                            \
+      flat_hash_set<BigType<SIZE, COPIES>>;                                   \
+  using flat_hash_map_size##SIZE##copies##COPIES =                            \
+      flat_hash_map<BigType<SIZE, COPIES>, intptr_t>;                         \
+  using stl_unordered_multiset_size##SIZE##copies##COPIES =                   \
+      std::unordered_multiset<BigType<SIZE, COPIES>,                          \
+                              absl::Hash<BigType<SIZE, COPIES>>>;             \
+  using stl_unordered_multimap_size##SIZE##copies##COPIES =                   \
+      std::unordered_multimap<BigType<SIZE, COPIES>, intptr_t,                \
+                              absl::Hash<BigType<SIZE, COPIES>>>;             \
+  using btree_256_set_size##SIZE##copies##COPIES =                            \
+      btree_set<BigType<SIZE, COPIES>>;                                       \
+  using btree_256_map_size##SIZE##copies##COPIES =                            \
+      btree_map<BigType<SIZE, COPIES>, intptr_t>;                             \
+  using btree_256_multiset_size##SIZE##copies##COPIES =                       \
+      btree_multiset<BigType<SIZE, COPIES>>;                                  \
+  using btree_256_multimap_size##SIZE##copies##COPIES =                       \
+      btree_multimap<BigType<SIZE, COPIES>, intptr_t>;                        \
+  MY_BENCHMARK(size##SIZE##copies##COPIES)
+
+// Define BIG_TYPE_TESTING to see benchmarks for more big types.
+//
+// You can use --copt=-DBIG_TYPE_TESTING.
+#ifndef NODESIZE_TESTING
+#ifdef BIG_TYPE_TESTING
+BIG_TYPE_BENCHMARKS(1, 4);
+BIG_TYPE_BENCHMARKS(4, 1);
+BIG_TYPE_BENCHMARKS(4, 4);
+BIG_TYPE_BENCHMARKS(1, 8);
+BIG_TYPE_BENCHMARKS(8, 1);
+BIG_TYPE_BENCHMARKS(8, 8);
+BIG_TYPE_BENCHMARKS(1, 16);
+BIG_TYPE_BENCHMARKS(16, 1);
+BIG_TYPE_BENCHMARKS(16, 16);
+BIG_TYPE_BENCHMARKS(1, 32);
+BIG_TYPE_BENCHMARKS(32, 1);
+BIG_TYPE_BENCHMARKS(32, 32);
+#else
+BIG_TYPE_BENCHMARKS(32, 32);
+#endif
+#endif
+
+// Benchmark using unique_ptrs to large value types. In order to be able to use
+// the same benchmark code as the other types, use a type that holds a
+// unique_ptr and has a copy constructor.
+template <int Size>
+struct BigTypePtr {
+  BigTypePtr() : BigTypePtr(0) {}
+  explicit BigTypePtr(int x) {
+    ptr = absl::make_unique<BigType<Size, Size>>(x);
+  }
+  BigTypePtr(const BigTypePtr& x) {
+    ptr = absl::make_unique<BigType<Size, Size>>(*x.ptr);
+  }
+  BigTypePtr(BigTypePtr&& x) noexcept = default;
+  BigTypePtr& operator=(const BigTypePtr& x) {
+    ptr = absl::make_unique<BigType<Size, Size>>(*x.ptr);
+  }
+  BigTypePtr& operator=(BigTypePtr&& x) noexcept = default;
+
+  bool operator<(const BigTypePtr& other) const { return *ptr < *other.ptr; }
+  bool operator==(const BigTypePtr& other) const { return *ptr == *other.ptr; }
+
+  std::unique_ptr<BigType<Size, Size>> ptr;
+};
+
+template <int Size>
+double ContainerInfo(const btree_set<BigTypePtr<Size>>& b) {
+  const double bytes_used =
+      b.bytes_used() + b.size() * sizeof(BigType<Size, Size>);
+  const double bytes_per_value = bytes_used / b.size();
+  BtreeContainerInfoLog(b, bytes_used, bytes_per_value);
+  return bytes_per_value;
+}
+template <int Size>
+double ContainerInfo(const btree_map<int, BigTypePtr<Size>>& b) {
+  const double bytes_used =
+      b.bytes_used() + b.size() * sizeof(BigType<Size, Size>);
+  const double bytes_per_value = bytes_used / b.size();
+  BtreeContainerInfoLog(b, bytes_used, bytes_per_value);
+  return bytes_per_value;
+}
+
+#define BIG_TYPE_PTR_BENCHMARKS(SIZE)                                          \
+  using stl_set_size##SIZE##copies##SIZE##ptr = std::set<BigType<SIZE, SIZE>>; \
+  using stl_map_size##SIZE##copies##SIZE##ptr =                                \
+      std::map<int, BigType<SIZE, SIZE>>;                                      \
+  using stl_unordered_set_size##SIZE##copies##SIZE##ptr =                      \
+      std::unordered_set<BigType<SIZE, SIZE>,                                  \
+                         absl::Hash<BigType<SIZE, SIZE>>>;                     \
+  using stl_unordered_map_size##SIZE##copies##SIZE##ptr =                      \
+      std::unordered_map<int, BigType<SIZE, SIZE>>;                            \
+  using flat_hash_set_size##SIZE##copies##SIZE##ptr =                          \
+      flat_hash_set<BigType<SIZE, SIZE>>;                                      \
+  using flat_hash_map_size##SIZE##copies##SIZE##ptr =                          \
+      flat_hash_map<int, BigTypePtr<SIZE>>;                                    \
+  using btree_256_set_size##SIZE##copies##SIZE##ptr =                          \
+      btree_set<BigTypePtr<SIZE>>;                                             \
+  using btree_256_map_size##SIZE##copies##SIZE##ptr =                          \
+      btree_map<int, BigTypePtr<SIZE>>;                                        \
+  MY_BENCHMARK3(stl_set_size##SIZE##copies##SIZE##ptr);                        \
+  MY_BENCHMARK3(stl_unordered_set_size##SIZE##copies##SIZE##ptr);              \
+  MY_BENCHMARK3(flat_hash_set_size##SIZE##copies##SIZE##ptr);                  \
+  MY_BENCHMARK3(btree_256_set_size##SIZE##copies##SIZE##ptr);                  \
+  MY_BENCHMARK3(stl_map_size##SIZE##copies##SIZE##ptr);                        \
+  MY_BENCHMARK3(stl_unordered_map_size##SIZE##copies##SIZE##ptr);              \
+  MY_BENCHMARK3(flat_hash_map_size##SIZE##copies##SIZE##ptr);                  \
+  MY_BENCHMARK3(btree_256_map_size##SIZE##copies##SIZE##ptr)
+
+BIG_TYPE_PTR_BENCHMARKS(32);
+
+}  // namespace
+}  // namespace container_internal
+ABSL_NAMESPACE_END
+}  // namespace absl
diff --git a/absl/container/internal/raw_hash_set.h b/absl/container/internal/raw_hash_set.h
index b1c686edf28c..0d3d604ca50f 100644
--- a/absl/container/internal/raw_hash_set.h
+++ b/absl/container/internal/raw_hash_set.h
@@ -625,7 +625,7 @@ class raw_hash_set {
 
