Fix to_bjdata() silently truncating out-of-range _ArrayData_ elements

write_bjdata_ndarray() validated that each _ArrayData_ element matched
the number kind (integer vs. float) named by _ArrayType_, but not its
range. An element that did not fit the target C++ type (e.g. 256 for
"uint8") was silently wrapped by the static_cast used to write it, or,
for "single", silently overflowed to infinity.

Range-check each element against the type named by _ArrayType_ before
writing it, reusing the existing fallback path that already encodes
the annotated object as a plain object for other invalid-annotation
cases in this function.

Fixes #5403.

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

include/nlohmann/detail/output/binary_writer.hpp

@@ -1647,6 +1647,20 @@ class binary_writer
         return 'D';  // float 64
     }
 
+    /*!
+    @brief checks whether a JSON number fits into @a TargetType
+    @param[in] el a JSON number of either the signed or unsigned integer kind
+    @return whether @a el's value can be represented by @a TargetType without
+            wrapping, regardless of which of the two kinds it is stored as
+    */
+    template<typename TargetType>
+    static bool bjdata_ndarray_value_in_range(const BasicJsonType& el)
+    {
+        return el.is_number_unsigned()
+               ? value_in_range_of<TargetType>(el.template get<std::uint64_t>())
+               : value_in_range_of<TargetType>(el.template get<std::int64_t>());
+    }
+
     /*!
     @return false if the object is successfully converted to a bjdata ndarray, true if the type or size is invalid
     */
@@ -1731,6 +1745,60 @@ class binary_writer
             }
         }
 
+        // every element is cast to the (possibly narrower) C++ type matching
+        // dtype below; a value that does not fit that type would silently
+        // wrap (integers) or overflow to infinity (the "single" precision
+        // float) instead of being reported, so such an object falls back to
+        // a plain object encoding as well
+        for (const auto& el : value.at(key))
+        {
+            bool in_range = true;
+            switch (dtype)
+            {
+                case 'U':
+                case 'C':
+                case 'B':
+                    in_range = bjdata_ndarray_value_in_range<std::uint8_t>(el);
+                    break;
+                case 'i':
+                    in_range = bjdata_ndarray_value_in_range<std::int8_t>(el);
+                    break;
+                case 'u':
+                    in_range = bjdata_ndarray_value_in_range<std::uint16_t>(el);
+                    break;
+                case 'I':
+                    in_range = bjdata_ndarray_value_in_range<std::int16_t>(el);
+                    break;
+                case 'm':
+                    in_range = bjdata_ndarray_value_in_range<std::uint32_t>(el);
+                    break;
+                case 'l':
+                    in_range = bjdata_ndarray_value_in_range<std::int32_t>(el);
+                    break;
+                case 'M':
+                    in_range = bjdata_ndarray_value_in_range<std::uint64_t>(el);
+                    break;
+                case 'L':
+                    in_range = bjdata_ndarray_value_in_range<std::int64_t>(el);
+                    break;
+                case 'd':
+                {
+                    const auto dval = el.template get<double>();
+                    in_range = !std::isfinite(dval) ||
+                               (dval >= static_cast<double>(std::numeric_limits<float>::lowest()) &&
+                                dval <= static_cast<double>((std::numeric_limits<float>::max)()));
+                    break;
+                }
+                default:
+                    // 'D' (double) already spans the full range of number_float_t
+                    break;
+            }
+            if (!in_range)
+            {
+                return true;
+            }
+        }
+
         oa->write_character('[');
         oa->write_character('$');
         oa->write_character(dtype);

single_include/nlohmann/json.hpp

@@ -18655,6 +18655,20 @@ class binary_writer
         return 'D';  // float 64
     }
 
+    /*!
+    @brief checks whether a JSON number fits into @a TargetType
+    @param[in] el a JSON number of either the signed or unsigned integer kind
+    @return whether @a el's value can be represented by @a TargetType without
+            wrapping, regardless of which of the two kinds it is stored as
+    */
+    template<typename TargetType>
+    static bool bjdata_ndarray_value_in_range(const BasicJsonType& el)
+    {
+        return el.is_number_unsigned()
+               ? value_in_range_of<TargetType>(el.template get<std::uint64_t>())
+               : value_in_range_of<TargetType>(el.template get<std::int64_t>());
+    }
+
     /*!
     @return false if the object is successfully converted to a bjdata ndarray, true if the type or size is invalid
     */
@@ -18739,6 +18753,60 @@ class binary_writer
             }
         }
 
