Extract current_is_whitespace() to deduplicate skip_whitespace()'s two whitespace checks

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

.github/workflows/ubuntu.yml

@@ -100,7 +100,7 @@ jobs:
     container: ubuntu:focal
     strategy:
       matrix:
-        target: [ci_cmake_flags, ci_test_diagnostics, ci_test_diagnostic_positions, ci_test_noexceptions, ci_test_noimplicitconversions, ci_test_legacycomparison, ci_test_noglobaludls, ci_test_disableenumserialization, ci_test_skiplibraryversioncheck]
+        target: [ci_cmake_flags, ci_test_diagnostics, ci_test_diagnostic_positions, ci_test_noexceptions, ci_test_noimplicitconversions, ci_test_legacycomparison, ci_test_noglobaludls]
     steps:
       - name: Install build-essential
         run: apt-get update ; apt-get install -y build-essential unzip wget git libssl-dev

cmake/ci.cmake

@@ -242,40 +242,6 @@ add_custom_target(ci_test_noglobaludls
     COMMENT "Compile and test with global UDLs disabled"
 )
 
-###############################################################################
-# Disable enum serialization.
-###############################################################################
-
-add_custom_target(ci_test_disableenumserialization
-    COMMAND ${CMAKE_COMMAND}
-    -DCMAKE_BUILD_TYPE=Debug -GNinja
-    -DJSON_BuildTests=ON -DJSON_FastTests=ON -DJSON_DisableEnumSerialization=ON
-    -S${PROJECT_SOURCE_DIR} -B${PROJECT_BINARY_DIR}/build_disableenumserialization
-    COMMAND ${CMAKE_COMMAND} --build ${PROJECT_BINARY_DIR}/build_disableenumserialization
-    COMMAND cd ${PROJECT_BINARY_DIR}/build_disableenumserialization && ${CMAKE_CTEST_COMMAND} --parallel ${N} --output-on-failure
-    COMMENT "Compile and test with enum serialization disabled"
-)
-
-###############################################################################
-# Skip the multiple-inclusion library version check.
-###############################################################################
-
-# tests/src/skip_library_version_check.cpp deliberately simulates a scenario
-# (mixing two differently-versioned inclusions of the library in one
-# translation unit) that unavoidably triggers the compiler's own "macro
-# redefined" warning, so -- unlike the ci_test_* targets above -- it is
-# compiled directly here, with a modest warning set, instead of being folded
-# into the library's own -Weverything/-Werror unit test matrix.
-add_custom_target(ci_test_skiplibraryversioncheck
-    COMMAND ${CMAKE_COMMAND} -E make_directory ${PROJECT_BINARY_DIR}/skip_library_version_check
-    COMMAND ${CMAKE_CXX_COMPILER} -std=c++11 -Wall -Wextra
-        -I${PROJECT_SOURCE_DIR}/include
-        ${PROJECT_SOURCE_DIR}/tests/src/skip_library_version_check.cpp
-        -o ${PROJECT_BINARY_DIR}/skip_library_version_check/skip_library_version_check
-    COMMAND ${PROJECT_BINARY_DIR}/skip_library_version_check/skip_library_version_check
-    COMMENT "Compile and run a translation unit simulating a mismatched library version, with JSON_SKIP_LIBRARY_VERSION_CHECK defined"
-)
-
 ###############################################################################
 # Coverage.
 ###############################################################################

include/nlohmann/detail/input/lexer.hpp

@@ -149,10 +149,11 @@ class lexer : public lexer_base<BasicJsonType>
   public:
     using token_type = typename lexer_base<BasicJsonType>::token_type;
 
-    explicit lexer(InputAdapterType&& adapter, bool ignore_comments_ = false) noexcept
+    explicit lexer(InputAdapterType&& adapter, bool ignore_comments_ = false, bool discard_number_values_ = false) noexcept
         : ia(std::move(adapter))
         , ignore_comments(ignore_comments_)
         , decimal_point_char(static_cast<char_int_type>(get_decimal_point()))
+        , discard_number_values(discard_number_values_)
     {}
 
     // deleted because of pointer members
@@ -1279,6 +1280,58 @@ scan_number_done:
         // we are done scanning a number)
         unget();
 