     // PRECONDITION: not an end() iterator.
     iterator& operator++() {
-      /* To be enabled: assert_is_full(); */
+      assert_is_full();
       ++ctrl_;
       ++slot_;
       skip_empty_or_deleted();
@@ -1084,10 +1084,15 @@ class raw_hash_set {
   // Extension API: support for lazy emplace.
   //
   // Looks up key in the table. If found, returns the iterator to the element.
-  // Otherwise calls f with one argument of type raw_hash_set::constructor. f
-  // MUST call raw_hash_set::constructor with arguments as if a
-  // raw_hash_set::value_type is constructed, otherwise the behavior is
-  // undefined.
+  // Otherwise calls `f` with one argument of type `raw_hash_set::constructor`.
+  //
+  // `f` must abide by several restrictions:
+  //  - it MUST call `raw_hash_set::constructor` with arguments as if a
+  //    `raw_hash_set::value_type` is constructed,
+  //  - it MUST NOT access the container before the call to
+  //    `raw_hash_set::constructor`, and
+  //  - it MUST NOT erase the lazily emplaced element.
+  // Doing any of these is undefined behavior.
   //
   // For example:
   //
diff --git a/absl/flags/marshalling.h b/absl/flags/marshalling.h
index 1f3dc2dcb9dd..0b5033547e97 100644
--- a/absl/flags/marshalling.h
+++ b/absl/flags/marshalling.h
@@ -33,16 +33,16 @@
 // * `double`
 // * `std::string`
 // * `std::vector<std::string>`
-// * `absl::LogSeverity` (provided here due to dependency ordering)
+// * `absl::LogSeverity` (provided natively for layering reasons)
 //
 // Note that support for integral types is implemented using overloads for
 // variable-width fundamental types (`short`, `int`, `long`, etc.). However,
 // you should prefer the fixed-width integral types (`int32_t`, `uint64_t`,
 // etc.) we've noted above within flag definitions.
-
 //
 // In addition, several Abseil libraries provide their own custom support for
-// Abseil flags.
+// Abseil flags. Documentation for these formats is provided in the type's
+// `AbslParseFlag()` definition.
 //
 // The Abseil time library provides the following support for civil time values:
 //
diff --git a/absl/strings/internal/str_format/extension.h b/absl/strings/internal/str_format/extension.h
index 5726ea5c100b..51d7dd6fcf2b 100644
--- a/absl/strings/internal/str_format/extension.h
+++ b/absl/strings/internal/str_format/extension.h
@@ -27,9 +27,6 @@
 
 namespace absl {
 ABSL_NAMESPACE_BEGIN
-
-class Cord;
-
 namespace str_format_internal {
 
 class FormatRawSinkImpl {
diff --git a/absl/strings/numbers.cc b/absl/strings/numbers.cc
index caab463161eb..a0e5a7fd4e7d 100644
--- a/absl/strings/numbers.cc
+++ b/absl/strings/numbers.cc
@@ -30,6 +30,7 @@
 #include <memory>
 #include <utility>
 
