Merge branch 'develop' into benchmarks/value-operations

Signed-off-by: Niels Lohmann <mail@nlohmann.me>

tests/benchmarks/src/benchmarks.cpp

@@ -8,10 +8,13 @@
 
 #include <benchmark/benchmark.h>
 #include <nlohmann/json.hpp>
+#include <array>
 #include <cstdint>
+#include <deque>
 #include <fstream>
 #include <map>
 #include <numeric>
+#include <set>
 #include <string>
 #include <vector>
 #include <test_data.hpp>
@@ -1144,4 +1147,27 @@ static void ToJsonStrings(benchmark::State& state, const bool move)
 BENCHMARK_CAPTURE(ToJsonStrings, copy, false);
 BENCHMARK_CAPTURE(ToJsonStrings, move, true);
 
+// conversion of a JSON array into standard containers
+template<typename Container>
+static void GetArray(benchmark::State& state)
+{
+    json j = json::array();
+    for (int i = 0; i < 1000; ++i)
+    {
+        j.push_back(i);
+    }
+
+    for (auto _ : state)
+    {
+        auto c = j.get<Container>();
+        benchmark::DoNotOptimize(c);
+    }
+    state.SetItemsProcessed(state.iterations() * 1000);
+}
+
+BENCHMARK_TEMPLATE(GetArray, std::vector<int>);
+BENCHMARK_TEMPLATE(GetArray, std::deque<int>);
+BENCHMARK_TEMPLATE(GetArray, std::set<int>);
+BENCHMARK_TEMPLATE(GetArray, std::array<int, 1000>);
+
 BENCHMARK_MAIN();

tests/fuzzing.md

@@ -9,7 +9,16 @@ Each fuzzer driver (`tests/src/fuzzer-parse_*.cpp`) parses its input twice: once
 once with exceptions. Both calls must agree. Where parsing with exceptions fails, the call without exceptions must
 return a discarded value (or throw the same kind of non-parse error), and it must never throw a `parse_error`. Where
 parsing succeeds, both calls must return the same value. The drivers then serialize the value, parse the result back,
-and check that nothing was lost. The drivers check all of this with `assert`, so they refuse to build with `NDEBUG`.
+and check that nothing was lost.
+
+Finally, each driver reads the same input through the other input adapters: a `std::istringstream`, a non-contiguous
+iterator range (`std::list<std::uint8_t>`), and, on POSIX systems, a `FILE*` from `fmemopen`. The pointer range takes the
+fast paths for contiguous input (bulk string scanning, multi-byte loads, lazy position tracking), while the other
+adapters read byte by byte. Every adapter must give the same value, or the same exception with the same id, message,
+and byte position. The JSON driver checks this for `parse` and `accept`, each with and without `ignore_comments`. The
+shared code is `check_input_adapters()` in `tests/src/fuzzer_common.hpp`.
+
+The drivers check all of this with `assert`, so they refuse to build with `NDEBUG`.
 
 ## Corpus creation
 

tests/src/fuzzer-parse_bjdata.cpp

@@ -47,12 +47,30 @@ dumps is stable under exactly the same values that break operator==.
 The unit tests run the same checks on a fixed corpus (see the "BJData round-trip
 invariants" test case), so keep both in sync.
 
+Finally, it reads the data from a stream, a non-contiguous iterator range, and
+a FILE*, which read byte by byte, and checks that each gives the same value or
+error (id, message, byte position) as the pointer range, which takes the bulk
+fast paths (see check_input_adapters()).
+
 The provided function `LLVMFuzzerTestOneInput` can be used in different fuzzer
 drivers.
 */
 
 #include "fuzzer_common.hpp"
 
+namespace
+{
+// json::from_bjdata for check_input_adapters()
+struct read_bjdata
+{
+    template<typename... Args>
+    json operator()(Args&& ... args) const
+    {
+        return json::from_bjdata(std::forward<Args>(args)...);
+    }
+};
+} // namespace
+
 // value-stable comparison for the round-trip checks below; see the note
 // above on why this compares dump()s rather than the json values directly
 static bool is_value_stable(const json& lhs, const json& rhs)
@@ -126,6 +144,9 @@ extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size)
         assert(parsed || noexcept_threw || j_noexcept.is_discarded());
     }
 
+    // step 3: the other input adapters must read the data alike
+    check_input_adapters(data, size, read_bjdata());
+
     // return 0 - non-zero return values are reserved for future use
     return 0;
 }

tests/src/fuzzer-parse_bon8.cpp

@@ -17,44 +17,33 @@ array data, it performs the following steps:
 - j2 = from_bon8(vec)
 - assert(to_bon8(j2) == vec)
 
-It also checks that reading the data from a stream, which reads strings byte by
-byte, gives the same value or error as reading it from contiguous memory, which
-copies strings in bulk.
+Finally, it reads the data from a stream, a non-contiguous iterator range, and
+a FILE*, which read byte by byte, and checks that each gives the same value or
+error (id, message, byte position) as the pointer range, which takes the bulk
+fast paths (see check_input_adapters()).
 
 The provided function `LLVMFuzzerTestOneInput` can be used in different fuzzer
 drivers.
 */
 
-#include <sstream>
 #include "fuzzer_common.hpp"
 
 namespace
 {
-// the serialization of the value read from @a input, or the error message
-template<typename InputType>
-std::string read_bon8(InputType&& input)
+// json::from_bon8 for check_input_adapters()
+struct read_bon8
 {
-    try
+    template<typename... Args>
+    json operator()(Args&& ... args) const
     {
-        const auto vec = json::to_bon8(json::from_bon8(std::forward<InputType>(input)));
-        return {vec.begin(), vec.end()};
+        return json::from_bon8(std::forward<Args>(args)...);
     }
-    catch (const json::exception& e)
-    {
-        return e.what();
-    }
-}
+};
 } // namespace
 
 // see http://llvm.org/docs/LibFuzzer.html
 extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size)
 {
-    // contiguous and stream input must be read alike
-    {
-        std::istringstream stream(std::string(reinterpret_cast<const char*>(data), size));
-        assert(read_bon8(std::vector<uint8_t>(data, data + size)) == read_bon8(stream));
-    }
-
     std::vector<uint8_t> const vec1(data, data + size);
 
     // step 0: parse input without exceptions
@@ -106,6 +95,9 @@ extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size)
         assert(parsed || noexcept_threw || j_noexcept.is_discarded());
     }
 
+    // step 3: the other input adapters must read the data alike
+    check_input_adapters(data, size, read_bon8());
+
     // return 0 - non-zero return values are reserved for future use
     return 0;
 }

tests/src/fuzzer-parse_bson.cpp

@@ -17,12 +17,30 @@ array data, it performs the following steps:
 - j2 = from_bson(vec)
 - assert(to_bson(j2) == vec)
 
+Finally, it reads the data from a stream, a non-contiguous iterator range, and
+a FILE*, which read byte by byte, and checks that each gives the same value or
+error (id, message, byte position) as the pointer range, which takes the bulk
+fast paths (see check_input_adapters()).
+
 The provided function `LLVMFuzzerTestOneInput` can be used in different fuzzer
 drivers.
 */
 
 #include "fuzzer_common.hpp"
 
+namespace
+{
+// json::from_bson for check_input_adapters()
+struct read_bson
+{
+    template<typename... Args>
+    json operator()(Args&& ... args) const
+    {
+        return json::from_bson(std::forward<Args>(args)...);
+    }
+};
+} // namespace
+
 // see http://llvm.org/docs/LibFuzzer.html
 extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size)
 {
@@ -77,6 +95,9 @@ extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size)
         assert(parsed || noexcept_threw || j_noexcept.is_discarded());
     }
 
+    // step 3: the other input adapters must read the data alike
+    check_input_adapters(data, size, read_bson());
+
     // return 0 - non-zero return values are reserved for future use
     return 0;
 }

tests/src/fuzzer-parse_cbor.cpp

@@ -17,12 +17,30 @@ array data, it performs the following steps:
 - j2 = from_cbor(vec)
 - assert(to_cbor(j2) == vec)
 
+Finally, it reads the data from a stream, a non-contiguous iterator range, and
+a FILE*, which read byte by byte, and checks that each gives the same value or
+error (id, message, byte position) as the pointer range, which takes the bulk
+fast paths (see check_input_adapters()).
+
 The provided function `LLVMFuzzerTestOneInput` can be used in different fuzzer
 drivers.
 */
 
 #include "fuzzer_common.hpp"
 
+namespace
+{
+// json::from_cbor for check_input_adapters()
+struct read_cbor
+{
+    template<typename... Args>
+    json operator()(Args&& ... args) const
+    {
+        return json::from_cbor(std::forward<Args>(args)...);
+    }
+};
+} // namespace
+
 // see http://llvm.org/docs/LibFuzzer.html
 extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size)
 {
@@ -77,6 +95,9 @@ extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size)
         assert(parsed || noexcept_threw || j_noexcept.is_discarded());
     }
 
+    // step 3: the other input adapters must read the data alike
+    check_input_adapters(data, size, read_cbor());
+
     // return 0 - non-zero return values are reserved for future use
     return 0;
 }

tests/src/fuzzer-parse_json.cpp

@@ -17,6 +17,12 @@ array data, it performs the following steps:
 - j2 = parse(s1)
 - s2 = serialize(j2)
 - assert(s1 == s2)
+- for ignore_comments = false and true: assert(parse(data) and accept(data)
+  give the same result for a stream, a non-contiguous iterator range, and a
+  FILE* as for the pointer range)
+
+The pointer range takes the bulk fast paths, the other input adapters read byte
+by byte; see check_input_adapters().
 
 The provided function `LLVMFuzzerTestOneInput` can be used in different fuzzer
 drivers.
@@ -24,6 +30,32 @@ drivers.
 
 #include "fuzzer_common.hpp"
 
+namespace
+{
+// json::parse and json::accept for check_input_adapters()
+struct parse_json
+{
+    bool ignore_comments;
+
+    template<typename... Args>
+    json operator()(Args&& ... args) const
+    {
+        return json::parse(std::forward<Args>(args)..., nullptr, true, ignore_comments);
+    }
+};
+
+struct accept_json
+{
+    bool ignore_comments;
+
+    template<typename... Args>
+    bool operator()(Args&& ... args) const
+    {
+        return json::accept(std::forward<Args>(args)..., ignore_comments);
+    }
+};
+} // namespace
+
 // see http://llvm.org/docs/LibFuzzer.html
 extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size)
 {
@@ -76,6 +108,16 @@ extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size)
         assert(parsed || noexcept_threw || j_noexcept.is_discarded());
     }
 
+    // step 3: the other input adapters must read the data alike
+    for (const bool ignore_comments :
+            {
+                false, true
+            })
+    {
+        check_input_adapters(data, size, parse_json{ignore_comments});
+        check_input_adapters(data, size, accept_json{ignore_comments});
+    }
+
     // return 0 - non-zero return values are reserved for future use
     return 0;
 }

tests/src/fuzzer-parse_msgpack.cpp

@@ -17,12 +17,30 @@ array data, it performs the following steps:
 - j2 = from_msgpack(vec)
 - assert(to_msgpack(j2) == vec)
 
+Finally, it reads the data from a stream, a non-contiguous iterator range, and
+a FILE*, which read byte by byte, and checks that each gives the same value or
+error (id, message, byte position) as the pointer range, which takes the bulk
+fast paths (see check_input_adapters()).
+
 The provided function `LLVMFuzzerTestOneInput` can be used in different fuzzer
 drivers.
 */
 
 #include "fuzzer_common.hpp"
 
+namespace
+{
+// json::from_msgpack for check_input_adapters()
+struct read_msgpack
+{
+    template<typename... Args>
+    json operator()(Args&& ... args) const
+    {
+        return json::from_msgpack(std::forward<Args>(args)...);
+    }
+};
+} // namespace
+
 // see http://llvm.org/docs/LibFuzzer.html
 extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size)
 {
@@ -77,6 +95,9 @@ extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size)
         assert(parsed || noexcept_threw || j_noexcept.is_discarded());
     }
 
+    // step 3: the other input adapters must read the data alike
+    check_input_adapters(data, size, read_msgpack());
+
     // return 0 - non-zero return values are reserved for future use
     return 0;
 }

tests/src/fuzzer-parse_ubjson.cpp

@@ -26,12 +26,30 @@ array data, it performs the following steps:
 The unit tests run the same checks on a fixed corpus (see the "UBJSON round-trip
 invariants" test case), so keep both in sync.
 