+        // If the caller does not need the converted value (only whether the
+        // input is syntactically valid; see json_sax_acceptor/accept()), an
+        // unsigned/integer token can be reported without calling
+        // strtoull()/strtoll() at all, *provided* we can already tell from
+        // the digit count alone that the conversion cannot overflow 64 bits.
+        // Such tokens are always finite and are accepted unconditionally by
+        // the parser regardless of their actual value (parser::sax_parse_internal()
+        // never checks finiteness for value_unsigned/value_integer), so the
+        // classification below is all that is needed.
+        //
+        // A decimal number with up to 18 digits is always representable in
+        // both std::uint64_t and std::int64_t (18 nines is ~1e18, well below
+        // both UINT64_MAX ~1.8e19 and INT64_MAX ~9.2e18), so strtoull()/strtoll()
+        // could not have set errno to ERANGE for it. Numbers with more digits
+        // (rare in practice) fall through to the exact code below, unchanged,
+        // so their handling -- including reclassification to value_float when
+        // the value overflows 64 bits, and rejection when it is not even
+        // finite as a double -- is bit-for-bit identical to before this
+        // optimization.
+        //
+        // Note this reasons about std::uint64_t/std::int64_t, not about
+        // number_unsigned_t/number_integer_t (BasicJsonType's own, possibly
+        // narrower, template parameters -- e.g. std::uint32_t). That is fine
+        // *only* because discard_number_values is exclusively set by
+        // accept() (see json.hpp), and accept() always parses through the
+        // library's own json_sax_acceptor -- never a user-supplied SAX
+        // consumer -- whose number_unsigned()/number_integer()/number_float()
+        // callbacks unconditionally discard their argument and return true.
+        // So for every caller that can reach this branch, neither the token
+        // classification below nor the eventual (possibly narrowed, and on
+        // this fast path left stale/unset) value_unsigned/value_integer is
+        // ever consulted -- an unsigned/integer token is accepted outright,
+        // and even a >18-digit token that this fast path deliberately falls
+        // through for is, once reclassified to value_float, still finite
+        // (and thus accepted) for any digit count that fits in number_unsigned_t
+        // or number_integer_t regardless of that type's width. If this
+        // function is ever taught to run with discard_number_values true for
+        // a caller that *does* read the converted value, this reasoning (and
+        // the fast path below) would need to be revisited.
+        if (discard_number_values)
+        {
+            constexpr std::size_t safe_digit_count = 18;
+            if (number_type == token_type::value_unsigned && token_buffer.size() <= safe_digit_count)
+            {
+                return token_type::value_unsigned;
+            }
+            if (number_type == token_type::value_integer && token_buffer.size() - 1 <= safe_digit_count)
+            {
+                return token_type::value_integer;
+            }
+        }
+
         char* endptr = nullptr; // NOLINT(misc-const-correctness,cppcoreguidelines-pro-type-vararg,hicpp-vararg)
         errno = 0;
 
@@ -1393,8 +1446,7 @@ scan_number_done:
     */
     char_int_type get()
     {
-        ++position.chars_read_total;
-        ++position.chars_read_current_line;
+        advance_position();
 
         if (next_unget)
         {
@@ -1406,6 +1458,23 @@ scan_number_done:
             current = ia.get_character();
         }
 
+        return track_after_read();
+    }
+
+    /// shared head of get() / get_ignoring_pending_unget(): bump the
+    /// per-character position counters (line-count-on-'\n' bookkeeping is
+    /// handled afterwards, in track_after_read(), once `current` is known)
+    void advance_position() noexcept
+    {
+        ++position.chars_read_total;
+        ++position.chars_read_current_line;
+    }
+
+    /// shared tail of get() / get_ignoring_pending_unget(): capture the
+    /// character for error messages (if needed) and update line/column
+    /// bookkeeping for the character now in `current`
+    char_int_type track_after_read()
+    {
         // seekable adapters reconstruct the token lazily on error (see
         // get_token_string), so the eager per-character copy is skipped
         capture_char(std::integral_constant<bool, lazy_token_string> {});
@@ -1419,6 +1488,29 @@ scan_number_done:
         return current;
     }
 