+        // every element is cast to the (possibly narrower) C++ type matching
+        // dtype below; a value that does not fit that type would silently
+        // wrap (integers) or overflow to infinity (the "single" precision
+        // float) instead of being reported, so such an object falls back to
+        // a plain object encoding as well
+        for (const auto& el : value.at(key))
+        {
+            bool in_range = true;
+            switch (dtype)
+            {
+                case 'U':
+                case 'C':
+                case 'B':
+                    in_range = bjdata_ndarray_value_in_range<std::uint8_t>(el);
+                    break;
+                case 'i':
+                    in_range = bjdata_ndarray_value_in_range<std::int8_t>(el);
+                    break;
+                case 'u':
+                    in_range = bjdata_ndarray_value_in_range<std::uint16_t>(el);
+                    break;
+                case 'I':
+                    in_range = bjdata_ndarray_value_in_range<std::int16_t>(el);
+                    break;
+                case 'm':
+                    in_range = bjdata_ndarray_value_in_range<std::uint32_t>(el);
+                    break;
+                case 'l':
+                    in_range = bjdata_ndarray_value_in_range<std::int32_t>(el);
+                    break;
+                case 'M':
+                    in_range = bjdata_ndarray_value_in_range<std::uint64_t>(el);
+                    break;
+                case 'L':
+                    in_range = bjdata_ndarray_value_in_range<std::int64_t>(el);
+                    break;
+                case 'd':
+                {
+                    const auto dval = el.template get<double>();
+                    in_range = !std::isfinite(dval) ||
+                               (dval >= static_cast<double>(std::numeric_limits<float>::lowest()) &&
+                                dval <= static_cast<double>((std::numeric_limits<float>::max)()));
+                    break;
+                }
+                default:
+                    // 'D' (double) already spans the full range of number_float_t
+                    break;
+            }
+            if (!in_range)
+            {
+                return true;
+            }
+        }
+
         oa->write_character('[');
         oa->write_character('$');
         oa->write_character(dtype);

tests/src/unit-bjdata.cpp

@@ -2776,6 +2776,53 @@ TEST_CASE("BJData")
                 CHECK(out_num.at(0) == '{');
                 CHECK(json::from_bjdata(out_num) == j_num);
             }
+
+            SECTION("ndarray with out-of-range _ArrayData_ elements stays as object")
+            {
+                // each element is cast to the (possibly narrower) C++ type
+                // named by _ArrayType_ before being written; a value that
+                // does not fit that type would silently wrap instead of
+                // being reported, so such an object falls back to a plain
+                // object encoding that still round-trips (see GitHub issue #5403)
+
+                // an unsigned element that does not fit uint8
+                json const j_uint8 = json({{"_ArrayType_", "uint8"}, {"_ArraySize_", {2}}, {"_ArrayData_", {1, 256}}});
+                const auto out_uint8 = json::to_bjdata(j_uint8);
+                CHECK(out_uint8.at(0) == '{');
+                CHECK(json::from_bjdata(out_uint8) == j_uint8);
+
+                // a signed element that does not fit int8
+                json const j_int8 = json({{"_ArrayType_", "int8"}, {"_ArraySize_", {2}}, {"_ArrayData_", {1, 200}}});
+                const auto out_int8 = json::to_bjdata(j_int8);
+                CHECK(out_int8.at(0) == '{');
+                CHECK(json::from_bjdata(out_int8) == j_int8);
+
+                // a negative element is likewise out of range for an
+                // unsigned _ArrayType_
+                json const j_uint16_neg = json({{"_ArrayType_", "uint16"}, {"_ArraySize_", {2}}, {"_ArrayData_", {1, -1}}});
+                const auto out_uint16_neg = json::to_bjdata(j_uint16_neg);
+                CHECK(out_uint16_neg.at(0) == '{');
+                CHECK(json::from_bjdata(out_uint16_neg) == j_uint16_neg);
+
+                // a double element that overflows to infinity when narrowed
+                // to the "single" (float) precision named by _ArrayType_
+                json const j_single = json({{"_ArrayType_", "single"}, {"_ArraySize_", {2}}, {"_ArrayData_", {1.5, 1e40}}});
+                const auto out_single = json::to_bjdata(j_single);
+                CHECK(out_single.at(0) == '{');
+                CHECK(json::from_bjdata(out_single) == j_single);
+
+                // in-range boundary values still use the compact ndarray encoding
+                json const j_uint8_ok = json({{"_ArrayType_", "uint8"}, {"_ArraySize_", {2}}, {"_ArrayData_", {0, 255}}});
+                CHECK(json::to_bjdata(j_uint8_ok) == std::vector<uint8_t>({'[', '$', 'U', '#', '[', 'i', 2, ']', 0, 255}));
+
+                json const j_int8_ok = json({{"_ArrayType_", "int8"}, {"_ArraySize_", {2}}, {"_ArrayData_", {-128, 127}}});
+                CHECK(json::to_bjdata(j_int8_ok) == std::vector<uint8_t>({'[', '$', 'i', '#', '[', 'i', 2, ']', 0x80, 0x7F}));
+
+                json const j_single_ok = json({{"_ArrayType_", "single"}, {"_ArraySize_", {1}}, {"_ArrayData_", {1.5}}});
+                const auto out_single_ok = json::to_bjdata(j_single_ok);
+                CHECK(out_single_ok.at(0) == '[');
+                CHECK(json::from_bjdata(out_single_ok) == json({1.5f}));
+            }
         }
     }
 