+Finally, it reads the data from a stream, a non-contiguous iterator range, and
+a FILE*, which read byte by byte, and checks that each gives the same value or
+error (id, message, byte position) as the pointer range, which takes the bulk
+fast paths (see check_input_adapters()).
+
 The provided function `LLVMFuzzerTestOneInput` can be used in different fuzzer
 drivers.
 */
 
 #include "fuzzer_common.hpp"
 
+namespace
+{
+// json::from_ubjson for check_input_adapters()
+struct read_ubjson
+{
+    template<typename... Args>
+    json operator()(Args&& ... args) const
+    {
+        return json::from_ubjson(std::forward<Args>(args)...);
+    }
+};
+} // namespace
+
 // see http://llvm.org/docs/LibFuzzer.html
 extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size)
 {
@@ -96,6 +114,9 @@ extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size)
         assert(parsed || noexcept_threw || j_noexcept.is_discarded());
     }
 
+    // step 3: the other input adapters must read the data alike
+    check_input_adapters(data, size, read_ubjson());
+
     // return 0 - non-zero return values are reserved for future use
     return 0;
 }

tests/src/fuzzer_common.hpp

@@ -11,6 +11,14 @@
 #pragma once
 
 #include <cassert> // assert
+#include <cstddef> // size_t
+#include <cstdint> // uint8_t
+#include <cstdio> // FILE, fmemopen, fclose
+#include <list> // list
+#include <sstream> // istringstream
+#include <string> // string
+#include <utility> // forward
+#include <vector> // vector
 #include <nlohmann/json.hpp>
 
 using nlohmann::json; // NOLINT(google-global-names-in-headers): shared by all fuzzer drivers
@@ -49,3 +57,89 @@ json parse_without_exceptions(Parse parse, bool& threw)
     }
     return {};
 }
+
+// the outcome of reading an input with exceptions: the serialization of the
+// value, or the id, message, and (for parse errors) byte of the exception
+struct read_outcome
+{
+    std::string value{};
+    int id = 0;
+    std::string what{};
+    std::size_t byte = 0;
+
+    bool operator==(const read_outcome& rhs) const
+    {
+        return value == rhs.value && id == rhs.id && what == rhs.what && byte == rhs.byte;
+    }
+};
+
+inline std::string outcome_value(const json& j)
+{
+    return j.dump(-1, ' ', false, json::error_handler_t::keep);
+}
+
+inline std::string outcome_value(bool b)
+{
+    return b ? "true" : "false";
+}
+
+template<typename Read, typename... Args>
+read_outcome read_with(const Read& parse, Args&& ... args)
+{
+    read_outcome result;
+    try
+    {
+        result.value = outcome_value(parse(std::forward<Args>(args)...));
+    }
+    catch (const json::parse_error& e)
+    {
+        result.id = e.id;
+        result.what = e.what();
+        result.byte = e.byte;
+    }
+    catch (const json::exception& e)
+    {
+        result.id = e.id;
+        result.what = e.what();
+    }
+    return result;
+}
+
+// The contiguous input (a pointer range) takes the bulk fast paths; the other
+// input adapters read byte by byte. Reads @a data through each of them and
+// checks that all give the same value, or the same exception with the same
+// id, message, and byte position.
+//
+// @a parse is called with either an iterator range or a single input (stream
+// or FILE*) and returns the parsed value (or the result of accept()).
+template<typename Read>
+void check_input_adapters(const std::uint8_t* data, std::size_t size, const Read& parse)
+{
+    const read_outcome expected = read_with(parse, data, data + size);
+
+    // stream adapter
+    {
+        std::istringstream stream(std::string(reinterpret_cast<const char*>(data), size));
+        assert(read_with(parse, stream) == expected);
+    }
+
+    // iterator adapter on a non-contiguous range
+    {
+        const std::list<std::uint8_t> list(data, data + size);
+        assert(read_with(parse, list.begin(), list.end()) == expected);
+    }
+
+#if defined(__unix__) || defined(__APPLE__)
+    // FILE* adapter; fmemopen may refuse an empty buffer
+    {
+        std::vector<std::uint8_t> buffer(data, data + size);
+        std::FILE* file = buffer.empty() ? nullptr : fmemopen(buffer.data(), buffer.size(), "rb");
+        if (file != nullptr)
+        {
+            const read_outcome actual = read_with(parse, file);
+            std::fclose(file);
+            assert(actual == expected);
+        }
+    }
+#endif
+}

tests/src/unit-algorithms.cpp

@@ -9,6 +9,7 @@
 #include "doctest_compatibility.h"
 
 #include <algorithm>
+#include <numeric>
 #include <nlohmann/json.hpp>
 using nlohmann::json;
 

tests/src/unit-comparison.cpp

@@ -19,6 +19,7 @@
 
 #include <cctype>
 #include <cstdint>
+#include <limits>
 #include <map>
 #include <string>
 #include <utility>
@@ -1149,3 +1150,157 @@ TEST_CASE("operator<=> of binary values with a different subtype does not depend
     }
 }
 #endif
+
+TEST_CASE("comparison of scalars (inline fast path)")
+{
+    const json::number_integer_t neg = -1;
+    const json::number_integer_t pos = 5;
+    const json::number_unsigned_t big = (std::numeric_limits<json::number_unsigned_t>::max)();
+
+    SECTION("same type")
+    {
+        CHECK(json(pos) == json(pos));
+        CHECK(json(pos) != json(neg));
+        CHECK(json(neg) < json(pos));
+        CHECK_FALSE(json(pos) < json(neg));
+        CHECK_FALSE(json(pos) < json(pos));
+        CHECK(json(pos) <= json(pos));
+
+        CHECK(json(1u) == json(1u));
+        CHECK(json(1u) < json(2u));
+        CHECK_FALSE(json(2u) < json(1u));
+
+        CHECK(json(1.5) == json(1.5));
+        CHECK(json(1.5) < json(2.5));
+        CHECK_FALSE(json(2.5) < json(1.5));
+        CHECK(json(-0.0) == json(0.0));
+    }
+
+    SECTION("mixed number types")
+    {
+        // negative signed vs unsigned must not wrap around
+        CHECK(json(neg) < json(1u));
+        CHECK(json(1u) > json(neg));
+        CHECK(json(neg) != json(big));
+        CHECK(json(neg) < json(big));
+        CHECK_FALSE(json(big) < json(neg));
+
+        CHECK(json(pos) == json(5u));
+        CHECK(json(5u) == json(pos));
+        CHECK(json(pos) < json(6u));
+        CHECK(json(pos) == json(5.0));
+        CHECK(json(5.0) == json(pos));
+        CHECK(json(5u) == json(5.0));
+        CHECK(json(neg) < json(0.5));
+        CHECK(json(0.5) < json(1u));
+    }
+
+    SECTION("boolean and null")
+    {
+        CHECK(json(true) == json(true));
+        CHECK(json(true) != json(false));
+        CHECK(json(false) < json(true));
+        CHECK_FALSE(json(true) < json(false));
+        CHECK_FALSE(json(true) < json(true));
+
+        CHECK(json(nullptr) == json(nullptr));
+        CHECK_FALSE(json(nullptr) < json(nullptr));
+        CHECK(json(nullptr) <= json(nullptr));
+        CHECK(json(nullptr) < json(false));
+        CHECK(json(nullptr) != json(false));
+        CHECK(json(nullptr) != json(0));
+    }
+
+    SECTION("string vs number")
+    {
+        CHECK(json("5") != json(5));
+        CHECK(json(5) != json("5"));
+        CHECK(json(5) < json("5")); // number < string by type order
+        CHECK_FALSE(json("5") < json(5));
+        CHECK(json(1.5) < json(""));
+        CHECK(json("a") == json("a"));
+        CHECK(json("a") < json("b"));
+    }
+
+    SECTION("NaN")
+    {
+        const json nan = std::numeric_limits<json::number_float_t>::quiet_NaN();
+        const json one = 1.0;
+
+        CHECK_FALSE(nan == nan);
+        CHECK(nan != nan);
+        CHECK_FALSE(nan < nan);
+        CHECK_FALSE(nan <= nan);
+        CHECK_FALSE(nan >= nan);
+        CHECK_FALSE(nan > nan);
+
+        CHECK_FALSE(nan == one);
+        CHECK_FALSE(one == nan);
+        CHECK_FALSE(nan < one);
+        CHECK_FALSE(one < nan);
+        CHECK_FALSE(nan <= one);
+        CHECK_FALSE(one <= nan);
+        CHECK_FALSE(nan >= one);
+        CHECK_FALSE(one >= nan);
+
+        // NaN against other number types
+        CHECK_FALSE(nan == json(1));
+        CHECK_FALSE(json(1) == nan);
+        CHECK_FALSE(nan < json(1u));
+        CHECK_FALSE(json(1u) < nan);
+
+#if JSON_HAS_THREE_WAY_COMPARISON
+        CHECK((nan <=> nan) == std::partial_ordering::unordered); // *NOPAD*
+        CHECK((nan <=> one) == std::partial_ordering::unordered); // *NOPAD*
+        CHECK((one <=> nan) == std::partial_ordering::unordered); // *NOPAD*
+        CHECK((one <=> json(2.0)) == std::partial_ordering::less); // *NOPAD*
+        CHECK((json(2.0) <=> one) == std::partial_ordering::greater); // *NOPAD*
+        CHECK((one <=> one) == std::partial_ordering::equivalent); // *NOPAD*
+        CHECK((json(neg) <=> json(1u)) == std::partial_ordering::less); // *NOPAD*
+        CHECK((json(nullptr) <=> json(nullptr)) == std::partial_ordering::equivalent); // *NOPAD*
+        CHECK((json(true) <=> json(false)) == std::partial_ordering::greater); // *NOPAD*
+#endif
+    }
+
+    SECTION("discarded")
+    {
+        const json d(json::value_t::discarded);
+
+        CHECK_FALSE(d == d);
+        CHECK_FALSE(d < d);
+        CHECK_FALSE(d == json(1));
+        CHECK_FALSE(json(1) == d);
+        CHECK_FALSE(d < json(1));
+        CHECK_FALSE(json(1) < d);
+        CHECK_FALSE(d == json(nullptr));
+        CHECK_FALSE(json(nullptr) == d);
+
+#if JSON_HAS_THREE_WAY_COMPARISON
+        CHECK((d <=> d) == std::partial_ordering::unordered); // *NOPAD*
+        CHECK((d <=> json(1)) == std::partial_ordering::unordered); // *NOPAD*
+        CHECK((json(1) <=> d) == std::partial_ordering::unordered); // *NOPAD*
+#endif
+    }
+
+    SECTION("scalar overloads")
+    {
+        CHECK(json(5) == 5);
+        CHECK(5 == json(5));
+        CHECK(json(5) < 6);
+        CHECK(4 < json(5));
+        CHECK(json(1.5) == 1.5);
+        CHECK(json(true) == true);
+        CHECK(json(5) != 6);
+        CHECK(json(1u) == 1.0);
+    }
+
+    SECTION("structured values still compare")
+    {
+        CHECK(json({1, 2}) == json({1, 2}));
+        CHECK(json({1, 2}) < json({1, 3}));
+        CHECK(json({{"a", 1}}) == json({{"a", 1}}));
+        CHECK(json::binary({1, 2}) == json::binary({1, 2}));
+        CHECK(json(1) < json::array());
+        CHECK(json::object() < json::array());
+    }
+}

tests/src/unit-constructor2.cpp

@@ -8,6 +8,9 @@
 
 #include "doctest_compatibility.h"
 
+#include <string>
+#include <vector>
+
 #include <nlohmann/json.hpp>
 using nlohmann::json;
 
@@ -184,3 +187,193 @@ TEST_CASE("other constructors and destructor")
         }
     }
 }
+
+TEST_CASE("copy constructor and destructor across value types")
+{
+    SECTION("copy of a scalar is independent of the source")
+    {
+        const json src_null = nullptr;
+        const json src_bool = true;
+        const json src_int = -7;
+        const json src_uint = 7u;
+        const json src_float = 2.5;
+
+        json c_null(src_null);
+        json c_bool(src_bool);
+        json c_int(src_int);
+        json c_uint(src_uint);
+        json c_float(src_float);
+
+        CHECK(c_null.is_null());
+        CHECK(c_bool == true);
+        CHECK(c_int == -7);
+        CHECK(c_uint == 7u);
+        CHECK(c_float == 2.5);
+
+        c_int = 8;
+        c_float = 3.5;
+        CHECK(src_int == -7);
+        CHECK(src_float == 2.5);
+    }
+
+    SECTION("copy of an owning value, then mutate the copy")
+    {
+        const json src_string = "hello";
+        const json src_array = {1, 2, 3};
+        const json src_object = {{"a", 1}, {"b", {{"c", 2}}}};
+        const json src_binary = json::binary({1, 2, 3});
+
+        json c_string(src_string);
+        json c_array(src_array);
+        json c_object(src_object);
+        json c_binary(src_binary);
+
+        CHECK(c_string == src_string);
+        CHECK(c_array == src_array);
+        CHECK(c_object == src_object);
+        CHECK(c_binary == src_binary);
+
+        c_string.get_ref<json::string_t&>() += " world";
+        c_array.push_back(4);
+        c_object["b"]["c"] = 3;
+        c_object["d"] = 4;
+        c_binary.get_binary().push_back(4);
+
+        CHECK(src_string == "hello");
+        CHECK(src_array.size() == 3);
+        CHECK(src_object["b"]["c"] == 2);
+        CHECK_FALSE(src_object.contains("d"));
+        CHECK(src_binary.get_binary().size() == 3);
+
+        CHECK(c_string == "hello world");
+        CHECK(c_array.size() == 4);
+        CHECK(c_object["b"]["c"] == 3);
+        CHECK(c_binary.get_binary().size() == 4);
+    }
+
+    SECTION("copy keeps parent pointers valid (JSON_DIAGNOSTICS)")
+    {
+        const json src = {{"a", {1, 2}}, {"b", "x"}};
+        json c(src);
+        c["a"].push_back(3);
+        CHECK(c["a"].size() == 3);
+        CHECK(src["a"].size() == 2);
+    }
+
+    SECTION("vector of mixed types: build, copy, and destroy")
+    {
+        {
+            std::vector<json> v;
+            for (int i = 0; i < 70; ++i)
+            {
+                switch (i % 9)
+                {
+                    case 0:
+                        v.emplace_back(nullptr);
+                        break;
+                    case 1:
+                        v.emplace_back(i % 2 == 0);
+                        break;
+                    case 2:
+                        v.emplace_back(-i);
+                        break;
+                    case 3:
+                        v.emplace_back(static_cast<json::number_unsigned_t>(i));
+                        break;
+                    case 4:
+                        v.emplace_back(i * 0.5);
+                        break;
+                    case 5:
+                        v.emplace_back(std::string(static_cast<std::size_t>(i), 'x'));
+                        break;
+                    case 6:
+                        v.emplace_back(json::array({i, "s", nullptr}));
+                        break;
+                    case 7:
+                        v.emplace_back(json::object({{"k", i}}));
+                        break;
+                    default:
+                        v.emplace_back(json::binary({1, 2, 3}));
+                        break;
+                }
+            }
+            REQUIRE(v.size() == 70);
+
+            std::vector<json> copy(v);
+            CHECK(copy == v);
+
+            // mutating the copy leaves the original alone
+            copy[5] = "changed";
+            copy[6].push_back(1);
+            CHECK(copy != v);
+            CHECK(v[5] == std::string(5, 'x'));
+            CHECK(v[6].size() == 3);
+
+            v.erase(v.begin(), v.begin() + 10);
+            CHECK(v.size() == 60);
+        }
+
+        // destroying the discarded and default-constructed values
+        std::vector<json> w(4);
+        w.emplace_back(json::value_t::discarded);
+        w.emplace_back(json::value_t::object);
+        w.emplace_back(json::value_t::array);
+        w.emplace_back(json::value_t::string);
+        w.emplace_back(json::value_t::binary);
+        CHECK(w.size() == 9);
+    }
+}
+
+TEST_CASE_TEMPLATE("destructor of nested containers", Json, nlohmann::json, nlohmann::ordered_json)
+{
+    SECTION("empty containers")
+    {
+        {
+            Json j = Json::array();
+        }
+        {
+            Json j = Json::object();
+        }
+        {
+            Json j = Json::array({Json::array(), Json::object(), Json::array({Json::array()})});
+        }
+    }
+
+    SECTION("leaves mixed with nested containers")
+    {
+        Json j = Json::object();
+        j["leaf"] = 1;
+        j["str"] = "text";
+        j["empty_a"] = Json::array();
+        j["empty_o"] = Json::object();
+        j["arr"] = Json::array({1, "a", Json::array({1, 2}), Json::object({{"k", Json::array({Json::object()})}}), nullptr});
+        j["obj"] = Json::object({{"x", Json::object({{"y", Json::array({1, Json::object({{"z", 1}})})}})}, {"w", 2}});
+        Json copy = j;
+        CHECK(copy == j);
+        j = Json::array({std::move(copy), j, 3, Json::array()});
+    }
+
+    SECTION("deep nesting")
+    {
+        for (int kind = 0; kind < 3; ++kind)
+        {
+            Json j = 1;
+            for (int i = 0; i < 5000; ++i)
+            {
+                switch (kind)
+                {
+                    case 0:
+                        j = Json::array({std::move(j)});
+                        break;
+                    case 1:
+                        j = Json::object({{"k", std::move(j)}});
+                        break;
+                    default:
+                        // siblings next to the nested child
+                        j = (i % 2 == 0) ? Json::array({1, Json::array(), std::move(j)}) : Json::object({{"a", 1}, {"b", Json::object()}, {"c", std::move(j)}});
+                        break;
+                }
+            }
+        }
+    }
+}