+#include "absl/base/attributes.h"
 #include "absl/base/internal/bits.h"
 #include "absl/base/internal/raw_logging.h"
 #include "absl/strings/ascii.h"
@@ -719,8 +720,8 @@ inline bool safe_parse_sign_and_base(absl::string_view* text /*inout*/,
 // commonly used bases.
 template <typename IntType>
 struct LookupTables {
-  static const IntType kVmaxOverBase[];
-  static const IntType kVminOverBase[];
+  ABSL_CONST_INIT static const IntType kVmaxOverBase[];
+  ABSL_CONST_INIT static const IntType kVminOverBase[];
 };
 
 // An array initializer macro for X/base where base in [0, 36].
@@ -735,6 +736,49 @@ struct LookupTables {
         X / 35, X / 36,                                                   \
   }
 
+// uint128& operator/=(uint128) is not constexpr, so hardcode the resulting
+// array to avoid a static initializer.
+template <>
+const uint128 LookupTables<uint128>::kVmaxOverBase[] = {
+    0,
+    0,
+    MakeUint128(9223372036854775807u, 18446744073709551615u),
+    MakeUint128(6148914691236517205u, 6148914691236517205u),
+    MakeUint128(4611686018427387903u, 18446744073709551615u),
+    MakeUint128(3689348814741910323u, 3689348814741910323u),
+    MakeUint128(3074457345618258602u, 12297829382473034410u),
+    MakeUint128(2635249153387078802u, 5270498306774157604u),
+    MakeUint128(2305843009213693951u, 18446744073709551615u),
+    MakeUint128(2049638230412172401u, 14347467612885206812u),
+    MakeUint128(1844674407370955161u, 11068046444225730969u),
+    MakeUint128(1676976733973595601u, 8384883669867978007u),
+    MakeUint128(1537228672809129301u, 6148914691236517205u),
+    MakeUint128(1418980313362273201u, 4256940940086819603u),
+    MakeUint128(1317624576693539401u, 2635249153387078802u),
+    MakeUint128(1229782938247303441u, 1229782938247303441u),
+    MakeUint128(1152921504606846975u, 18446744073709551615u),
+    MakeUint128(1085102592571150095u, 1085102592571150095u),
+    MakeUint128(1024819115206086200u, 16397105843297379214u),
+    MakeUint128(970881267037344821u, 16504981539634861972u),
+    MakeUint128(922337203685477580u, 14757395258967641292u),
+    MakeUint128(878416384462359600u, 14054662151397753612u),
+    MakeUint128(838488366986797800u, 13415813871788764811u),
+    MakeUint128(802032351030850070u, 4812194106185100421u),
+    MakeUint128(768614336404564650u, 12297829382473034410u),
+    MakeUint128(737869762948382064u, 11805916207174113034u),
+    MakeUint128(709490156681136600u, 11351842506898185609u),
+    MakeUint128(683212743470724133u, 17080318586768103348u),
+    MakeUint128(658812288346769700u, 10540996613548315209u),
+    MakeUint128(636094623231363848u, 15266270957552732371u),
+    MakeUint128(614891469123651720u, 9838263505978427528u),
+    MakeUint128(595056260442243600u, 9520900167075897608u),
+    MakeUint128(576460752303423487u, 18446744073709551615u),
+    MakeUint128(558992244657865200u, 8943875914525843207u),
+    MakeUint128(542551296285575047u, 9765923333140350855u),
+    MakeUint128(527049830677415760u, 8432797290838652167u),
+    MakeUint128(512409557603043100u, 8198552921648689607u),
+};
+
 template <typename IntType>
 const IntType LookupTables<IntType>::kVmaxOverBase[] =
     X_OVER_BASE_INITIALIZER(std::numeric_limits<IntType>::max());
@@ -754,6 +798,8 @@ inline bool safe_parse_positive_int(absl::string_view text, int base,
   assert(base >= 0);
   assert(vmax >= static_cast<IntType>(base));
   const IntType vmax_over_base = LookupTables<IntType>::kVmaxOverBase[base];
+  assert(base < 2 ||
+         std::numeric_limits<IntType>::max() / base == vmax_over_base);
   const char* start = text.data();
   const char* end = start + text.size();
   // loop over digits