+    /*!
+    @brief like get(), but for call sites that can prove no unget() is pending
+
+    get() has to check the `next_unget` flag on every call, because a
+    previous token may have ended with unget() (e.g. scan_number() always
+    ungets the character that terminated the number, so the next call to
+    scan() can see it again). skip_whitespace() reads that first,
+    possibly-ungotten character via a plain get(), but every further
+    character it reads is guaranteed to be a fresh read: nothing between
+    those calls invokes unget(). This variant skips the (otherwise always
+    false) next_unget branch for those calls; it is not a general
+    replacement for get().
+    */
+    char_int_type get_ignoring_pending_unget()
+    {
+        JSON_ASSERT(!next_unget);
+
+        advance_position();
+        current = ia.get_character();
+
+        return track_after_read();
+    }
+
     /// seekable adapter: nothing to capture, the token is rebuilt on error
     void capture_char(std::true_type /*lazy*/) const noexcept {}
 
@@ -1612,13 +1704,37 @@ scan_number_done:
         return true;
     }
 
+    /// whether `current` is one of the four JSON whitespace characters
+    bool current_is_whitespace() const noexcept
+    {
+        return current == ' ' || current == '\t' || current == '\n' || current == '\r';
+    }
+
     void skip_whitespace()
     {
+        // the first character may be a pending unget() left over from the
+        // previous token (see get_ignoring_pending_unget()); every
+        // subsequent character read by this loop is guaranteed fresh, since
+        // nothing below calls unget()
+        get();
+
+        if (!current_is_whitespace())
+        {
+            return;
+        }
+
+        // this is written as an if-guarded do-while (rather than a plain
+        // while loop) because that shape is what lets both GCC and Clang
+        // keep the input adapter's read pointer in a register across
+        // iterations; the equivalent while-loop measurably defeated that
+        // optimization in testing, turning long whitespace runs (e.g. the
+        // indentation of pretty-printed JSON) from a register-only loop
+        // into one that reloads the pointer from memory every character
         do
         {
-            get();
+            get_ignoring_pending_unget();
         }
-        while (current == ' ' || current == '\t' || current == '\n' || current == '\r');
+        while (current_is_whitespace());
     }
 
     token_type scan()
@@ -1754,6 +1870,13 @@ scan_number_done:
     const char_int_type decimal_point_char = '.';
     /// the position of the decimal point in the input
     std::size_t decimal_point_position = std::string::npos;
+
+    /// whether the caller (e.g. accept()/json_sax_acceptor) only needs the
+    /// token classification and never looks at the converted numeric value;
+    /// when set, scan_number() may skip strtoull()/strtoll() for
+    /// value_unsigned/value_integer tokens whose digit count guarantees they
+    /// fit into 64 bits (see scan_number())
+    const bool discard_number_values = false;
 };
 
 }  // namespace detail

include/nlohmann/detail/input/parser.hpp

@@ -72,9 +72,10 @@ class parser
                     parser_callback_t<BasicJsonType> cb = nullptr,
                     const bool allow_exceptions_ = true,
                     const bool ignore_comments = false,
-                    const bool ignore_trailing_commas_ = false)
+                    const bool ignore_trailing_commas_ = false,
+                    const bool discard_number_values_ = false)
         : callback(std::move(cb))
-        , m_lexer(std::move(adapter), ignore_comments)
+        , m_lexer(std::move(adapter), ignore_comments, discard_number_values_)
         , allow_exceptions(allow_exceptions_)
         , ignore_trailing_commas(ignore_trailing_commas_)
     {

include/nlohmann/json.hpp

@@ -164,11 +164,12 @@ class basic_json // NOLINT(cppcoreguidelines-special-member-functions,hicpp-spec
         detail::parser_callback_t<basic_json>cb = nullptr,
         const bool allow_exceptions = true,
         const bool ignore_comments = false,
-        const bool ignore_trailing_commas = false
+        const bool ignore_trailing_commas = false,
+        const bool discard_number_values = false
     )
     {
         return ::nlohmann::detail::parser<basic_json, InputAdapterType>(std::move(adapter),
-               std::move(cb), allow_exceptions, ignore_comments, ignore_trailing_commas);
+               std::move(cb), allow_exceptions, ignore_comments, ignore_trailing_commas, discard_number_values);
     }
 