tests/src/unit-diagnostics.cpp

@@ -13,9 +13,11 @@
 #endif
 
 #define JSON_DIAGNOSTICS 1
+#define JSON_TESTS_PRIVATE
 
 #include <nlohmann/json.hpp>
 using nlohmann::json;
+using nlohmann::ordered_json;
 
 #include <map>
 #include <unordered_map>
@@ -631,3 +633,238 @@ TEST_CASE("Better diagnostics past the descent bound of update(), merge_patch()
         CHECK_THROWS_WITH_AS(p->at("y").at(0), expected_y.c_str(), json::type_error);
     }
 }
+
+namespace
+{
+// check that every child's parent pointer points to the containing value
+template<typename BasicJsonType>
+bool parents_consistent(const BasicJsonType& j)
+{
+    bool ok = true;
+    if (j.is_array() || j.is_object())
+    {
+        for (const auto& child : j)
+        {
+            ok = ok && child.m_parent == &j && parents_consistent(child);
+        }
+    }
+    return ok;
+}
+} // namespace
+
+TEST_CASE_TEMPLATE("Parent pointers stay consistent", J, json, ordered_json)
+{
+    SECTION("parse, copy, move, assign, swap")
+    {
+        J j = J::parse(R"({"a":[1,{"b":[2,3]},[4]],"c":{"d":{"e":[]}}})");
+        CHECK(parents_consistent(j));
+
+        J copy(j);
+        CHECK(parents_consistent(copy));
+
+        J moved(std::move(copy));
+        CHECK(parents_consistent(moved));
+
+        J assigned;
+        assigned = j;
+        CHECK(parents_consistent(assigned));
+
+        J other = {1, 2, {{"k", J::array({1, 2})}}};
+        other.swap(assigned);
+        CHECK(parents_consistent(other));
+        CHECK(parents_consistent(assigned));
+    }
+
+    SECTION("push_back growth")
+    {
+        J arr = J::array();
+        for (int i = 0; i < 500; ++i)
+        {
+            arr.push_back(J::array({i}));
+        }
+        CHECK(parents_consistent(arr));
+
+        J obj = J::object();
+        for (int i = 0; i < 500; ++i)
+        {
+            obj.push_back({std::to_string(i), J::array({i})});
+        }
+        CHECK(parents_consistent(obj));
+    }
+
+    SECTION("emplace_back and emplace growth")
+    {
+        J arr = J::array();
+        J obj = J::object();
+        for (int i = 0; i < 500; ++i)
+        {
+            arr.emplace_back(J::array({i}));
+            obj.emplace(std::to_string(i), J::array({i}));
+        }
+        CHECK(parents_consistent(arr));
+        CHECK(parents_consistent(obj));
+    }
+
+    SECTION("operator[] growth and at()")
+    {
+        J obj = J::object();
+        for (int i = 0; i < 500; ++i)
+        {
+            obj[std::to_string(i)] = J::array({i});
+            CHECK(obj[std::to_string(i / 2)].m_parent == &obj);
+        }
+        CHECK(parents_consistent(obj));
+
+        J arr = J::array();
+        for (int i = 0; i < 500; ++i)
+        {
+            arr[static_cast<size_t>(i)] = J::object({{"x", i}});
+        }
+        for (size_t i = 0; i < 500; ++i)
+        {
+            CHECK(arr.at(i).m_parent == &arr);
+        }
+        CHECK(obj.at("3").m_parent == &obj);
+        CHECK(parents_consistent(arr));
+        CHECK(parents_consistent(obj));
+    }
+}
+
+TEST_CASE_TEMPLATE("update() and merge_patch() keep parent pointers and diagnostic paths", J, json, ordered_json)
+{
+    // enough keys to make the vector behind ordered_json reallocate several times
+    const auto make_object = [](int first, int last)
+    {
+        J o = J::object();
+        for (int i = first; i < last; ++i)
+        {
+            o[std::to_string(i)] = J::object({{"x", J::array({i})}});
+        }
+        return o;
+    };
+
+    // const access, so operator[] cannot repair the parent pointers
+    const auto check_path = [](const J & j, const std::string & key)
+    {
+        const std::string expected = "[json.exception.type_error.304] (/" + key + "/x) cannot use at() with array";
+        CHECK_THROWS_WITH_AS(j[key]["x"].at("y"), expected.c_str(), json::type_error);
+    };
+
+    SECTION("update(const_reference)")
+    {
+        J j = make_object(0, 200);
+        j.update(make_object(100, 400));
+        CHECK(j.size() == 400);
+        CHECK(parents_consistent(j));
+        check_path(j, "0");
+        check_path(j, "150");
+        check_path(j, "399");
+    }
+
+    SECTION("update(const_reference, merge_objects)")
+    {
+        J j = make_object(0, 200);
+        j["50"]["y"] = 1;
+        J other = make_object(100, 400);
+        other["50"] = J::object({{"z", J::array({1})}});
+        j.update(other, true);
+        CHECK(j.size() == 400);
+        CHECK(j["50"].size() == 3);
+        CHECK(parents_consistent(j));
+        check_path(j, "0");
+        check_path(j, "399");
+    }
+
+    SECTION("update(first, last)")
+    {
+        J j = make_object(0, 200);
+        J other = make_object(100, 400);
+        j.update(other.begin(), other.end());
+        CHECK(j.size() == 400);
+        CHECK(parents_consistent(j));
+        check_path(j, "0");
+        check_path(j, "399");
+
+        J k = make_object(0, 200);
+        k.update(other.begin(), other.end(), true);
+        CHECK(k.size() == 400);
+        CHECK(parents_consistent(k));
+        check_path(k, "399");
+    }
+
+    SECTION("merge_patch()")
+    {
+        J j = make_object(0, 200);
+        J patch = make_object(100, 400);
+        patch["0"] = nullptr;
+        j.merge_patch(patch);
+        CHECK(j.size() == 399);
+        CHECK_FALSE(j.contains("0"));
+        CHECK(parents_consistent(j));
+        check_path(j, "1");
+        check_path(j, "399");
+    }
+}
+
+namespace
+{
+// an object type that, like a sorted flat map, moves its members when a key is
+// added even if its capacity does not change; it has a capacity() member, but
+// is neither node-based nor nlohmann::ordered_map, so the library must not
+// trust its capacity to tell whether the members moved
+template<class Key, class T, class IgnoredLess = std::less<Key>,
+         class Allocator = std::allocator<std::pair<const Key, T>>>
+             struct relocating_map : nlohmann::ordered_map<Key, T, IgnoredLess, Allocator>
+{
+    using base_type = nlohmann::ordered_map<Key, T, IgnoredLess, Allocator>;
+    using typename base_type::iterator;
+    using typename base_type::Container;
+    using base_type::base_type;
+
+    template<typename... Args>
+    std::pair<iterator, bool> emplace(Args&& ... args)
+    {
+        auto result = base_type::emplace(std::forward<Args>(args)...);
+        const auto index = result.first - this->begin();
+        if (result.second)
+        {
+            // move every member to a new buffer of the same capacity
+            Container relocated;
+            relocated.reserve(this->capacity());
+            for (auto& member : *this)
+            {
+                relocated.emplace_back(member.first, std::move(member.second));
+            }
+            this->Container::swap(relocated);
+        }
+        return {this->begin() + index, result.second};
+    }
+};
+
+using relocating_json = nlohmann::basic_json<relocating_map>;
+} // namespace
+
+TEST_CASE("Parent pointers stay consistent for object types not known to keep their members in place")
+{
+    relocating_json obj = relocating_json::object();
+    for (int i = 0; i < 100; ++i)
+    {
+        obj[std::to_string(i)] = relocating_json::array({i});
+        obj.emplace(std::to_string(i + 1000), relocating_json::array({i}));
+    }
+    CHECK(parents_consistent(obj));
+
+    obj.erase("0");
+    CHECK(parents_consistent(obj));
+
+    relocating_json other = relocating_json::object();
+    for (int i = 0; i < 100; ++i)
+    {
+        other[std::to_string(i + 2000)] = relocating_json::array({i});
+    }
+    obj.update(other);
+    CHECK(parents_consistent(obj));
+
+    obj["1"].push_back(42);
+    CHECK_THROWS_WITH_AS(obj["1"].at(5), "[json.exception.out_of_range.401] (/1) array index 5 is out of range", relocating_json::out_of_range&);
+}

tests/src/unit-iterators1.cpp

@@ -10,6 +10,8 @@
 
 #define JSON_TESTS_PRIVATE
 #include <nlohmann/json.hpp>
+#include <string>
+#include <vector>
 using nlohmann::json;
 
 TEST_CASE("iterators 1")
@@ -1628,3 +1630,108 @@ TEST_CASE("iterators 1")
         }
     }
 }
+
+TEST_CASE_TEMPLATE("reverse iteration and member access", Json, nlohmann::json, nlohmann::ordered_json)
+{
+    SECTION("array: -- and rbegin/rend")
+    {
+        Json j = {1, 2, 3, 4};
+
+        auto it = j.end();
+        std::vector<int> seen;
+        while (it != j.begin())
+        {
+            --it;
+            seen.push_back(it->template get<int>());
+        }
+        CHECK(seen == std::vector<int> {4, 3, 2, 1});
+
+        auto it2 = j.end();
+        auto old = it2--;
+        CHECK(old == j.end());
+        CHECK(*it2 == 4);
+
+        seen.clear();
+        for (auto r = j.rbegin(); r != j.rend(); ++r)
+        {
+            seen.push_back(r->template get<int>());
+        }
+        CHECK(seen == std::vector<int> {4, 3, 2, 1});
+
+        seen.clear();
+        for (auto r = j.crbegin(); r != j.crend(); ++r)
+        {
+            seen.push_back(*r);
+        }
+        CHECK(seen == std::vector<int> {4, 3, 2, 1});
+    }
+
+    SECTION("object: -- and rbegin/rend")
+    {
+        Json j = {{"a", 1}, {"b", 2}, {"c", 3}};
+        // order of keys depends on the object type; use the forward order as reference
+        std::vector<std::string> fwd;
+        for (auto it = j.begin(); it != j.end(); ++it)
+        {
+            fwd.push_back(it.key());
+        }
+        std::vector<std::string> rev(fwd.rbegin(), fwd.rend());
+
+        std::vector<std::string> seen;
+        auto it = j.end();
+        while (it != j.begin())
+        {
+            --it;
+            seen.push_back(it.key());
+        }
+        CHECK(seen == rev);
+
+        auto it2 = j.end();
+        auto old = it2--;
+        CHECK(old == j.end());
+        CHECK(it2.key() == rev.front());
+
+        seen.clear();
+        for (auto r = j.rbegin(); r != j.rend(); ++r)
+        {
+            seen.push_back(r.key());
+        }
+        CHECK(seen == rev);
+    }
+
+    SECTION("object: operator-> reads and writes the mapped value")
+    {
+        Json j = {{"a", 1}, {"s", "str"}};
+        auto it = j.find("s");
+        REQUIRE(it != j.end());
+        CHECK(it->is_string());
+        CHECK(it->size() == 1);
+        *it = Json::array({1, 2});
+        CHECK(it->is_array());
+        it->push_back(3);
+        CHECK(j["s"].size() == 3);
+
+        const Json& cj = j;
+        auto cit = cj.find("a");
+        REQUIRE(cit != cj.end());
+        CHECK(cit->is_number_integer());
+        CHECK(cit->template get<int>() == 1);
+    }
+
+    SECTION("primitives: -- and ->")
+    {
+        Json j = 42;
+        auto it = j.end();
+        --it;
+        CHECK(it == j.begin());
+        CHECK(it->template get<int>() == 42);
+    }
+
+    SECTION("comparing iterators of different containers throws")
+    {
+        Json a = {1, 2};
+        Json b = {1, 2};
+        CHECK_THROWS_WITH_AS(a.begin() == b.begin(), "[json.exception.invalid_iterator.212] cannot compare iterators of different containers", nlohmann::json::invalid_iterator);
+        CHECK_THROWS_WITH_AS(a.begin() < b.begin(), "[json.exception.invalid_iterator.212] cannot compare iterators of different containers", nlohmann::json::invalid_iterator);
+    }
+}