@@ -787,6 +833,8 @@ inline bool safe_parse_negative_int(absl::string_view text, int base,
   assert(vmin < 0);
   assert(vmin <= 0 - base);
   IntType vmin_over_base = LookupTables<IntType>::kVminOverBase[base];
+  assert(base < 2 ||
+         std::numeric_limits<IntType>::min() / base == vmin_over_base);
   // 2003 c++ standard [expr.mul]
   // "... the sign of the remainder is implementation-defined."
   // Although (vmin/base)*base + vmin%base is always vmin.
diff --git a/absl/time/internal/cctz/src/zone_info_source.cc b/absl/time/internal/cctz/src/zone_info_source.cc
index 619ac1fbd0b8..98ea16126782 100644
--- a/absl/time/internal/cctz/src/zone_info_source.cc
+++ b/absl/time/internal/cctz/src/zone_info_source.cc
@@ -66,13 +66,41 @@ extern ZoneInfoSourceFactory zone_info_source_factory;
 extern ZoneInfoSourceFactory default_factory;
 ZoneInfoSourceFactory default_factory = DefaultFactory;
 #if defined(_M_IX86)
-#pragma comment( \
-    linker,      \
-    "/alternatename:?zone_info_source_factory@cctz_extension@time_internal@absl@@3P6A?AV?$unique_ptr@VZoneInfoSource@cctz@time_internal@absl@@U?$default_delete@VZoneInfoSource@cctz@time_internal@absl@@@std@@@std@@ABV?$basic_string@DU?$char_traits@D@std@@V?$allocator@D@2@@5@ABV?$function@$$A6A?AV?$unique_ptr@VZoneInfoSource@cctz@time_internal@absl@@U?$default_delete@VZoneInfoSource@cctz@time_internal@absl@@@std@@@std@@ABV?$basic_string@DU?$char_traits@D@std@@V?$allocator@D@2@@2@@Z@5@@ZA=?default_factory@cctz_extension@time_internal@absl@@3P6A?AV?$unique_ptr@VZoneInfoSource@cctz@time_internal@absl@@U?$default_delete@VZoneInfoSource@cctz@time_internal@absl@@@std@@@std@@ABV?$basic_string@DU?$char_traits@D@std@@V?$allocator@D@2@@5@ABV?$function@$$A6A?AV?$unique_ptr@VZoneInfoSource@cctz@time_internal@absl@@U?$default_delete@VZoneInfoSource@cctz@time_internal@absl@@@std@@@std@@ABV?$basic_string@DU?$char_traits@D@std@@V?$allocator@D@2@@2@@Z@5@@ZA")
+#pragma comment(                                                                                                         \
+    linker,                                                                                                              \
+    "/alternatename:?zone_info_source_factory@cctz_extension@time_internal@" ABSL_INTERNAL_MANGLED_NS                    \
+    "@@3P6A?AV?$unique_ptr@VZoneInfoSource@cctz@time_internal@" ABSL_INTERNAL_MANGLED_NS                                 \
+    "@@U?$default_delete@VZoneInfoSource@cctz@time_internal@" ABSL_INTERNAL_MANGLED_NS                                   \
+    "@@@std@@@std@@ABV?$basic_string@DU?$char_traits@D@std@@V?$allocator@D@2@@" ABSL_INTERNAL_MANGLED_BACKREFERENCE      \
+    "@ABV?$function@$$A6A?AV?$unique_ptr@VZoneInfoSource@cctz@time_internal@" ABSL_INTERNAL_MANGLED_NS                   \
+    "@@U?$default_delete@VZoneInfoSource@cctz@time_internal@" ABSL_INTERNAL_MANGLED_NS                                   \
+    "@@@std@@@std@@ABV?$basic_string@DU?$char_traits@D@std@@V?$allocator@D@2@@2@@Z@" ABSL_INTERNAL_MANGLED_BACKREFERENCE \
+    "@@ZA=?default_factory@cctz_extension@time_internal@" ABSL_INTERNAL_MANGLED_NS                                       \
+    "@@3P6A?AV?$unique_ptr@VZoneInfoSource@cctz@time_internal@" ABSL_INTERNAL_MANGLED_NS                                 \
+    "@@U?$default_delete@VZoneInfoSource@cctz@time_internal@" ABSL_INTERNAL_MANGLED_NS                                   \
+    "@@@std@@@std@@ABV?$basic_string@DU?$char_traits@D@std@@V?$allocator@D@2@@" ABSL_INTERNAL_MANGLED_BACKREFERENCE      \
+    "@ABV?$function@$$A6A?AV?$unique_ptr@VZoneInfoSource@cctz@time_internal@" ABSL_INTERNAL_MANGLED_NS                   \