tests/src/unit-bson.cpp

@@ -38,54 +38,6 @@ class huge_binary_t : public std::vector<std::uint8_t>
 using huge_binary_json = nlohmann::basic_json <
                          std::map, std::vector, std::string, bool, std::int64_t, std::uint64_t,
                          double, std::allocator, nlohmann::adl_serializer, huge_binary_t, void >;
-
-// a string type that can be made to report a size beyond INT32_MAX without
-// allocating that much memory, so BSON length overflow can be tested for
-// strings and (embedded) documents as well, following the same idea as
-// huge_binary_t.
-//
-// Unlike huge_binary_t (which is only ever used as the BSON *value* type),
-// this type doubles as basic_json's StringType and is therefore also used
-// for *object keys* (e.g. "s" or "nested" below). Only the designated test
-// value is meant to lie about its size - if every huge_string_t (including
-// keys) reported a huge size, the running totals computed while walking the
-// BSON document (see calc_bson_object_size & friends in binary_writer.hpp)
-// would need more than 32 bits, and on platforms where std::size_t is only
-// 32 bits wide that arithmetic would silently wrap around, producing wrong
-// (or even unguarded) lengths. The fake size is therefore opt-in via
-// as_huge(), and plain strings - in particular object keys - keep reporting
-// their real, small size.
-class huge_string_t : public std::string
-{
-  public:
-    using std::string::string;
-    huge_string_t(const std::string& s) : std::string(s) {} // NOLINT(google-explicit-constructor,hicpp-explicit-conversions)
-
-    // returns a copy of @a s whose size() pretends to be huge
-    static huge_string_t as_huge(const std::string& s)
-    {
-        huge_string_t result(s);
-        result.pretend_huge = true;
-        return result;
-    }
-
-    size_type size() const noexcept
-    {
-        if (pretend_huge)
-        {
-            // one byte more than the BSON length field can represent
-            return static_cast<size_type>((std::numeric_limits<std::int32_t>::max)()) + 1;
-        }
-        return std::string::size();
-    }
-
-  private:
-    bool pretend_huge = false;
-};
-
-using huge_string_json = nlohmann::basic_json <
-                         std::map, std::vector, huge_string_t, bool, std::int64_t, std::uint64_t,
-                         double, std::allocator, nlohmann::adl_serializer, std::vector<std::uint8_t>, void >;
 } // namespace
 
 TEST_CASE("BSON")
@@ -153,36 +105,10 @@ TEST_CASE("BSON")
 
     SECTION("lengths exceeding INT32_MAX cannot be serialized to BSON")
     {
-        // out_of_range.412 is thrown from a single shared helper
-        // (to_bson_length) that guards the BSON length fields of binary
-        // values, strings, and (embedded) documents alike
-        SECTION("binary")
-        {
-            huge_binary_json j;
-            j["b"] = huge_binary_json::binary(huge_binary_t{});
+        huge_binary_json j;
+        j["b"] = huge_binary_json::binary(huge_binary_t{});
 