tests/src/unit-json_patch.cpp

@@ -2066,3 +2066,40 @@ TEST_CASE("JSON patch - every operation on ordered_json")
         }
     }
 }
+
+TEST_CASE_TEMPLATE("JSON patch - diff round trip on hand-made documents", Json, nlohmann::json, nlohmann::ordered_json)
+{
+    const auto roundtrip = [](const Json & a, const Json & b)
+    {
+        CAPTURE(a.dump())
+        CAPTURE(b.dump())
+        const Json p = Json::diff(a, b);
+        CHECK(a.patch(p) == b);
+        CHECK(Json::diff(b, b).empty());
+    };
+
+    SECTION("reordered keys")
+    {
+        const Json a = Json::parse(R"({"x":1,"y":2,"z":{"p":1,"q":2}})");
+        const Json b = Json::parse(R"({"z":{"q":2,"p":1},"y":2,"x":1})");
+        roundtrip(a, b);
+        roundtrip(b, a);
+    }
+
+    SECTION("nested add, remove and replace")
+    {
+        const Json a = Json::parse(R"({"a":[1,2,{"k":"v","gone":true}],"b":{"c":{"d":1}},"e":"s"})");
+        const Json b = Json::parse(R"({"a":[1,3,{"k":"w","new":[]}],"b":{"c":{"d":[1,2]}},"f":null})");
+        roundtrip(a, b);
+        roundtrip(b, a);
+    }
+
+    SECTION("arrays of different length and type changes")
+    {
+        roundtrip(Json::parse("[1,2,3,4,5]"), Json::parse("[1,3]"));
+        roundtrip(Json::parse("[1]"), Json::parse("[1,[2],{\"a\":3}]"));
+        roundtrip(Json::parse("[]"), Json::parse("{}"));
+        roundtrip(Json::parse("{\"a\":[]}"), Json::parse("{\"a\":{}}"));
+        roundtrip(Json(1), Json::parse("[1]"));
+    }
+}

tests/src/unit-json_pointer.cpp

@@ -23,8 +23,14 @@ using nlohmann::json;
     using namespace nlohmann::literals; // NOLINT(google-build-using-namespace)
 #endif
 
+#include <cerrno>
+#include <cstdlib>
+#include <limits>
 #include <map>
+#include <random>
 #include <sstream>
+#include <string>
+#include <vector>
 
 TEST_CASE("JSON pointers")
 {
@@ -1071,3 +1077,242 @@ TEST_CASE("flatten of structured values")
               R"({"/z":"s","/a~1b/~0":1,"/a~1b/k":null,"/a/0":1,"/a/1/c":null,"/a/1/b":null,"/a/2/0/x~0~1/0":true,"/a/2/0/x~0~1/1":null,"/a/2/0/w":2})");
     }
 }
+
+TEST_CASE("flatten of ordered_json")
+{
+    using nlohmann::ordered_json;
+
+    SECTION("keeps the document order of the leaves")
+    {
+        const ordered_json j = ordered_json::parse(R"({
+            "z": 1, "a": {"y": [true, null, {}], "b": "x", "": []},
+            "m~n": 2.5, "o/p": {"k": {}}, "0": [[1], [2, 3]]
+        })");
+
+        const ordered_json flat = j.flatten();
+        std::vector<std::string> keys;
+        for (const auto& el : flat.items())
+        {
+            keys.push_back(el.key());
+        }
+        const std::vector<std::string> expected =
+        {
+            "/z", "/a/y/0", "/a/y/1", "/a/y/2", "/a/b", "/a/", "/m~0n", "/o~1p/k", "/0/0/0", "/0/1/0", "/0/1/1"
+        };
+        CHECK(keys == expected);
+        CHECK(flat.size() == expected.size());
+        CHECK(flat["/m~0n"] == 2.5);
+        CHECK(flat["/a/y/2"].is_null());
+    }
+
+    SECTION("round-trips and agrees with json")
+    {
+        ordered_json j;
+        for (int i = 0; i < 200; ++i)
+        {
+            j["k" + std::to_string(i)] = {{"v", i}, {"l", {i, i + 1, "s"}}};
+        }
+        const ordered_json flat = j.flatten();
+        CHECK(flat.size() == 800);
+        CHECK(flat.unflatten() == j);
+
+        // same entries as for a sorted object
+        const nlohmann::json plain = nlohmann::json::parse(j.dump());
+        CHECK(nlohmann::json::parse(flat.dump()) == plain.flatten());
+    }
+
+    SECTION("primitive and empty root")
+    {
+        CHECK(ordered_json(1).flatten() == ordered_json({{"", 1}}));
+        CHECK(ordered_json::array().flatten() == ordered_json({{"", nullptr}}));
+    }
+}
+namespace
+{
+// reference implementation of the array index rules (RFC 6901, Sect. 4) based on strtoull
+enum class index_class
+{
+    ok,
+    leading_zero,
+    not_a_number,
+    unresolved,
+    exceeds_size_type
+};
+
+index_class classify_index(const std::string& s, std::size_t& idx)
+{
+    if (s.size() > 1 && s[0] == '0')
+    {
+        return index_class::leading_zero;
+    }
+    if (!s.empty() && s != "-" && !(s[0] >= '0' && s[0] <= '9'))
+    {
+        return index_class::not_a_number;
+    }
+    char* end = nullptr;
+    errno = 0;
+    const unsigned long long res = std::strtoull(s.c_str(), &end, 10); // NOLINT(runtime/int)
+    if (end == s.c_str() || errno == ERANGE || static_cast<std::size_t>(end - s.c_str()) != s.size())
+    {
+        return index_class::unresolved;
+    }
+    if (res >= static_cast<unsigned long long>((std::numeric_limits<std::size_t>::max)())) // NOLINT(runtime/int)
+    {
+        return index_class::exceeds_size_type;
+    }
+    idx = static_cast<std::size_t>(res);
+    return index_class::ok;
+}
+
+// the observable behavior of json_pointer::at() and contains() for an array index token
+void check_index_token(const std::string& token)
+{
+    const json array = {10, 20, 30};
+    const json::json_pointer ptr("/" + token);
+
+    std::size_t idx = 0;
+    const auto cls = classify_index(token, idx);
+    const bool exists = cls == index_class::ok && idx < array.size();
+
+    CAPTURE(token)
+    CHECK(array.contains(ptr) == (exists && token != "-"));
+
+    if (token == "-")
+    {
+        CHECK_THROWS_AS(array.at(ptr), json::out_of_range&);
+        return;
+    }
+
+    switch (cls)
+    {
+        case index_class::leading_zero:
+            CHECK_THROWS_AS(array.at(ptr), json::parse_error&);
+            break;
+        case index_class::not_a_number:
+            CHECK_THROWS_AS(array.at(ptr), json::parse_error&);
+            break;
+        case index_class::unresolved:
+        case index_class::exceeds_size_type:
+            CHECK_THROWS_AS(array.at(ptr), json::out_of_range&);
+            break;
+        case index_class::ok:
+        default:
+            if (exists)
+            {
+                CHECK(array.at(ptr) == array[idx]);
+            }
+            else
+            {
+                CHECK_THROWS_AS(array.at(ptr), json::out_of_range&);
+            }
+            break;
+    }
+}
+} // namespace
+
+TEST_CASE("JSON pointer array index tokens match strtoull semantics")
+{
+    SECTION("boundary values")
+    {
+        const std::string max_digits = std::to_string((std::numeric_limits<std::size_t>::max)());
+        const std::vector<std::string> tokens =
+        {
+            "", "-", "0", "1", "2", "3", "00", "01", "007", "+1", "-1", " 1", "1 ", "1a", "a1", "0x1", "1e1", "1.0",
+            "999999999", "1000000000", "99999999999999999", "999999999999999999", "1000000000000000000",
+            "9999999999999999999", "18446744073709551614", max_digits, "18446744073709551616",
+            "99999999999999999999999", "000000000000000000000"
+        };
+        for (const auto& token : tokens)
+        {
+            check_index_token(token);
+        }
+    }
+
+    SECTION("random tokens")
+    {
+        std::mt19937 gen(42);
+        const std::string alphabet = "0123456789012345678901234567890123456789-+ ex";
+        for (int i = 0; i < 5000; ++i)
+        {
+            std::string token(static_cast<std::size_t>(gen() % 23), '0');
+            for (auto& c : token)
+            {
+                c = alphabet[gen() % alphabet.size()];
+            }
+            check_index_token(token);
+        }
+    }
+}
+
+TEST_CASE("JSON pointer operations on pointers with escaped tokens")
+{
+    const std::vector<std::string> strings =
+    {
+        "", "/", "//", "/a", "/a/b", "/a~1b", "/~0", "/~01", "/~10", "/a~0~1/b/", "/0/1/2", "/ ", "/a/", "/~1/~0"
+    };
+
+    for (const auto& str : strings)
+    {
+        CAPTURE(str)
+        const json::json_pointer ptr(str);
+        CHECK(ptr.to_string() == str);
+        CHECK(ptr / "x" / "y" / std::size_t{3} == json::json_pointer(str + "/x/y/3"));
+        CHECK(ptr.parent_pointer() == (str.empty() ? ptr : json::json_pointer(str.substr(0, str.rfind('/')))));
+
+        // comparison with a string, in place, and with every other pointer
+#if !defined(JSON_TEST_DEPRECATED_FUNCTIONS_DELETED)
+        for (const auto& other : strings)
+        {
+            CAPTURE(other)
+            CHECK((ptr == other) == (str == other));
+            CHECK((other == ptr) == (str == other));
+            CHECK((ptr != other) == (str != other));
+        }
+        CHECK_THROWS_AS(ptr == "x", json::parse_error&);
+        CHECK_THROWS_AS(ptr == "/~", json::parse_error&);
+        CHECK_THROWS_AS(ptr == "/b/~2", json::parse_error&);
+#endif
+    }
+}
+
+TEST_CASE("JSON pointer lookups")
+{
+    json j = {{"a", {{"b", {1, 2, 3}}}}, {"", 7}, {"~/", {{"x", nullptr}}}};
+    const json& cj = j;
+
+    CHECK(cj[json::json_pointer("/a/b/2")] == 3);
+    CHECK(cj.at(json::json_pointer("/a/b/1")) == 2);
+    CHECK(cj.at(json::json_pointer("/")) == 7);
+    CHECK(cj.contains(json::json_pointer("/~0~1/x")));
+    CHECK(!cj.contains(json::json_pointer("/a/c")));
+    CHECK(cj.value(json::json_pointer("/a/b/0"), 5) == 1);
+    CHECK(cj.value(json::json_pointer("/a/b/3"), 5) == 5);
+    CHECK(cj.value(json::json_pointer("/a/zz"), 5) == 5);
+
+    CHECK_THROWS_AS(cj.at(json::json_pointer("/a/c")), json::out_of_range&);
+    CHECK_THROWS_AS(cj.at(json::json_pointer("/a/b/3")), json::out_of_range&);
+
+    // non-const access reuses existing entries and creates missing ones
+    j[json::json_pointer("/a/b/1")] = 20;
+    j[json::json_pointer("/n/m")] = 1;
+    CHECK(j["a"]["b"] == json({1, 20, 3}));
+    CHECK(j["n"]["m"] == 1);
+}
+
+TEST_CASE("JSON patch with moves and copies")
+{
+    json doc = {{"a", {{"b", {1, 2, 3}}}}, {"c", {{"d", "x"}}}};
+
+    CHECK(doc.patch(json::parse(R"([{"op":"move","from":"/a/b","path":"/c/e"}])")) ==
+          json::parse(R"({"a":{},"c":{"d":"x","e":[1,2,3]}})"));
+    CHECK(doc.patch(json::parse(R"([{"op":"copy","from":"/a/b","path":"/c/e"}])")) ==
+          json::parse(R"({"a":{"b":[1,2,3]},"c":{"d":"x","e":[1,2,3]}})"));
+    CHECK(doc.patch(json::parse(R"([{"op":"move","from":"/a/b/0","path":"/a/b/2"}])")) ==
+          json::parse(R"({"a":{"b":[2,3,1]},"c":{"d":"x"}})"));
+
+    // a failing move leaves a copy-patched document untouched
+    CHECK_THROWS_AS(doc.patch(json::parse(R"([{"op":"move","from":"","path":""}])")), json::out_of_range&);
+    json inplace = doc;
+    CHECK_THROWS_AS(inplace.patch_inplace(json::parse(R"([{"op":"move","from":"","path":""}])")), json::out_of_range&);
+    CHECK(inplace == doc);
+}