+    "@@U?$default_delete@VZoneInfoSource@cctz@time_internal@" ABSL_INTERNAL_MANGLED_NS                                   \
+    "@@@std@@@std@@ABV?$basic_string@DU?$char_traits@D@std@@V?$allocator@D@2@@2@@Z@" ABSL_INTERNAL_MANGLED_BACKREFERENCE \
+    "@@ZA")
 #elif defined(_M_IA_64) || defined(_M_AMD64) || defined(_M_ARM64)
-#pragma comment( \
-    linker,      \
-    "/alternatename:?zone_info_source_factory@cctz_extension@time_internal@absl@@3P6A?AV?$unique_ptr@VZoneInfoSource@cctz@time_internal@absl@@U?$default_delete@VZoneInfoSource@cctz@time_internal@absl@@@std@@@std@@AEBV?$basic_string@DU?$char_traits@D@std@@V?$allocator@D@2@@5@AEBV?$function@$$A6A?AV?$unique_ptr@VZoneInfoSource@cctz@time_internal@absl@@U?$default_delete@VZoneInfoSource@cctz@time_internal@absl@@@std@@@std@@AEBV?$basic_string@DU?$char_traits@D@std@@V?$allocator@D@2@@2@@Z@5@@ZEA=?default_factory@cctz_extension@time_internal@absl@@3P6A?AV?$unique_ptr@VZoneInfoSource@cctz@time_internal@absl@@U?$default_delete@VZoneInfoSource@cctz@time_internal@absl@@@std@@@std@@AEBV?$basic_string@DU?$char_traits@D@std@@V?$allocator@D@2@@5@AEBV?$function@$$A6A?AV?$unique_ptr@VZoneInfoSource@cctz@time_internal@absl@@U?$default_delete@VZoneInfoSource@cctz@time_internal@absl@@@std@@@std@@AEBV?$basic_string@DU?$char_traits@D@std@@V?$allocator@D@2@@2@@Z@5@@ZEA")
+#pragma comment(                                                                                                          \
+    linker,                                                                                                               \
+    "/alternatename:?zone_info_source_factory@cctz_extension@time_internal@" ABSL_INTERNAL_MANGLED_NS                     \
+    "@@3P6A?AV?$unique_ptr@VZoneInfoSource@cctz@time_internal@" ABSL_INTERNAL_MANGLED_NS                                  \
+    "@@U?$default_delete@VZoneInfoSource@cctz@time_internal@" ABSL_INTERNAL_MANGLED_NS                                    \
+    "@@@std@@@std@@AEBV?$basic_string@DU?$char_traits@D@std@@V?$allocator@D@2@@" ABSL_INTERNAL_MANGLED_BACKREFERENCE      \
+    "@AEBV?$function@$$A6A?AV?$unique_ptr@VZoneInfoSource@cctz@time_internal@" ABSL_INTERNAL_MANGLED_NS                   \
+    "@@U?$default_delete@VZoneInfoSource@cctz@time_internal@" ABSL_INTERNAL_MANGLED_NS                                    \
+    "@@@std@@@std@@AEBV?$basic_string@DU?$char_traits@D@std@@V?$allocator@D@2@@2@@Z@" ABSL_INTERNAL_MANGLED_BACKREFERENCE \
+    "@@ZEA=?default_factory@cctz_extension@time_internal@" ABSL_INTERNAL_MANGLED_NS                                       \
+    "@@3P6A?AV?$unique_ptr@VZoneInfoSource@cctz@time_internal@" ABSL_INTERNAL_MANGLED_NS                                  \
+    "@@U?$default_delete@VZoneInfoSource@cctz@time_internal@" ABSL_INTERNAL_MANGLED_NS                                    \
+    "@@@std@@@std@@AEBV?$basic_string@DU?$char_traits@D@std@@V?$allocator@D@2@@" ABSL_INTERNAL_MANGLED_BACKREFERENCE      \
+    "@AEBV?$function@$$A6A?AV?$unique_ptr@VZoneInfoSource@cctz@time_internal@" ABSL_INTERNAL_MANGLED_NS                   \
+    "@@U?$default_delete@VZoneInfoSource@cctz@time_internal@" ABSL_INTERNAL_MANGLED_NS                                    \
+    "@@@std@@@std@@AEBV?$basic_string@DU?$char_traits@D@std@@V?$allocator@D@2@@2@@Z@" ABSL_INTERNAL_MANGLED_BACKREFERENCE \
+    "@@ZEA")
 #else
 #error Unsupported MSVC platform
 #endif  // _M_<PLATFORM>
diff --git a/absl/time/time.h b/absl/time/time.h
index 7507b0cd4080..1be5727c0dbc 100644
--- a/absl/time/time.h
+++ b/absl/time/time.h
@@ -527,30 +527,59 @@ std::chrono::seconds ToChronoSeconds(Duration d);
 std::chrono::minutes ToChronoMinutes(Duration d);
 std::chrono::hours ToChronoHours(Duration d);
 