-            CHECK_THROWS_WITH_AS(huge_binary_json::to_bson(j), "[json.exception.out_of_range.412] BSON length 2147483661 exceeds maximum of 2147483647", huge_binary_json::out_of_range&);
-        }
-
-        SECTION("string")
-        {
-            huge_string_json j;
-            j["s"] = huge_string_t::as_huge("value");
-
-            CHECK_THROWS_WITH_AS(huge_string_json::to_bson(j), "[json.exception.out_of_range.412] BSON length 2147483661 exceeds maximum of 2147483647", huge_string_json::out_of_range&);
-        }
-
-        SECTION("document")
-        {
-            // an oversized string nested one level deep makes the
-            // *embedded* document's own length exceed INT32_MAX as well
-            huge_string_json nested;
-            nested["s"] = huge_string_t::as_huge("value");
-            huge_string_json j;
-            j["nested"] = nested;
-
-            CHECK_THROWS_WITH_AS(huge_string_json::to_bson(j), "[json.exception.out_of_range.412] BSON length 2147483674 exceeds maximum of 2147483647", huge_string_json::out_of_range&);
-        }
+        CHECK_THROWS_WITH_AS(huge_binary_json::to_bson(j), "[json.exception.out_of_range.412] BSON length 2147483661 exceeds maximum of 2147483647", huge_binary_json::out_of_range&);
     }
 
     SECTION("string length must be at least 1")
@@ -267,23 +193,6 @@ TEST_CASE("BSON")
             CHECK(json::from_bson(result, true, false) == j);
         }
 
-        SECTION("non-empty object with bool from a non-0/1 byte (lenient parsing)")
-        {
-            // documented lenient behavior (see gh-5333): any non-zero byte
-            // is accepted as `true`, not just 0x01
-            std::vector<std::uint8_t> const input =
-            {
-                0x0D, 0x00, 0x00, 0x00, // size (little endian)
-                0x08,               // entry: boolean
-                'e', 'n', 't', 'r', 'y', '\x00',
-                0x02,           // value = 0x02 (neither 0x00 nor 0x01)
-                0x00                    // end marker
-            };
-
-            const json expected = { { "entry", true } };
-            CHECK(json::from_bson(input) == expected);
-        }
-
         SECTION("non-empty object with double")
         {
             json const j =
@@ -590,29 +499,6 @@ TEST_CASE("BSON")
             CHECK(json::from_bson(result, true, false) == j);
         }
 
-        SECTION("array elements with non-conforming keys (lenient parsing)")
-        {
-            // documented lenient behavior (see gh-5333): BSON array element
-            // keys are not checked against the required decimal sequence
-            // "0", "1", "2", ... - elements are taken in encoded order
-            std::vector<std::uint8_t> const input =
-            {
-                0x26, 0x00, 0x00, 0x00, // size (little endian)
-                0x04, 'e', 'n', 't', 'r', 'y', '\x00', // entry: embedded array
-
-                0x1A, 0x00, 0x00, 0x00, // size (little endian)
-                0x10, '5', 0x00, 0x0A, 0x00, 0x00, 0x00, // key "5" (bogus)      -> 10
-                0x10, 'x', 0x00, 0x14, 0x00, 0x00, 0x00, // key "x" (non-numeric) -> 20
-                0x10, '1', 0x00, 0x1E, 0x00, 0x00, 0x00, // key "1" (out of order) -> 30
-                0x00, // end marker (embedded array)
-
-                0x00 // end marker
-            };
-
-            const json expected = { { "entry", json::array({10, 20, 30}) } };
-            CHECK(json::from_bson(input) == expected);
-        }
-
         SECTION("non-empty object with binary member")
         {
             const size_t N = 10;
@@ -708,31 +594,6 @@ TEST_CASE("BSON")
             CHECK(json::from_bson(result, true, false) == j);
         }
 
-        SECTION("binary member with subtype 0x02 (old binary) keeps its inner length prefix (lenient parsing)")
-        {
-            // documented lenient behavior (see gh-5333): the payload for
-            // binary subtype 0x02 ("old binary") is returned as-is,
-            // including its own inner 4-byte length prefix; it is not
-            // stripped or reinterpreted
-            std::vector<std::uint8_t> const input =
-            {
-                0x17, 0x00, 0x00, 0x00, // size (little endian)
-                0x05, 'e', 'n', 't', 'r', 'y', '\x00', // entry: binary
-
-                0x06, 0x00, 0x00, 0x00, // size of binary (little endian)
-                0x02, // "old binary" subtype
-                0x02, 0x00, 0x00, 0x00, // inner length prefix (part of the old-binary payload)
-                0x68, 0x69, // payload ('h', 'i')
-
-                0x00 // end marker
-            };
-
-            // the inner length prefix is part of the (unmodified) payload
-            const std::vector<std::uint8_t> expected_payload = {0x02, 0x00, 0x00, 0x00, 0x68, 0x69};
-            const json expected = { { "entry", json::binary(expected_payload, 0x02) } };
-            CHECK(json::from_bson(input) == expected);
-        }
-
         SECTION("Some more complex document")
         {
             json const j =