tests/src/unit-same-number-types.cpp

@@ -0,0 +1,201 @@
+//     __ _____ _____ _____
+//  __|  |   __|     |   | |  JSON for Modern C++ (supporting code)
+// |  |  |__   |  |  | | | |  version 3.12.0
+// |_____|_____|_____|_|___|  https://github.com/nlohmann/json
+//
+// SPDX-FileCopyrightText: 2013-2026 Niels Lohmann <https://nlohmann.me>
+// SPDX-License-Identifier: MIT
+
+#include "doctest_compatibility.h"
+
+#include <nlohmann/json.hpp>
+
+#include <cstdint>
+#include <functional>
+#include <limits>
+#include <string>
+#include <vector>
+
+// basic_json with the same type for number_integer_t and number_unsigned_t
+// (#2573): every integer is stored as number_integer
+
+namespace
+{
+
+template<typename IntegerType>
+using same_types_json = nlohmann::json::with_integers_t<IntegerType, IntegerType>;
+
+using json64 = same_types_json<std::int64_t>;
+using json32 = same_types_json<std::int32_t>;
+
+} // namespace
+
+TEST_CASE_TEMPLATE("same type for number_integer_t and number_unsigned_t", Json, json64, json32)
+{
+    using integer_t = typename Json::number_integer_t;
+    using limits = std::numeric_limits<integer_t>;
+
+    SECTION("parsing stores integers as number_integer")
+    {
+        const Json j = Json::parse("[0, 1, -1, 2147483647, -2147483648]");
+        for (const auto& element : j)
+        {
+            CHECK(element.is_number_integer());
+            CHECK(!element.is_number_unsigned());
+            CHECK(element.type() == nlohmann::detail::value_t::number_integer);
+        }
+        CHECK(j[1].template get<integer_t>() == 1);
+        CHECK(j[2].template get<integer_t>() == -1);
+        CHECK(j[3].template get<integer_t>() == 2147483647);
+        CHECK(j[4].template get<std::int64_t>() == -2147483648LL);
+    }
+
+    SECTION("integers above the maximum are parsed as floats")
+    {
+        // a positive literal is parsed as unsigned first; it must not wrap
+        // around to a negative number_integer_t
+        const auto above_max = std::to_string(static_cast<std::uint64_t>(limits::max()) + 1);
+        const Json j1 = Json::parse(above_max);
+        CHECK(j1.is_number_float());
+        CHECK(j1.template get<double>() == static_cast<double>(static_cast<std::uint64_t>(limits::max()) + 1));
+
+        const Json j2 = Json::parse("18446744073709551615");
+        CHECK(j2.is_number_float());
+        CHECK(j2.template get<double>() == 18446744073709551615.0);
+
+        const Json j3 = Json::parse(std::to_string(limits::max()));
+        CHECK(j3.is_number_integer());
+        CHECK(j3.template get<integer_t>() == limits::max());
+    }
+
+    SECTION("serialization of the extremes")
+    {
+        CHECK(Json(limits::max()).dump() == std::to_string(limits::max()));
+        // the absolute value of the minimum does not fit into number_unsigned_t
+        CHECK(Json(limits::min()).dump() == std::to_string(limits::min()));
+        CHECK(Json::parse(std::to_string(limits::min())) == Json(limits::min()));
+        CHECK(Json(0).dump() == "0");
+    }
+
+    SECTION("construction from unsigned types")
+    {
+        const Json j1 = 5u;
+        CHECK(j1.is_number_integer());
+        CHECK(!j1.is_number_unsigned());
+        CHECK(j1 == 5);
+
+        const Json j2 = static_cast<std::uint8_t>(200);
+        CHECK(j2.is_number_integer());
+        CHECK(j2.template get<integer_t>() == 200);
+
+        const Json j3 = std::vector<unsigned> {1, 2, 3};
+        CHECK(j3.dump() == "[1,2,3]");
+        CHECK(j3[0].is_number_integer());
+
+        const Json j4 = true;
+        CHECK(j4.is_boolean());
+    }
+
+    SECTION("value_t::number_unsigned creates a number_integer")
+    {
+        const Json j(nlohmann::detail::value_t::number_unsigned);
+        CHECK(j.is_number_integer());
+        CHECK(!j.is_number_unsigned());
+        CHECK(j == 0);
+        CHECK(j.dump() == "0");
+    }
+
+    SECTION("conversions")
+    {
+        Json j = 42;
+        CHECK(j.template get<unsigned>() == 42u);
+        CHECK(j.template get<std::uint64_t>() == 42u);
+        CHECK(j.template get<int>() == 42);
+        CHECK(j.template get<double>() == 42.0);
+        CHECK(j.template get_ptr<integer_t*>() != nullptr);
+        CHECK(*j.template get_ptr<integer_t*>() == 42);
+        CHECK(j.template get_ref<integer_t&>() == 42);
+        j.template get_ref<integer_t&>() = 43;
+        CHECK(j == 43);
+
+        integer_t i = 0;
+        j.get_to(i);
+        CHECK(i == 43);
+
+        const Json object = {{"a", 1}};
+        CHECK(object.value("a", 0u) == 1u);
+        CHECK(object.value("b", 7u) == 7u);
+    }
+
+    SECTION("conversion from a basic_json with distinct types")
+    {
+        const nlohmann::json j = {1, 2u, -3, 4.5};
+        CHECK(j[1].is_number_unsigned());
+        const Json converted = j;
+        CHECK(converted.dump() == "[1,2,-3,4.5]");
+        CHECK(converted[1].is_number_integer());
+        CHECK(!converted[1].is_number_unsigned());
+
+        const nlohmann::json back = converted;
+        CHECK(back == j);
+    }
+
+    SECTION("comparison and hashing")
+    {
+        const Json j1 = 1;
+        const Json j2 = 2u;
+        CHECK(j1 < j2);
+        CHECK(j1 != j2);
+        CHECK(j2 == 2);
+        CHECK(j2 == 2u);
+        CHECK(j1 < 1.5);
+        CHECK(Json(-1) < Json(0u));
+        CHECK(std::hash<Json> {}(Json(5u)) == std::hash<Json> {}(Json(5)));
+    }
+
+    SECTION("JSON Patch, diff, and flatten")
+    {
+        const Json source = Json::parse(R"({"a": [1, 2, 3], "b": 4})");
+        const Json target = Json::parse(R"({"a": [1, 5], "c": 6})");
+        CHECK(source.patch(Json::diff(source, target)) == target);
+        CHECK(source.flatten().unflatten() == source);
+    }
+
+    SECTION("binary formats")
+    {
+        const Json j = {{"min", limits::min()}, {"max", limits::max()}, {"zero", 0}, {"small", 100}, {"negative", -100}, {"float", 1.5}};
+
+        CHECK(Json::from_cbor(Json::to_cbor(j)) == j);
+        CHECK(Json::from_msgpack(Json::to_msgpack(j)) == j);
+        CHECK(Json::from_ubjson(Json::to_ubjson(j)) == j);
+        CHECK(Json::from_bjdata(Json::to_bjdata(j)) == j);
+        CHECK(Json::from_bson(Json::to_bson(j)) == j);
+        CHECK(Json::from_bon8(Json::to_bon8(j)) == j);
+
+        for (const auto& element : Json::from_cbor(Json::to_cbor(j)))
+        {
+            CHECK(!element.is_number_unsigned());
+        }
+
+        // CBOR unsigned integer 2^64-1 does not fit and is read as float
+        const std::vector<std::uint8_t> cbor_max = {0x1B, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF};
+        const Json from_cbor = Json::from_cbor(cbor_max);
+        CHECK(from_cbor.is_number_float());
+        CHECK(from_cbor.template get<double>() == 18446744073709551615.0);
+
+        // MessagePack uint 8 is read as number_integer
+        const std::vector<std::uint8_t> msgpack_uint8 = {0xCC, 0xC8};
+        const Json from_msgpack = Json::from_msgpack(msgpack_uint8);
+        CHECK(from_msgpack.is_number_integer());
+        CHECK(from_msgpack == 200);
+    }
+}
+
+TEST_CASE("same type for number_integer_t and number_unsigned_t: type traits")
+{
+    CHECK(std::is_same<nlohmann::detail::number_unsigned_overload_t<json64>, nlohmann::detail::same_as_number_integer_type>::value);
+    CHECK(std::is_same<nlohmann::detail::number_unsigned_overload_t<nlohmann::json>, std::uint64_t>::value);
+    CHECK(nlohmann::detail::is_compatible_number_integer_type<json64, unsigned>::value);
+    CHECK(!nlohmann::detail::is_compatible_number_integer_type<nlohmann::json, unsigned>::value);
+    CHECK(!nlohmann::detail::is_compatible_number_integer_type<json64, bool>::value);
+}

tests/src/unit-serialization.cpp

@@ -20,8 +20,11 @@
 using nlohmann::json;
 
 #include <array>
+#include <cstdint>
 #include <sstream>
 #include <iomanip>
+#include <string>
+#include <vector>
 
 #include "test_utils.hpp"
 
@@ -923,66 +926,68 @@ TEST_CASE("serialization boundary values for the string buffer")
     // mid-string once fewer than 13 bytes remain (`string_buffer.size() - bytes
     // < 13`), 13 being one more than the most a single code point can ever
     // write at once (a surrogate pair: two back-to-back "\uXXXX" escapes, 12
-    // bytes). Every write into string_buffer that this check protects happens
-    // in steps of 2 (a simple "\\x" escape) or 6 (one "\uXXXX" unit), so
-    // `bytes` only ever takes even values at the point the check runs - the
-    // tightest values actually reachable are therefore 498 (512 - 498 == 14,
-    // one simple escape away from the threshold) and 500 (512 - 500 == 12,
-    // where the flush fires immediately and resets bytes to 0).
+    // bytes). Characters below 0x60 (control characters, '"' and '\\') are
+    // escaped straight into the write buffer and never reach string_buffer, so
+    // under ensure_ascii it fills in steps of 6 (one "\uXXXX" unit for a
+    // character below U+10000) or 12 (a surrogate pair). With 2-byte characters
+    // such as U+00E4, `bytes` takes multiples of 6 at the point the check runs;
+    // the tightest values reachable are therefore 498 (512 - 498 == 14, after
+    // 83 characters) and 504 (512 - 504 == 8, after 84 characters, where the
+    // flush fires immediately and resets bytes to 0).
 
-    SECTION("a run of 2-byte escapes lands bytes on, and one step past, the flush threshold")
+    SECTION("a run of 6-byte escapes lands bytes on, and one step past, the flush threshold")
     {
         for (const int count :
                 {
-                    249, 250, 251
+                    83, 84, 85
                 })
         {
             CAPTURE(count)
-            const json j = std::string(static_cast<std::size_t>(count), '\n');
+            std::string umlauts;
             std::string expected = "\"";
             for (int i = 0; i < count; ++i)
             {
-                expected += "\\n";
+                umlauts += "\xC3\xA4";
+                expected += "\\u00e4";
             }
             expected += '"';
-            CHECK(j.dump() == expected);
+            CHECK(json(umlauts).dump(-1, ' ', true) == expected);
         }
     }
 
-    SECTION("an ASCII prefix leaves the tightest reachable margin before a 12-byte surrogate pair")
+    SECTION("a prefix leaves the tightest reachable margin before a 12-byte surrogate pair")
     {
         // U+1F600 (the "\xF0\x9F\x98\x80" UTF-8 bytes) is dumped under
         // ensure_ascii as the 12-byte surrogate pair "\ud83d\ude00"; that
         // write happens in a single step with no intermediate flush check, so
         // it is the write most exposed by an off-by-one in the "< 13" guard.
-        // A prefix of 249 newlines leaves exactly 14 bytes of headroom
+        // A prefix of 83 U+00E4 characters leaves exactly 14 bytes of headroom
         // (512 - 498), the smallest margin the guard ever actually allows
-        // into a new code point; 250 newlines instead trigger the guard's own
+        // into a new code point; 84 characters instead trigger the guard's own
         // flush first, so the emoji starts from a freshly emptied (512-byte)
-        // buffer, and 251 repeats that with one more escape already past the
+        // buffer, and 85 repeats that with one more escape already past the
         // reset. Together they cover the margin the guard allows landing on,
         // one step before, and one step after - all must still produce the
         // identical, correct escapes.
         for (const int prefix_count :
                 {
-                    249, 250, 251
+                    83, 84, 85
                 })
         {
             CAPTURE(prefix_count)
-            const std::string prefix(static_cast<std::size_t>(prefix_count), '\n');
-            const std::string emoji = "\xF0\x9F\x98\x80";
-            const json j = prefix + emoji;
-
+            std::string prefix;
             std::string expected_prefix;
             for (int i = 0; i < prefix_count; ++i)
             {
-                expected_prefix += "\\n";
+                prefix += "\xC3\xA4";
+                expected_prefix += "\\u00e4";
             }
+            const std::string emoji = "\xF0\x9F\x98\x80";
+            const json j = prefix + emoji;
 
