Read ND-arrays of half floats and enforce the char limit in ND-arrays

BJData N-dimensional arrays with element type 'h' (float16) failed to
read with a misleading parse_error.112, because the reader's dtype name
table lacked 'h'. It now maps to "half", the JData name. The writer still
does not pack "half"; such an object is written as a plain object and
reads back equal.

BJData limits 'C' (char) to 0..127 in Draft 2 and Draft 3, and the 1-D
reader enforces this, but ND-arrays with _ArrayType_ "char" wrote values
up to 255 as 'C' elements and read them back without a check. The writer
now only packs values in 0..127 (others fall back to a plain object that
round-trips), and the reader rejects 'C' elements above 127 with
parse_error.113 like the 1-D reader.

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

include/nlohmann/detail/input/binary_reader.hpp

@@ -3277,6 +3277,8 @@ class binary_reader
             // get_ubjson_size_value() (the type marker is known before the dimension vector
             // that determines size_and_type.first is read, so it is emitted first there to
             // match the documented _ArrayType_, _ArraySize_, _ArrayData_ key order)
+            // 'C' elements are read as unsigned bytes, but BJData limits a char to 0..127
+            const bool nd_is_char = size_and_type.second == 'C';
             if (size_and_type.second == 'C' || size_and_type.second == 'B')
             {
                 size_and_type.second = 'U';
@@ -3290,6 +3292,24 @@ class binary_reader
 
             for (std::size_t i = 0; i < size_and_type.first; ++i)
             {
+                if (nd_is_char)
+                {
+                    get();
+                    if (JSON_HEDLEY_UNLIKELY(!unexpect_eof("char")))
+                    {
+                        return false;
+                    }
+                    if (JSON_HEDLEY_UNLIKELY(current > 127))
+                    {
+                        return last_byte_error(exception_id::invalid_string_or_size, "byte after 'C' must be in range 0x00..0x7F", "char", "");
+                    }
+                    if (JSON_HEDLEY_UNLIKELY(!sax->number_unsigned(static_cast<number_unsigned_t>(current))))
+                    {
+                        return false;
+                    }
+                    continue;
+                }
+
                 if (JSON_HEDLEY_UNLIKELY(!get_ubjson_value(size_and_type.second)))
                 {
                     return false;
@@ -4655,6 +4675,8 @@ class binary_reader
                 return "uint8";
             case 'd':
                 return "single";
+            case 'h':
+                return "half";
             case 'i':
                 return "int8";
             case 'l':

include/nlohmann/detail/output/binary_writer.hpp

@@ -3848,7 +3848,7 @@ class binary_writer
     A C++11 `constexpr` function cannot contain a `switch`, so this is a plain
     comparison chain instead; it is only reached once per ND-array candidate
     object. Keep in sync with binary_reader's `bjd_type_name()`, which maps
-    the other way.
+    the other way (it also reads "half", which is never written).
     */
     static CharType bjdata_ndarray_type_marker(const string_t& name)
     {
@@ -4056,10 +4056,13 @@ class binary_writer
             switch (dtype)
             {
                 case 'U':
-                case 'C':
                 case 'B':
                     ok = write_bjdata_ndarray_element<std::uint8_t>(el, dry_run);
                     break;
+                case 'C':
+                    // BJData limits a char to 0..127 (the intersection of the int8 and uint8 ranges)
+                    ok = bjdata_ndarray_value_in_range<std::int8_t>(el) && bjdata_ndarray_value_in_range<std::uint8_t>(el);
+                    break;
                 case 'i':
                     ok = write_bjdata_ndarray_element<std::int8_t>(el, dry_run);
                     break;

tests/src/unit-bjdata.cpp

@@ -2916,6 +2916,29 @@ TEST_CASE("BJData")
                 CHECK(json::from_bjdata(out_single_ok) == json({{"_ArrayType_", "single"}, {"_ArraySize_", {2, 1}}, {"_ArrayData_", {1.5f, -1.5f}}}));
             }
 