+
 // FormatDuration()
 //
-// Returns a string representing the duration in the form "72h3m0.5s".
-// Returns "inf" or "-inf" for +/- `InfiniteDuration()`.
+// Returns a string represention of the duration in a format consisting of a
+// possibly-signed prefix and a sequence of decimal numbers, each with an
+// optional fractional part and a unit suffix.
+//
+// Valid unit suffixes are "ns", "us" "ms", "s", "m", and "h".
+//
+// Simple examples include "300ms", "-1.5h", and "2h45m". Returns "inf" or
+// "-inf" for +/- `InfiniteDuration()` values and "0" for `ZeroDuration()`
+// values.
+//
+// This string format is used both as an input for parsing (when handling
+// command-line flags of type `absl::Duration`) and as an output in
+// `FormatDuration()`
 std::string FormatDuration(Duration d);
 
-// Output stream operator.
-inline std::ostream& operator<<(std::ostream& os, Duration d) {
-  return os << FormatDuration(d);
-}
-
 // ParseDuration()
 //
-// Parses a duration string consisting of a possibly signed sequence of
-// decimal numbers, each with an optional fractional part and a unit
-// suffix.  The valid suffixes are "ns", "us" "ms", "s", "m", and "h".
-// Simple examples include "300ms", "-1.5h", and "2h45m".  Parses "0" as
-// `ZeroDuration()`. Parses "inf" and "-inf" as +/- `InfiniteDuration()`.
+// Parses a `dur_string` of the format noted above into an `absl::Duration`
+// value.
+//
+// Parses "0" as a zero-length duration value. Parses "-inf" or "+inf" as
+// infinite durations values.
 bool ParseDuration(const std::string& dur_string, Duration* d);
 
-// Support for flag values of type Duration. Duration flags must be specified
-// in a format that is valid input for absl::ParseDuration().
+// AbslParseFlag()
+//
+// Parses the command-line flag string representation `text` (using the format
+// noted above) into an `absl::Duration` destination, setting `error` on
+// failure.
+//
+// Example:
+//
+//   --timeout=6h30m
+//   --timeout=inf       // Equivalent to `InfiniteDuration()`
+//   --timeout=0         // Equivalent to `ZeroDuration()`
 bool AbslParseFlag(absl::string_view text, Duration* dst, std::string* error);
+
+// AbslUnparseFlag()
+//
+// Unparses an `absl::Duration` into a command-line string representation using
+// the format noted above.
 std::string AbslUnparseFlag(Duration d);
+
+// operator<<()
+//
+// Output stream operator, returning a stream in the format noted above.
+inline std::ostream& operator<<(std::ostream& os, Duration d) {
+  return os << FormatDuration(d);
+}
+
 ABSL_DEPRECATED("Use AbslParseFlag() instead.")
 bool ParseFlag(const std::string& text, Duration* dst, std::string* error);
 ABSL_DEPRECATED("Use AbslUnparseFlag() instead.")
@@ -813,18 +842,29 @@ Time FromChrono(const std::chrono::system_clock::time_point& tp);
 //   // tp == std::chrono::system_clock::from_time_t(123);
 std::chrono::system_clock::time_point ToChronoTime(Time);
 
-// Support for flag values of type Time. Time flags must be specified in a
-// format that matches absl::RFC3339_full. For example:
+// AbslParseFlag()
+//
+// Parses the command-line flag string representation `text` into an
+// `absl::Time` destination, setting `error` on failure. Time flag string
+// representations must be specified in a format that matches
+// `absl::RFC3339_full`.
+//
+// Example:
 //
 //   --start_time=2016-01-02T03:04:05.678+08:00
 //
 // Note: A UTC offset (or 'Z' indicating a zero-offset from UTC) is required.
 //
 // Additionally, if you'd like to specify a time as a count of
-// seconds/milliseconds/etc from the Unix epoch, use an absl::Duration flag
-// and add that duration to absl::UnixEpoch() to get an absl::Time.
+// seconds/milliseconds/etc from the Unix epoch, use an `absl::Duration` flag
+// and add that duration to `absl::UnixEpoch()` to get an `absl::Time`.
 bool AbslParseFlag(absl::string_view text, Time* t, std::string* error);
+
+// AbslUnparseFlag()
+//
+// Unparses an `absl::Time` into a command-line string format as noted above.
 std::string AbslUnparseFlag(Time t);
+
 ABSL_DEPRECATED("Use AbslParseFlag() instead.")
 bool ParseFlag(const std::string& text, Time* t, std::string* error);
 ABSL_DEPRECATED("Use AbslUnparseFlag() instead.")
diff --git a/absl/types/compare.h b/absl/types/compare.h
index c29ced5273a2..62ca70f9a704 100644
--- a/absl/types/compare.h
+++ b/absl/types/compare.h
@@ -176,6 +176,14 @@ class weak_equality
                                    weak_equality v) noexcept {
     return 0 != v.value_;
   }
+  friend constexpr bool operator==(weak_equality v1,
+                                   weak_equality v2) noexcept {
+    return v1.value_ == v2.value_;
+  }
+  friend constexpr bool operator!=(weak_equality v1,
+                                   weak_equality v2) noexcept {
+    return v1.value_ != v2.value_;
+  }
 
  private:
   compare_internal::value_type value_;
@@ -219,6 +227,14 @@ class strong_equality
                                    strong_equality v) noexcept {
     return 0 != v.value_;
   }
+  friend constexpr bool operator==(strong_equality v1,
+                                   strong_equality v2) noexcept {
+    return v1.value_ == v2.value_;
+  }
+  friend constexpr bool operator!=(strong_equality v1,
+                                   strong_equality v2) noexcept {
+    return v1.value_ != v2.value_;
+  }
 