-            // newline escaping does not depend on ensure_ascii: only the
-            // emoji differs (raw UTF-8 bytes vs. a \u-escaped surrogate pair)
+            // without ensure_ascii, the whole string is copied unchanged
             std::string expected_raw = "\"";
-            expected_raw += expected_prefix;
+            expected_raw += prefix;
             expected_raw += emoji;
             expected_raw += '"';
             std::string expected_ascii = "\"";
@@ -1027,3 +1032,524 @@ TEST_CASE("serialization boundary values for the string buffer")
         }
     }
 }
+
+namespace
+{
+// appends the escaped form of the ASCII character c to out; written
+// independently of the table in serializer::ascii_escapes()
+void append_escaped_ascii(std::string& out, const char c, const bool ensure_ascii)
+{
+    const auto byte = static_cast<unsigned char>(c);
+    switch (byte)
+    {
+        case '\b':
+            out += "\\b";
+            break;
+        case '\t':
+            out += "\\t";
+            break;
+        case '\n':
+            out += "\\n";
+            break;
+        case '\f':
+            out += "\\f";
+            break;
+        case '\r':
+            out += "\\r";
+            break;
+        case '"':
+            out += "\\\"";
+            break;
+        case '\\':
+            out += "\\\\";
+            break;
+        default:
+            if (byte < 0x20 || (ensure_ascii && byte == 0x7F))
+            {
+                // byte < 0x80, so the first two hex digits are 0
+                const char* const hex = "0123456789abcdef";
+                out += "\\u00";
+                out += hex[byte >> 4U];
+                out += hex[byte & 0x0FU];
+            }
+            else
+            {
+                out += c;
+            }
+            break;
+    }
+}
+
+// reference escaper for ASCII strings
+std::string reference_escape_ascii(const std::string& s, const bool ensure_ascii)
+{
+    std::string result = "\"";
+    for (const char c : s)
+    {
+        append_escaped_ascii(result, c, ensure_ascii);
+    }
+    result += '"';
+    return result;
+}
+} // namespace
+
+TEST_CASE("characters below 0x60 are escaped from a table straight into the write buffer")
+{
+    SECTION("every ASCII character at every position")
+    {
+        // lengths up to 17 cover the 8-byte strides of the bulk scanner on
+        // both sides of the character, so the scanner stops at it from every
+        // offset
+        for (std::size_t length = 1; length <= 17; ++length)
+        {
+            for (std::size_t pos = 0; pos < length; ++pos)
+            {
+                for (unsigned int c = 0; c < 0x80; ++c)
+                {
+                    std::string s(length, 'a');
+                    s[pos] = static_cast<char>(c);
+                    CAPTURE(length)
+                    CAPTURE(pos)
+                    CAPTURE(c)
+                    const json j = s;
+                    CHECK(j.dump(-1, ' ', false) == reference_escape_ascii(s, false));
+                    CHECK(j.dump(-1, ' ', true) == reference_escape_ascii(s, true));
+                }
+            }
+        }
+    }
+
+    SECTION("runs of escaped characters")
+    {
+        std::string s;
+        for (unsigned int c = 0; c < 0x20; ++c)
+        {
+            s += static_cast<char>(c);
+        }
+        s += R"("\"\)";
+        CHECK(json(s).dump() == reference_escape_ascii(s, false));
+        CHECK(json(s).dump(-1, ' ', true) == reference_escape_ascii(s, true));
+    }
+
+    SECTION("escapes straddle the end of the write buffer")
+    {
+        // the 8-byte store of a table entry needs room for all 8 bytes, even
+        // when the escape itself is shorter; a prefix of k plain characters
+        // shifts the 6-byte escapes across every offset of the 1024-byte
+        // write buffer
+        for (std::size_t k = 0; k < 8; ++k)
+        {
+            CAPTURE(k)
+            const std::string s = std::string(k, 'a') + std::string(400, '\x01');
+            CHECK(json(s).dump() == reference_escape_ascii(s, false));
+            CHECK(json::array({s, s}).dump() == "[" + reference_escape_ascii(s, false) + "," + reference_escape_ascii(s, false) + "]");
+        }
+    }
+
+    SECTION("pending \\u escapes of non-ASCII characters are written first")
+    {
+        // under ensure_ascii, U+00E4 is \u-escaped into the string buffer;
+        // a following escape from the table must not overtake it
+        const json j = "\xC3\xA4\n\xC3\xA4\"\xF0\x9F\x98\x80\\";
+        CHECK(j.dump(-1, ' ', true) == "\"\\u00e4\\n\\u00e4\\\"\\ud83d\\ude00\\\\\"");
+        CHECK(j.dump(-1, ' ', false) == "\"\xC3\xA4\\n\xC3\xA4\\\"\xF0\x9F\x98\x80\\\\\"");
+    }
+
+    SECTION("escapes after invalid UTF-8 for every error handler")
+    {
+        const json j = "a\xFF\n\xC3\"";
+        CHECK_THROWS_WITH_AS(utils::ignore_return_value(j.dump()), "[json.exception.type_error.316] invalid UTF-8 byte at index 1: 0xFF", json::type_error&);
+        CHECK(j.dump(-1, ' ', false, json::error_handler_t::ignore) == "\"a\\n\\\"\"");
+        CHECK(j.dump(-1, ' ', true, json::error_handler_t::ignore) == "\"a\\n\\\"\"");
+        CHECK(j.dump(-1, ' ', false, json::error_handler_t::replace) == "\"a\xEF\xBF\xBD\\n\xEF\xBF\xBD\\\"\"");
+        CHECK(j.dump(-1, ' ', true, json::error_handler_t::replace) == "\"a\\ufffd\\n\\ufffd\\\"\"");
+        CHECK(j.dump(-1, ' ', false, json::error_handler_t::keep) == "\"a\xFF\\n\xC3\\\"\"");
+        CHECK(j.dump(-1, ' ', true, json::error_handler_t::keep) == "\"a\xFF\\n\xC3\\\"\"");
+    }
+}
+
+namespace
+{
+// expected result of dumping a string: the output, or (with
+// error_handler_t::strict) the message of the type_error.316 exception
+struct expected_dump
+{
+    bool throws = false;
+    std::string text{}; // NOLINT(readability-redundant-member-init)
+};
+
+std::string upper_hex_byte(const unsigned char byte)
+{
+    const char* const hex = "0123456789ABCDEF";
+    return std::string(1, hex[byte >> 4U]) + hex[byte & 0x0FU];
+}
+
+void append_u_escape(std::string& out, const std::uint32_t code_unit)
+{
+    const char* const hex = "0123456789abcdef";
+    out += "\\u";
+    for (int shift = 12; shift >= 0; shift -= 4)
+    {
+        out += hex[(code_unit >> static_cast<unsigned int>(shift)) & 0x0FU];
+    }
+}
+
+void append_replacement_character(std::string& out, const bool ensure_ascii)
+{
+    out += ensure_ascii ? "\\ufffd" : "\xEF\xBF\xBD";
+}
+
+// reference serializer for arbitrary byte strings, written from the documented
+// behavior of dump() and independently of serializer::dump_escaped(): UTF-8 is
+// decoded strictly (well-formed sequences as in Table 3-7 of the Unicode
+// Standard), and each ill-formed sequence is handled as the error handler says
+expected_dump reference_dump_string(const std::string& s, const bool ensure_ascii, const json::error_handler_t error_handler)
+{
+    expected_dump result;
+    std::string& out = result.text;
+    out = "\"";
+
+    // handle the ill-formed bytes s[begin, end), of which s[index] is the byte
+    // that could not be decoded; returns false if an exception is expected
+    const auto ill_formed = [&](const std::size_t begin, const std::size_t end, const std::size_t index)
+    {
+        switch (error_handler)
+        {
+            case json::error_handler_t::strict:
+                result.throws = true;
+                out = "[json.exception.type_error.316] invalid UTF-8 byte at index " + std::to_string(index) + ": 0x" + upper_hex_byte(static_cast<unsigned char>(s[index]));
+                return false;
+            case json::error_handler_t::ignore:
+                break;
+            case json::error_handler_t::replace:
+                append_replacement_character(out, ensure_ascii);
+                break;
+            case json::error_handler_t::keep:
+            default:
+                // kept as is, even with ensure_ascii
+                out.append(s, begin, end - begin);
+                break;
+        }
+        return true;
+    };
+
+    std::size_t i = 0;
+    while (i < s.size())
+    {
+        const auto lead = static_cast<unsigned char>(s[i]);
+        if (lead < 0x80)
+        {
+            append_escaped_ascii(out, s[i], ensure_ascii);
+            ++i;
+            continue;
+        }
+
+        // number of bytes of the sequence and the range of its second byte
+        std::size_t length = 0;
+        std::uint32_t code_point = 0;
+        unsigned char lower = 0x80;
+        unsigned char upper = 0xBF;
+        if (lead >= 0xC2 && lead <= 0xDF)
+        {
+            length = 2;
+            code_point = lead & 0x1FU;
+        }
+        else if (lead >= 0xE0 && lead <= 0xEF)
+        {
+            length = 3;
+            code_point = lead & 0x0FU;
+            lower = (lead == 0xE0) ? 0xA0 : 0x80; // no overlong encodings
+            upper = (lead == 0xED) ? 0x9F : 0xBF; // no surrogates
+        }
+        else if (lead >= 0xF0 && lead <= 0xF4)
+        {
+            length = 4;
+            code_point = lead & 0x07U;
+            lower = (lead == 0xF0) ? 0x90 : 0x80; // no overlong encodings
+            upper = (lead == 0xF4) ? 0x8F : 0xBF; // nothing above U+10FFFF
+        }
+        else
+        {
+            // a byte that cannot start a sequence is consumed
+            if (!ill_formed(i, i + 1, i))
+            {
+                return result;
+            }
+            ++i;
+            continue;
+        }
+
+        std::size_t j = i + 1;
+        for (; j < i + length; ++j)
+        {
+            if (j == s.size())
+            {
+                // the string ends in the middle of the sequence
+                switch (error_handler)
+                {
+                    case json::error_handler_t::strict:
+                        result.throws = true;
+                        out = "[json.exception.type_error.316] incomplete UTF-8 string; last byte: 0x" + upper_hex_byte(static_cast<unsigned char>(s.back()));
+                        return result;
+                    case json::error_handler_t::ignore:
+                        break;
+                    case json::error_handler_t::replace:
+                        append_replacement_character(out, ensure_ascii);
+                        break;
+                    case json::error_handler_t::keep:
+                    default:
+                        out.append(s, i, s.size() - i);
+                        break;
+                }
+                out += '"';
+                return result;
+            }
+
+            const auto byte = static_cast<unsigned char>(s[j]);
+            if (byte < lower || byte > upper)
+            {
+                break;
+            }
+            code_point = (code_point << 6U) | (byte & 0x3FU);
+            lower = 0x80;
+            upper = 0xBF;
+        }
+
+        if (j < i + length)
+        {
+            // the sequence was abandoned at s[j]; that byte is read again as
+            // the start of a new character
+            if (!ill_formed(i, j, j))
+            {
+                return result;
+            }
+            i = j;
+            continue;
+        }
+
+        if (!ensure_ascii)
+        {
+            out.append(s, i, length);
+        }
+        else if (code_point <= 0xFFFF)
+        {
+            append_u_escape(out, code_point);
+        }
+        else
+        {
+            append_u_escape(out, 0xD800U + ((code_point - 0x10000U) >> 10U));
+            append_u_escape(out, 0xDC00U + ((code_point - 0x10000U) & 0x3FFU));
+        }
+        i = j;
+    }
+
+    out += '"';
+    return result;
+}
+
+std::string hex_dump(const std::string& s)
+{
+    std::string result;
+    for (const char c : s)
+    {
+        result += upper_hex_byte(static_cast<unsigned char>(c));
+        result += ' ';
+    }
+    return result;
+}
+
+// compares dump() of s with the reference for every error handler, with
+// ensure_ascii on and off; CHECKs only on a mismatch to keep the sweep fast;
+// error_handler_t::strict can be skipped where the position of the string
+// buffer does not matter, as throwing is slow on some platforms
+void check_dump_against_reference(const std::string& s, const bool with_strict = true)
+{
+    const json j = s;
+    const std::array<json::error_handler_t, 4> error_handlers =
+    {
+        {
+            json::error_handler_t::strict, json::error_handler_t::ignore,
+            json::error_handler_t::replace, json::error_handler_t::keep
+        }
+    };
+    for (const auto error_handler : error_handlers)
+    {
+        if (!with_strict && error_handler == json::error_handler_t::strict)
+        {
+            continue;
+        }
+        for (const bool ensure_ascii :
+                {
+                    false, true
+                })
+        {
+            const expected_dump expected = reference_dump_string(s, ensure_ascii, error_handler);
+            expected_dump actual;
+            int id = 0;
+            try
+            {
+                actual.text = j.dump(-1, ' ', ensure_ascii, error_handler);
+            }
+            catch (const json::exception& e)
+            {
+                actual.throws = true;
+                actual.text = e.what();
+                id = e.id;
+            }
+            if (actual.throws != expected.throws || actual.text != expected.text || (actual.throws && id != 316))
+            {
+                const std::string input = hex_dump(s);
+                CAPTURE(input)
+                CAPTURE(ensure_ascii)
+                CAPTURE(static_cast<int>(error_handler))
+                CHECK(actual.throws == expected.throws);
+                CHECK(actual.text == expected.text);
+                CHECK((!actual.throws || id == 316));
+            }
+        }
+    }
+}
+} // namespace
+
+TEST_CASE("invalid UTF-8 at every position is handled like the reference")
+{
+    // ill-formed byte sequences; those that end in a truncated sequence are
+    // also cut short by the filler that follows, or by the end of the string
+    const std::vector<std::string> patterns =
+    {
+        // stray continuation bytes
+        "\x80", "\xBF",
+        // bytes that cannot start a sequence
+        "\xC0", "\xC1", "\xF5", "\xF6", "\xF7", "\xF8", "\xF9", "\xFA", "\xFB", "\xFC", "\xFD", "\xFE", "\xFF",
+        // overlong encodings
+        "\xC0\xAF", "\xE0\x80\x80", "\xF0\x80\x80\x80",
+        // UTF-16 surrogates
+        "\xED\xA0\x80", "\xED\xBF\xBF",
+        // a code point above U+10FFFF
+        "\xF4\x90\x80\x80",
+        // sequences cut short by ASCII, a control character, a quote, a
+        // backslash, DEL, or the lead byte of a well-formed character
+        "\xC3" "a", "\xE2\x82" "\n", "\xE2\x82\"", "\xF0\x9F\x98\\", "\xC3\x7F", "\xC3\xC3\xA4", "\xE2\xF0\x9F\x98\x80",
+        // truncated sequences
+        "\xC3", "\xE2\x82", "\xF0\x9F\x98"
+    };
+
+    // well-formed characters of 1, 2, 3, and 4 bytes
+    const std::vector<std::string> fillers = {"a", "\xC3\xA4", "\xE2\x82\xAC", "\xF0\x9F\x98\x80"};
+
+    const auto make_string = [](const std::string & filler, const std::size_t before, const std::string & pattern, const std::size_t after)
+    {
+        std::string s;
+        for (std::size_t k = 0; k < before; ++k)
+        {
+            s += filler;
+        }
+        s += pattern;
+        for (std::size_t k = 0; k < after; ++k)
+        {
+            s += filler;
+        }
+        return s;
+    };
+
+    SECTION("the reference")
+    {
+        // the reference itself, against hand-written results
+        const std::string s = "a\xFF\n\xC3\"\xE0\x80\xF0\x9F\x98\x80\xE2\x82";
+        CHECK(reference_dump_string(s, false, json::error_handler_t::strict).text == "[json.exception.type_error.316] invalid UTF-8 byte at index 1: 0xFF");
+        CHECK(reference_dump_string(s.substr(2), false, json::error_handler_t::strict).text == "[json.exception.type_error.316] invalid UTF-8 byte at index 2: 0x22");
+        CHECK(reference_dump_string("\xF0\x9F\x98\x80\xE2\x82", true, json::error_handler_t::strict).text == "[json.exception.type_error.316] incomplete UTF-8 string; last byte: 0x82");
+        CHECK(reference_dump_string(s, false, json::error_handler_t::ignore).text == "\"a\\n\\\"\xF0\x9F\x98\x80\"");
+        CHECK(reference_dump_string(s, true, json::error_handler_t::ignore).text == "\"a\\n\\\"\\ud83d\\ude00\"");
+        CHECK(reference_dump_string(s, false, json::error_handler_t::replace).text == "\"a\xEF\xBF\xBD\\n\xEF\xBF\xBD\\\"\xEF\xBF\xBD\xEF\xBF\xBD\xF0\x9F\x98\x80\xEF\xBF\xBD\"");
+        CHECK(reference_dump_string(s, true, json::error_handler_t::replace).text == "\"a\\ufffd\\n\\ufffd\\\"\\ufffd\\ufffd\\ud83d\\ude00\\ufffd\"");
+        CHECK(reference_dump_string(s, false, json::error_handler_t::keep).text == "\"" + std::string("a\xFF\\n\xC3\\\"\xE0\x80\xF0\x9F\x98\x80\xE2\x82") + "\"");
+        CHECK(reference_dump_string(s, true, json::error_handler_t::keep).text == "\"" + std::string("a\xFF\\n\xC3\\\"\xE0\x80\\ud83d\\ude00\xE2\x82") + "\"");
+        check_dump_against_reference(s);
+    }
+
+    SECTION("short strings")
+    {
+        // strings of up to 17 bytes put the pattern at every offset of an
+        // 8-byte stride of the bulk scanners, with whole strides on both
+        // sides; lengths around 24, 32, and 40 bytes cross the 16- and 32-byte
+        // strides; error_handler_t::strict is checked for up to 17 bytes of
+        // ASCII and 8 bytes of other characters only, as throwing is slow on
+        // some platforms
+        for (const auto& filler : fillers)
+        {
+            for (std::size_t length = 0; length * filler.size() <= 40; ++length)
+            {
+                const std::size_t bytes = length * filler.size();
+                if (filler.size() == 1 && bytes > 17 && (bytes + 1) % 8 > 2)
+                {
+                    continue;
+                }
+                for (std::size_t pos = 0; pos <= length; ++pos)
+                {
+                    for (const auto& pattern : patterns)
+                    {
+                        check_dump_against_reference(make_string(filler, pos, pattern, length - pos), bytes <= (filler.size() == 1 ? 17 : 8));
+                    }
+                }
+            }
+        }
+    }
+
+    SECTION("strings crossing the end of the string buffer")
+    {
+        // characters that are not copied in bulk go through a 512-byte string
+        // buffer that is written out once fewer than 13 bytes are left; with
+        // ensure_ascii, each 2-byte character takes 6 bytes of it and each
+        // 4-byte character 12, so the pattern meets the end of the buffer after
+        // about 84 or 42 characters (3-byte characters behave like 2-byte
+        // ones); plain ASCII, copied in bulk, is checked around 512 bytes;
+        // positions further away are sampled; error_handler_t::strict throws
+        // at the first ill-formed byte, regardless of the string buffer, and is
+        // covered above
+        for (const auto& filler : fillers)
+        {
+            if (filler.size() == 3)
+            {
+                continue;
+            }
+            const std::size_t escaped_size = filler.size() == 1 ? 1 : filler.size() * 3;
+            const std::size_t middle = 506 / escaped_size;
+            const std::size_t window = filler.size() == 1 ? 8 : 16;
+            for (std::size_t pos = 0; pos <= middle + window; pos += (pos + window >= middle) ? 1 : 29)
+            {
+                for (const std::size_t after :
+                        {
+                            std::size_t{0}, std::size_t{12}
+                        })
+                {
+                    for (const auto& pattern : patterns)
+                    {
+                        check_dump_against_reference(make_string(filler, pos, pattern, after), false);
+                    }
+                }
+            }
+        }
+    }
+
+    SECTION("repeated invalid sequences at every offset of the string buffer")
+    {
+        // ill-formed sequences that are kept or replaced fill the string
+        // buffer without a bulk copy in between; a prefix of 0 to 12 bytes
+        // 0xFF (kept as 1 byte, replaced by 3 or 6) shifts them across every
+        // offset of the end of the buffer
+        for (const auto& pattern : patterns)
+        {
+            for (std::size_t shift = 0; shift <= 12; ++shift)
+            {
+                std::string s(shift, '\xFF');
+                while (s.size() < 600)
+                {
+                    s += pattern;
+                }
+                check_dump_against_reference(s, shift == 0);
+                check_dump_against_reference(s + "\xF0\x9F\x98\x80", false);
+            }
+        }
+    }
+}