+            SECTION("ndarray with _ArrayType_ \"char\" is limited to 0..127")
+            {
+                // both BJData drafts limit a char to 0..127; larger values must not
+                // be written as 'C' elements (the reader rejects them), so such an
+                // object stays a plain object that round-trips
+                for (const auto version : {json::bjdata_version_t::draft2, json::bjdata_version_t::draft3})
+                {
+                    json const j_ok = json({{"_ArrayType_", "char"}, {"_ArraySize_", {2, 2}}, {"_ArrayData_", {0, 97, 98, 127}}});
+                    const auto out_ok = json::to_bjdata(j_ok, true, true, version);
+                    CHECK(out_ok == std::vector<uint8_t>({'[', '$', 'C', '#', '[', '$', 'i', '#', 'i', 2, 2, 2, 0, 97, 98, 127}));
+                    CHECK(json::from_bjdata(out_ok) == j_ok);
+
+                    for (const int bad : {128, 200, 255, 256, -1})
+                    {
+                        CAPTURE(bad)
+                        json const j_bad = json({{"_ArrayType_", "char"}, {"_ArraySize_", {2, 2}}, {"_ArrayData_", {97, 98, 99, bad}}});
+                        const auto out_bad = json::to_bjdata(j_bad, true, true, version);
+                        CHECK(out_bad.at(0) == '{');
+                        CHECK(json::from_bjdata(out_bad) == j_bad);
+                    }
+                }
+            }
+
             SECTION("ndarray annotation keys are read back in the documented order")
             {
                 // from_bjdata() must emit the annotation object's keys in the order
@@ -3324,6 +3347,44 @@ TEST_CASE("BJData")
                 }
             }
 
+            SECTION("ndarray of half-precision floats can be read")
+            {
+                // the 'h' element type is read as _ArrayType_ "half"
+                json _;
+                const std::vector<uint8_t> half = {'[', '$', 'h', '#', '[', '$', 'U', '#', 'U', 2, 2, 2, 0x00, 0x3C, 0x00, 0x40, 0x00, 0x42, 0x00, 0x44};
+                const json j_half = json::parse(R"({"_ArrayType_":"half","_ArraySize_":[2,2],"_ArrayData_":[1.0,2.0,3.0,4.0]})");
+                CHECK(json::from_bjdata(half) == j_half);
+
+                // the writer does not pack "half": it writes a plain object that reads back equal
+                CHECK(json::from_bjdata(json::to_bjdata(j_half)) == j_half);
+                CHECK(json::from_bjdata(json::to_bjdata(j_half, true, true, json::bjdata_version_t::draft3)) == j_half);
+
+                // truncated half data
+                CHECK_THROWS_AS(_ = json::from_bjdata(std::vector<uint8_t>(half.begin(), half.end() - 1)), json::parse_error&);
+            }
+
+            SECTION("ndarray of char rejects values above 127")
+            {
+                json _;
+                const std::vector<uint8_t> ok = {'[', '$', 'C', '#', '[', '$', 'U', '#', 'U', 2, 2, 2, 'a', 'b', 'c', 0x7F};
+                CHECK(json::from_bjdata(ok) == json::parse(R"({"_ArrayType_":"char","_ArraySize_":[2,2],"_ArrayData_":[97,98,99,127]})"));
+
+                std::vector<uint8_t> bad = ok;
+                bad.back() = 0xC8;
+                CHECK_THROWS_AS(_ = json::from_bjdata(bad), json::parse_error&);
+                CHECK(json::from_bjdata(bad, true, false).is_discarded());
+
+                // the same restriction holds for 1-D arrays
+                CHECK_THROWS_AS(_ = json::from_bjdata(std::vector<uint8_t>({'[', '$', 'C', '#', 'i', 1, 0xC8})), json::parse_error&);
+
+                // truncated input
+                CHECK_THROWS_AS(_ = json::from_bjdata(std::vector<uint8_t>(ok.begin(), ok.end() - 1)), json::parse_error&);
+
+                // 'B' (byte) elements may use the full range
+                const std::vector<uint8_t> byte_nd = {'[', '$', 'B', '#', '[', '$', 'U', '#', 'U', 2, 2, 2, 1, 2, 3, 0xC8};
+                CHECK(json::from_bjdata(byte_nd) == json::parse(R"({"_ArrayType_":"byte","_ArraySize_":[2,2],"_ArrayData_":[1,2,3,200]})"));
+            }
+
             SECTION("do not accept NTFZ markers in ndarray optimized type (with count)")
             {
                 json _;