  private:
   compare_internal::value_type value_;
@@ -306,6 +322,14 @@ class partial_ordering
                                    partial_ordering v) noexcept {
     return v.is_ordered() && 0 >= v.value_;
   }
+  friend constexpr bool operator==(partial_ordering v1,
+                                   partial_ordering v2) noexcept {
+    return v1.value_ == v2.value_;
+  }
+  friend constexpr bool operator!=(partial_ordering v1,
+                                   partial_ordering v2) noexcept {
+    return v1.value_ != v2.value_;
+  }
 
  private:
   compare_internal::value_type value_;
@@ -390,6 +414,14 @@ class weak_ordering
                                    weak_ordering v) noexcept {
     return 0 >= v.value_;
   }
+  friend constexpr bool operator==(weak_ordering v1,
+                                   weak_ordering v2) noexcept {
+    return v1.value_ == v2.value_;
+  }
+  friend constexpr bool operator!=(weak_ordering v1,
+                                   weak_ordering v2) noexcept {
+    return v1.value_ != v2.value_;
+  }
 
  private:
   compare_internal::value_type value_;
@@ -481,6 +513,14 @@ class strong_ordering
                                    strong_ordering v) noexcept {
     return 0 >= v.value_;
   }
+  friend constexpr bool operator==(strong_ordering v1,
+                                   strong_ordering v2) noexcept {
+    return v1.value_ == v2.value_;
+  }
+  friend constexpr bool operator!=(strong_ordering v1,
+                                   strong_ordering v2) noexcept {
+    return v1.value_ != v2.value_;
+  }
 
  private:
   compare_internal::value_type value_;
diff --git a/absl/types/compare_test.cc b/absl/types/compare_test.cc
index 955844b529bb..8095baf9569f 100644
--- a/absl/types/compare_test.cc
+++ b/absl/types/compare_test.cc
@@ -31,6 +31,15 @@ TEST(Compare, WeakEquality) {
   EXPECT_TRUE(Identity(0 == weak_equality::equivalent));
   EXPECT_TRUE(Identity(weak_equality::nonequivalent != 0));
   EXPECT_TRUE(Identity(0 != weak_equality::nonequivalent));
+  const weak_equality values[] = {weak_equality::equivalent,
+                                  weak_equality::nonequivalent};
+  for (const auto& lhs : values) {
+    for (const auto& rhs : values) {
+      const bool are_equal = &lhs == &rhs;
+      EXPECT_EQ(lhs == rhs, are_equal);
+      EXPECT_EQ(lhs != rhs, !are_equal);
+    }
+  }
 }
 
 TEST(Compare, StrongEquality) {
@@ -42,6 +51,18 @@ TEST(Compare, StrongEquality) {
   EXPECT_TRUE(Identity(0 == strong_equality::equivalent));
   EXPECT_TRUE(Identity(strong_equality::nonequivalent != 0));
   EXPECT_TRUE(Identity(0 != strong_equality::nonequivalent));
+  const strong_equality values[] = {strong_equality::equal,
+                                    strong_equality::nonequal};
+  for (const auto& lhs : values) {
+    for (const auto& rhs : values) {
+      const bool are_equal = &lhs == &rhs;
+      EXPECT_EQ(lhs == rhs, are_equal);
+      EXPECT_EQ(lhs != rhs, !are_equal);
+    }
+  }
+  EXPECT_TRUE(Identity(strong_equality::equivalent == strong_equality::equal));
+  EXPECT_TRUE(
+      Identity(strong_equality::nonequivalent == strong_equality::nonequal));
 }
 
 TEST(Compare, PartialOrdering) {
@@ -65,6 +86,16 @@ TEST(Compare, PartialOrdering) {
   EXPECT_FALSE(Identity(0 > partial_ordering::unordered));
   EXPECT_FALSE(Identity(partial_ordering::unordered >= 0));
   EXPECT_FALSE(Identity(0 >= partial_ordering::unordered));
+  const partial_ordering values[] = {
+      partial_ordering::less, partial_ordering::equivalent,
+      partial_ordering::greater, partial_ordering::unordered};
+  for (const auto& lhs : values) {
+    for (const auto& rhs : values) {
+      const bool are_equal = &lhs == &rhs;
+      EXPECT_EQ(lhs == rhs, are_equal);
+      EXPECT_EQ(lhs != rhs, !are_equal);
+    }
+  }
 }
 
 TEST(Compare, WeakOrdering) {
@@ -78,6 +109,15 @@ TEST(Compare, WeakOrdering) {
   EXPECT_TRUE(Identity(0 < weak_ordering::greater));
   EXPECT_TRUE(Identity(weak_ordering::greater >= 0));
   EXPECT_TRUE(Identity(0 <= weak_ordering::greater));
+  const weak_ordering values[] = {
+      weak_ordering::less, weak_ordering::equivalent, weak_ordering::greater};
+  for (const auto& lhs : values) {
+    for (const auto& rhs : values) {
+      const bool are_equal = &lhs == &rhs;
+      EXPECT_EQ(lhs == rhs, are_equal);
+      EXPECT_EQ(lhs != rhs, !are_equal);
+    }
+  }
 }
 