tests/src/unit-throw_messages.cpp

@@ -0,0 +1,135 @@
+//     __ _____ _____ _____
+//  __|  |   __|     |   | |  JSON for Modern C++ (supporting code)
+// |  |  |__   |  |  | | | |  version 3.12.0
+// |_____|_____|_____|_|___|  https://github.com/nlohmann/json
+//
+// SPDX-FileCopyrightText: 2013-2026 Niels Lohmann <https://nlohmann.me>
+// SPDX-License-Identifier: MIT
+
+#include "doctest_compatibility.h"
+
+#include <nlohmann/json.hpp>
+using nlohmann::json;
+
+#include <string>
+
+// The exceptions in hot core functions are constructed out of line. These tests pin the
+// exact what() texts of the sites that moved, so the refactoring cannot change them.
+
+#define CHECK_MSG(expr, msg) CHECK_THROWS_WITH(expr, msg)
+
+TEST_CASE("exception messages of out-of-line throw sites")
+{
+    json arr = {1, 2, 3};
+    json arr2 = {4, 5};
+    json obj = {{"a", 1}, {"b", 2}};
+    json obj2 = {{"c", 3}};
+    json num = 1;
+
+    SECTION("iterators")
+    {
+        CHECK_MSG((void)(arr.begin() == arr2.begin()), "[json.exception.invalid_iterator.212] cannot compare iterators of different containers");
+        CHECK_MSG((void)(arr.begin() < arr2.begin()), "[json.exception.invalid_iterator.212] cannot compare iterators of different containers");
+        CHECK_MSG((void)(obj.begin() < obj.end()), "[json.exception.invalid_iterator.213] cannot compare order of object iterators");
+        CHECK_MSG(obj.begin() += 1, "[json.exception.invalid_iterator.209] cannot use offsets with object iterators");
+        CHECK_MSG((void)(obj.begin() - 1), "[json.exception.invalid_iterator.209] cannot use offsets with object iterators");
+        CHECK_MSG((void)obj.begin()[0], "[json.exception.invalid_iterator.208] cannot use operator[] for object iterators");
+        CHECK_MSG((void)arr.begin().key(), "[json.exception.invalid_iterator.207] cannot use key() for non-object iterators");
+        json null;
+        CHECK_MSG((void)*null.end(), "[json.exception.invalid_iterator.214] cannot get value");
+    }
+
+    SECTION("constructors")
+    {
+        CHECK_MSG(json(arr.begin(), arr2.end()), "[json.exception.invalid_iterator.201] iterators are not compatible");
+        CHECK_MSG(json(num.begin() + 1, num.end() + 1), "[json.exception.invalid_iterator.204] iterators out of range");
+        CHECK_MSG(json(json::initializer_list_t{1, 2}, false, json::value_t::object), "[json.exception.type_error.301] cannot create object from initializer list");
+    }
+
+    SECTION("get_ref and operator[]")
+    {
+        CHECK_MSG((void)num.get_ref<std::string&>(), "[json.exception.type_error.303] incompatible ReferenceType for get_ref, actual type is number");
+        CHECK_MSG((void)num[0], "[json.exception.type_error.305] cannot use operator[] with a numeric argument with number");
+    }
+
+    SECTION("at")
+    {
+        const json carr = arr;
+        const json cobj = obj;
+        CHECK_MSG((void)arr.at(static_cast<json::size_type>(10)), "[json.exception.out_of_range.401] array index 10 is out of range");
+        CHECK_MSG((void)carr.at(static_cast<json::size_type>(3)), "[json.exception.out_of_range.401] array index 3 is out of range");
+        CHECK_MSG((void)num.at(static_cast<json::size_type>(0)), "[json.exception.type_error.304] cannot use at() with number");
+        CHECK_MSG((void)obj.at("z"), "[json.exception.out_of_range.403] key 'z' not found");
+        CHECK_MSG((void)cobj.at("z"), "[json.exception.out_of_range.403] key 'z' not found");
+        CHECK_MSG((void)obj.at(std::string("zz")), "[json.exception.out_of_range.403] key 'zz' not found");
+        CHECK_MSG((void)num.at("a"), "[json.exception.type_error.304] cannot use at() with number");
+
+        nlohmann::ordered_json oobj = {{"a", 1}};
+        const nlohmann::ordered_json coobj = oobj;
+        CHECK_MSG((void)oobj.at("z"), "[json.exception.out_of_range.403] key 'z' not found");
+        CHECK_MSG((void)coobj.at("z"), "[json.exception.out_of_range.403] key 'z' not found");
+        CHECK_MSG((void)coobj.at(std::string("zz")), "[json.exception.out_of_range.403] key 'zz' not found");
+    }
+
+    SECTION("erase")
+    {
+        CHECK_MSG(arr.erase(arr2.begin()), "[json.exception.invalid_iterator.202] iterator does not fit current value");
+        CHECK_MSG(num.erase(num.end()), "[json.exception.invalid_iterator.205] iterator out of range");
+        CHECK_MSG(arr.erase(arr.begin(), arr2.end()), "[json.exception.invalid_iterator.203] iterators do not fit current value");
+        CHECK_MSG(num.erase(num.begin() + 1, num.end()), "[json.exception.invalid_iterator.204] iterators out of range");
+        CHECK_MSG(arr.erase(static_cast<json::size_type>(10)), "[json.exception.out_of_range.401] array index 10 is out of range");
+    }
+
+    SECTION("insert")
+    {
+        CHECK_MSG(arr.insert(arr2.begin(), 1), "[json.exception.invalid_iterator.202] iterator does not fit current value");
+        CHECK_MSG(arr.insert(arr2.begin(), 2, 1), "[json.exception.invalid_iterator.202] iterator does not fit current value");
+        CHECK_MSG(arr.insert(arr2.begin(), arr2.begin(), arr2.end()), "[json.exception.invalid_iterator.202] iterator does not fit current value");
+        CHECK_MSG(arr.insert(arr.begin(), arr.begin(), arr2.end()), "[json.exception.invalid_iterator.210] iterators do not fit");
+        CHECK_MSG(arr.insert(arr.begin(), arr.begin(), arr.end()), "[json.exception.invalid_iterator.211] passed iterators may not belong to container");
+        CHECK_MSG(arr.insert(arr.begin(), obj.begin(), obj.end()), "[json.exception.invalid_iterator.202] iterators first and last must point to arrays");
+        CHECK_MSG(arr.insert(arr2.begin(), {1, 2}), "[json.exception.invalid_iterator.202] iterator does not fit current value");
+        CHECK_MSG(obj.insert(obj2.begin(), arr.end()), "[json.exception.invalid_iterator.210] iterators do not fit");
+        CHECK_MSG(obj.insert(arr.begin(), arr.end()), "[json.exception.invalid_iterator.202] iterators first and last must point to objects");
+        CHECK_MSG(obj.update(arr.begin(), arr2.end()), "[json.exception.invalid_iterator.210] iterators do not fit");
+    }
+
+    SECTION("patch")
+    {
+        const auto patch_msg = [](const char* doc, const char* patch)
+        {
+            try
+            {
+                json d = json::parse(doc);
+                d.patch(json::parse(patch));
+            }
+            catch (const json::exception& e)
+            {
+                // with JSON_DIAGNOSTICS, what() names the offending value like "(/0) "; drop it
+                std::string msg = e.what();
+                const auto pos = msg.find(" (/");
+                if (pos != std::string::npos)
+                {
+                    msg.erase(pos + 1, msg.find(") ", pos) - pos + 1);
+                }
+                return msg;
+            }
+            return std::string();
+        };
+
+        CHECK(patch_msg("{}", "{}") == "[json.exception.parse_error.104] parse error: JSON patch must be an array of objects");
+        CHECK(patch_msg("{}", "[1]") == "[json.exception.parse_error.104] parse error: JSON patch must be an array of objects");
+        CHECK(patch_msg("{}", "[{}]") == "[json.exception.parse_error.105] parse error: operation must have member 'op'");
+        CHECK(patch_msg("{}", "[{\"op\":1}]") == "[json.exception.parse_error.105] parse error: operation must have string member 'op'");
+        CHECK(patch_msg("{}", "[{\"op\":\"add\"}]") == "[json.exception.parse_error.105] parse error: operation 'add' must have member 'path'");
+        CHECK(patch_msg("{}", "[{\"op\":\"add\",\"path\":1}]") == "[json.exception.parse_error.105] parse error: operation 'add' must have string member 'path'");
+        CHECK(patch_msg("{}", "[{\"op\":\"foo\",\"path\":\"\"}]") == "[json.exception.parse_error.105] parse error: operation value 'foo' is invalid");
+        CHECK(patch_msg("{\"a\":1}", "[{\"op\":\"remove\",\"path\":\"/b\"}]") == "[json.exception.out_of_range.403] key 'b' not found");
+        CHECK(patch_msg("[1]", "[{\"op\":\"add\",\"path\":\"/5\",\"value\":0}]") == "[json.exception.out_of_range.401] array index 5 is out of range");
+        CHECK(patch_msg("{\"a\":1}", "[{\"op\":\"add\",\"path\":\"/a/b\",\"value\":0}]") == "[json.exception.out_of_range.411] cannot add value: the JSON Patch 'add' target's parent is of type number, but must be an object or array");
+        CHECK(patch_msg("{\"a\":1}", "[{\"op\":\"remove\",\"path\":\"/a/b\"}]") == "[json.exception.out_of_range.413] cannot remove value: the JSON Patch 'remove' target's parent is of type number, but must be an object or array");
+        CHECK(patch_msg("{\"a\":{}}", "[{\"op\":\"move\",\"from\":\"/a\",\"path\":\"/a/b\"}]") == "[json.exception.out_of_range.414] cannot move value: 'from' path '/a' is a proper prefix of 'path' '/a/b'");
+        CHECK(patch_msg("{\"a\":1}", "[{\"op\":\"test\",\"path\":\"/a\",\"value\":2}]") == "[json.exception.other_error.501] unsuccessful: {\"op\":\"test\",\"path\":\"/a\",\"value\":2}");
+        CHECK(patch_msg("{\"a\":1}", "[{\"op\":\"move\",\"path\":\"/a\"}]") == "[json.exception.parse_error.105] parse error: operation 'move' must have member 'from'");
+    }
+}

tests/src/unit-to_chars.cpp

@@ -21,11 +21,8 @@ using nlohmann::detail::dtoa_impl::reinterpret_bits;
 #include <cstdint>
 #include <cstdio>
 #include <cstdlib>
-#include <iomanip>
 #include <limits>
-#include <locale>
 #include <random>
-#include <sstream>
 #include <string>
 #include <utility>
 #include <vector>
@@ -689,20 +686,13 @@ void check_shortest(double v)
     const std::string& digits = de.first;
     if (digits.size() > 1)
     {
-        // the decimals of one digit fewer next to the value
-        // (a stream rather than snprintf("%.*e"), whose output GCC cannot bound)
-        std::ostringstream shorter;
-        shorter.imbue(std::locale::classic());
-        shorter << std::scientific << std::setprecision(static_cast<int>(digits.size()) - 2) << v;
-        const auto near = digits_and_exponent(shorter.str());
-        // as an integer with digits.size() - 1 digits
-        std::string m = near.first;
-        int e = near.second;
-        while (m.size() < digits.size() - 1)
-        {
-            m += '0';
-            --e;
-        }
+        // the decimals of one digit fewer next to the value: v lies within
+        // half a unit of the last digit of text, so the closest ones are the
+        // truncated digits and their successor
+        // (not a stream with std::setprecision(digits.size() - 2): the
+        // standard library of MSVC 2017 ignores a precision of 0)
+        const std::string m = digits.substr(0, digits.size() - 1);
+        const int e = de.second + 1;
         const std::uint64_t mid = std::stoull(m);
         for (const std::uint64_t candidate :
                 {

tests/src/unit-to_json-rvalue.cpp

@@ -0,0 +1,257 @@
+//     __ _____ _____ _____
+//  __|  |   __|     |   | |  JSON for Modern C++ (supporting code)
+// |  |  |__   |  |  | | | |  version 3.12.0
+// |_____|_____|_____|_|___|  https://github.com/nlohmann/json
+//
+// SPDX-FileCopyrightText: 2013-2026 Niels Lohmann <https://nlohmann.me>
+// SPDX-License-Identifier: MIT
+
+#include "doctest_compatibility.h"
+
+#include <nlohmann/json.hpp>
+using nlohmann::json;
+
+#include <array>
+#include <deque>
+#include <forward_list>
+#include <list>
+#include <map>
+#include <set>
+#include <string>
+#include <unordered_map>
+#include <utility>
+#include <valarray>
+#include <vector>
+
+#if (defined(__cplusplus) && __cplusplus >= 201703L) || (defined(_HAS_CXX17) && _HAS_CXX17 == 1)
+    #if __has_include(<filesystem>)
+        #include <filesystem>
+        #define TO_JSON_RVALUE_HAS_FS 1
+    #endif
+#endif
+
+namespace
+{
+// element type that counts how it is converted to JSON: a conversion from an
+// lvalue copies, a conversion from an rvalue moves
+struct tracked
+{
+    std::string payload;
+    bool moved_from = false;
+
+    explicit tracked(std::string p) : payload(std::move(p)) {}
+    tracked(const tracked&) = default;
+    tracked& operator=(const tracked&) = default;
+    tracked(tracked&& other) noexcept : payload(std::move(other.payload)), moved_from(other.moved_from)
+    {
+        other.moved_from = true;
+    }
+    tracked& operator=(tracked&& other) noexcept
+    {
+        payload = std::move(other.payload);
+        moved_from = other.moved_from;
+        other.moved_from = true;
+        return *this;
+    }
+    ~tracked() = default;
+};
+
+int tracked_copies = 0; // NOLINT(cppcoreguidelines-avoid-non-const-global-variables,misc-use-internal-linkage)
+int tracked_moves = 0;  // NOLINT(cppcoreguidelines-avoid-non-const-global-variables,misc-use-internal-linkage)
+
+void to_json(json& j, const tracked& t)
+{
+    ++tracked_copies;
+    j = t.payload;
+}
+
+void to_json(json& j, tracked&& t)
+{
+    ++tracked_moves;
+    j = std::move(t.payload);
+    t.moved_from = true;
+}
+
+// user type with its own rvalue-aware serializer
+struct udt_container
+{
+    std::vector<int> values;
+};
+} // namespace
+
+namespace nlohmann
+{
+template<>
+struct adl_serializer<udt_container>
+{
+    static void to_json(json& j, const udt_container& c)
+    {
+        j = c.values;
+        j.push_back("copied");
+    }
+    static void to_json(json& j, udt_container&& c)
+    {
+        j = std::move(c.values);
+        j.push_back("moved");
+    }
+};
+} // namespace nlohmann
+
+TEST_CASE("to_json from rvalue containers")
+{
+    const std::string long_a(64, 'a');
+    const std::string long_b(64, 'b');
+
+    SECTION("vector<std::string>: elements are moved")
+    {
+        std::vector<std::string> v{long_a, long_b};
+        const json j = std::move(v);
+        CHECK(j == json::array({long_a, long_b}));
+        for (const auto& s : v) // NOLINT(bugprone-use-after-move,hicpp-invalid-access-moved)
+        {
+            CHECK(s.empty()); // long strings leave the source element empty when moved
+        }
+    }
+
+    SECTION("vector<std::string>: lvalues are copied and left untouched")
+    {
+        const std::vector<std::string> v{long_a, long_b};
+        const json j = v;
+        CHECK(j == json::array({long_a, long_b}));
+        CHECK(v[0] == long_a);
+        CHECK(v[1] == long_b);
+
+        std::vector<std::string> v2{long_a, long_b};
+        const json j2 = v2;
+        CHECK(j2 == j);
+        CHECK(v2[0] == long_a);
+        CHECK(v2[1] == long_b);
+    }
+
+    SECTION("tracked elements: rvalue moves, lvalue copies")
+    {
+        tracked_copies = 0;
+        tracked_moves = 0;
+
+        std::vector<tracked> v;
+        v.emplace_back("x");
+        v.emplace_back("y");
+
+        const json copy = v;
+        CHECK(copy == json::array({"x", "y"}));
+        CHECK(tracked_copies == 2);
+        CHECK(tracked_moves == 0);
+        CHECK_FALSE(v[0].moved_from);
+
+        const json moved = std::move(v);
+        CHECK(moved == copy);
+        CHECK(tracked_copies == 2);
+        CHECK(tracked_moves == 2);
+    }
+
+    SECTION("const rvalue is copied")
+    {
+        tracked_copies = 0;
+        tracked_moves = 0;
+        const std::vector<tracked> v(2, tracked("z"));
+        const json j = std::move(v); // NOLINT(performance-move-const-arg,hicpp-move-const-arg)
+        CHECK(j == json::array({"z", "z"}));
+        CHECK(tracked_copies == 2);
+        CHECK(tracked_moves == 0);
+    }
+
+    SECTION("other sequence containers")
+    {
+        CHECK(json(std::deque<std::string> {long_a, long_b}) == json::array({long_a, long_b}));
+        CHECK(json(std::list<std::string> {long_a, long_b}) == json::array({long_a, long_b}));
+        CHECK(json(std::forward_list<std::string> {long_a, long_b}) == json::array({long_a, long_b}));
+        CHECK(json(std::array<std::string, 2> {{long_a, long_b}}) == json::array({long_a, long_b}));
+        CHECK(json(std::set<std::string> {long_a, long_b}) == json::array({long_a, long_b}));
+        CHECK(json(std::vector<bool> {true, false}) == json::array({true, false}));
+        CHECK(json(std::vector<int> {1, 2, 3}) == json::array({1, 2, 3}));
+        CHECK(json(std::vector<std::vector<std::string>> {{long_a}, {long_b}}) == json::array({json::array({long_a}), json::array({long_b})}));
+        CHECK(json(std::vector<json> {1, "a"}) == json::array({1, "a"}));
+    }
+
+    SECTION("valarray")
+    {
+        std::valarray<int> va(3);
+        va[0] = 1;
+        va[1] = 2;
+        va[2] = 3;
+        CHECK(json(std::move(va)) == json::array({1, 2, 3})); // NOLINT(performance-move-const-arg,hicpp-move-const-arg)
+        CHECK(json(std::valarray<int>(2)) == json::array({0, 0}));
+    }
+
+    SECTION("array_t and json arrays")
+    {
+        json::array_t arr{1, 2};
+        CHECK(json(std::move(arr)) == json::array({1, 2}));
+        json a = json::array({1, 2});
+        const json b = std::move(a);
+        CHECK(b == json::array({1, 2}));
+    }
+
+    SECTION("object-like containers")
+    {
+        std::map<std::string, std::string> m{{"a", long_a}, {"b", long_b}};
+        const json j = std::move(m);
+        CHECK(j == json({{"a", long_a}, {"b", long_b}}));
+        for (const auto& kv : m) // NOLINT(bugprone-use-after-move,hicpp-invalid-access-moved)
+        {
+            CHECK(kv.second.empty()); // values moved, keys (const) copied
+        }
+
+        const std::map<std::string, std::string> cm{{"a", long_a}};
+        const json jc = cm;
+        CHECK(jc["a"] == long_a);
+        CHECK(cm.at("a") == long_a);
+
+        std::unordered_map<std::string, std::string> um{{"a", long_a}};
+        CHECK(json(std::move(um)) == json({{"a", long_a}}));
+
+        std::multimap<std::string, int> mm{{"a", 1}};
+        CHECK(json(std::move(mm)) == json({{"a", 1}}));
+
+        tracked_copies = 0;
+        tracked_moves = 0;
+        std::map<std::string, tracked> mt;
+        mt.emplace("k", tracked("v"));
+        const json jt = std::move(mt);
+        CHECK(jt == json({{"k", "v"}}));
+        CHECK(tracked_moves == 1);
+        CHECK(tracked_copies == 0);
+
+        json::object_t obj{{"a", 1}};
+        CHECK(json(std::move(obj)) == json({{"a", 1}}));
+    }
+
+    SECTION("user type with its own to_json overloads")
+    {
+        udt_container c{{1, 2}};
+        const json lv = c;
+        CHECK(lv == json::array({1, 2, "copied"}));
+        const json rv = std::move(c);
+        CHECK(rv == json::array({1, 2, "moved"}));
+    }
+
+    SECTION("assignment and implicit conversion")
+    {
+        std::vector<std::string> v{long_a};
+        json j;
+        j = std::move(v);
+        CHECK(j == json::array({long_a}));
+        json k = {std::vector<std::string>{long_b}};
+        CHECK(k.size() == 1);
+    }
+
+#ifdef TO_JSON_RVALUE_HAS_FS
+    SECTION("std::filesystem::path rvalue is still a string")
+    {
+        CHECK(json(std::filesystem::path("a/b")) == json("a/b"));
+#ifdef __cpp_lib_char8_t
+        CHECK(json(std::u8string(u8"a/b")) == json("a/b"));
+#endif
+    }
+#endif
+}