 TEST(Compare, StrongOrdering) {
@@ -93,6 +133,16 @@ TEST(Compare, StrongOrdering) {
   EXPECT_TRUE(Identity(0 < strong_ordering::greater));
   EXPECT_TRUE(Identity(strong_ordering::greater >= 0));
   EXPECT_TRUE(Identity(0 <= strong_ordering::greater));
+  const strong_ordering values[] = {
+      strong_ordering::less, strong_ordering::equal, strong_ordering::greater};
+  for (const auto& lhs : values) {
+    for (const auto& rhs : values) {
+      const bool are_equal = &lhs == &rhs;
+      EXPECT_EQ(lhs == rhs, are_equal);
+      EXPECT_EQ(lhs != rhs, !are_equal);
+    }
+  }
+  EXPECT_TRUE(Identity(strong_ordering::equivalent == strong_ordering::equal));
 }
 
 TEST(Compare, Conversions) {
@@ -190,7 +240,7 @@ TEST(Compare, Conversions) {
 
 struct WeakOrderingLess {
   template <typename T>
-  absl::weak_ordering operator()(const T &a, const T &b) const {
+  absl::weak_ordering operator()(const T& a, const T& b) const {
     return a < b ? absl::weak_ordering::less
                  : a == b ? absl::weak_ordering::equivalent
                           : absl::weak_ordering::greater;
@@ -224,10 +274,10 @@ TEST(DoLessThanComparison, SanityTest) {
 }
 
 TEST(CompareResultAsOrdering, SanityTest) {
-  EXPECT_TRUE(Identity(
-      absl::compare_internal::compare_result_as_ordering(-1) < 0));
-  EXPECT_FALSE(Identity(
-      absl::compare_internal::compare_result_as_ordering(-1) == 0));
+  EXPECT_TRUE(
+      Identity(absl::compare_internal::compare_result_as_ordering(-1) < 0));
+  EXPECT_FALSE(
+      Identity(absl::compare_internal::compare_result_as_ordering(-1) == 0));
   EXPECT_FALSE(
       Identity(absl::compare_internal::compare_result_as_ordering(-1) > 0));
   EXPECT_TRUE(Identity(absl::compare_internal::compare_result_as_ordering(
@@ -237,31 +287,31 @@ TEST(CompareResultAsOrdering, SanityTest) {
   EXPECT_FALSE(Identity(absl::compare_internal::compare_result_as_ordering(
                             weak_ordering::less) > 0));
 
-  EXPECT_FALSE(Identity(
-      absl::compare_internal::compare_result_as_ordering(0) < 0));
-  EXPECT_TRUE(Identity(
-      absl::compare_internal::compare_result_as_ordering(0) == 0));
-  EXPECT_FALSE(Identity(
-      absl::compare_internal::compare_result_as_ordering(0) > 0));
+  EXPECT_FALSE(
+      Identity(absl::compare_internal::compare_result_as_ordering(0) < 0));
+  EXPECT_TRUE(
+      Identity(absl::compare_internal::compare_result_as_ordering(0) == 0));
+  EXPECT_FALSE(
+      Identity(absl::compare_internal::compare_result_as_ordering(0) > 0));
   EXPECT_FALSE(Identity(absl::compare_internal::compare_result_as_ordering(
-                           weak_ordering::equivalent) < 0));
+                            weak_ordering::equivalent) < 0));
   EXPECT_TRUE(Identity(absl::compare_internal::compare_result_as_ordering(
-                            weak_ordering::equivalent) == 0));
+                           weak_ordering::equivalent) == 0));
   EXPECT_FALSE(Identity(absl::compare_internal::compare_result_as_ordering(
                             weak_ordering::equivalent) > 0));
 
-  EXPECT_FALSE(Identity(
-      absl::compare_internal::compare_result_as_ordering(1) < 0));
-  EXPECT_FALSE(Identity(
-      absl::compare_internal::compare_result_as_ordering(1) == 0));
-  EXPECT_TRUE(Identity(
-      absl::compare_internal::compare_result_as_ordering(1) > 0));
+  EXPECT_FALSE(
+      Identity(absl::compare_internal::compare_result_as_ordering(1) < 0));
+  EXPECT_FALSE(
+      Identity(absl::compare_internal::compare_result_as_ordering(1) == 0));
+  EXPECT_TRUE(
+      Identity(absl::compare_internal::compare_result_as_ordering(1) > 0));
   EXPECT_FALSE(Identity(absl::compare_internal::compare_result_as_ordering(
-                           weak_ordering::greater) < 0));
+                            weak_ordering::greater) < 0));
   EXPECT_FALSE(Identity(absl::compare_internal::compare_result_as_ordering(
                             weak_ordering::greater) == 0));
   EXPECT_TRUE(Identity(absl::compare_internal::compare_result_as_ordering(
-                            weak_ordering::greater) > 0));
+                           weak_ordering::greater) > 0));
 }
 
 TEST(DoThreeWayComparison, SanityTest) {