   private:
@@ -4133,7 +4134,7 @@ class basic_json // NOLINT(cppcoreguidelines-special-member-functions,hicpp-spec
                        const bool ignore_comments = false,
                        const bool ignore_trailing_commas = false)
     {
-        return parser(detail::input_adapter(std::forward<InputType>(i)), nullptr, false, ignore_comments, ignore_trailing_commas).accept(true);
+        return parser(detail::input_adapter(std::forward<InputType>(i)), nullptr, false, ignore_comments, ignore_trailing_commas, true).accept(true);
     }
 
     /// @brief check if the input is valid JSON (iterator pair, or iterator+sentinel pair for C++20 ranges support)
@@ -4144,7 +4145,7 @@ class basic_json // NOLINT(cppcoreguidelines-special-member-functions,hicpp-spec
                        const bool ignore_comments = false,
                        const bool ignore_trailing_commas = false)
     {
-        return parser(detail::input_adapter(std::move(first), std::move(last)), nullptr, false, ignore_comments, ignore_trailing_commas).accept(true);
+        return parser(detail::input_adapter(std::move(first), std::move(last)), nullptr, false, ignore_comments, ignore_trailing_commas, true).accept(true);
     }
 
     JSON_HEDLEY_WARN_UNUSED_RESULT
@@ -4153,7 +4154,7 @@ class basic_json // NOLINT(cppcoreguidelines-special-member-functions,hicpp-spec
                        const bool ignore_comments = false,
                        const bool ignore_trailing_commas = false)
     {
-        return parser(i.get(), nullptr, false, ignore_comments, ignore_trailing_commas).accept(true);
+        return parser(i.get(), nullptr, false, ignore_comments, ignore_trailing_commas, true).accept(true);
     }
 
     /// @brief generate SAX events

single_include/nlohmann/json.hpp

@@ -7932,10 +7932,11 @@ class lexer : public lexer_base<BasicJsonType>
   public:
     using token_type = typename lexer_base<BasicJsonType>::token_type;
 
-    explicit lexer(InputAdapterType&& adapter, bool ignore_comments_ = false) noexcept
+    explicit lexer(InputAdapterType&& adapter, bool ignore_comments_ = false, bool discard_number_values_ = false) noexcept
         : ia(std::move(adapter))
         , ignore_comments(ignore_comments_)
         , decimal_point_char(static_cast<char_int_type>(get_decimal_point()))
+        , discard_number_values(discard_number_values_)
     {}
 
     // deleted because of pointer members
@@ -9062,6 +9063,58 @@ scan_number_done:
         // we are done scanning a number)
         unget();
 
+        // If the caller does not need the converted value (only whether the
+        // input is syntactically valid; see json_sax_acceptor/accept()), an
+        // unsigned/integer token can be reported without calling
+        // strtoull()/strtoll() at all, *provided* we can already tell from
+        // the digit count alone that the conversion cannot overflow 64 bits.
+        // Such tokens are always finite and are accepted unconditionally by
+        // the parser regardless of their actual value (parser::sax_parse_internal()
+        // never checks finiteness for value_unsigned/value_integer), so the
+        // classification below is all that is needed.
+        //
+        // A decimal number with up to 18 digits is always representable in
+        // both std::uint64_t and std::int64_t (18 nines is ~1e18, well below
+        // both UINT64_MAX ~1.8e19 and INT64_MAX ~9.2e18), so strtoull()/strtoll()
+        // could not have set errno to ERANGE for it. Numbers with more digits
+        // (rare in practice) fall through to the exact code below, unchanged,
+        // so their handling -- including reclassification to value_float when
+        // the value overflows 64 bits, and rejection when it is not even
+        // finite as a double -- is bit-for-bit identical to before this
+        // optimization.
+        //
+        // Note this reasons about std::uint64_t/std::int64_t, not about
+        // number_unsigned_t/number_integer_t (BasicJsonType's own, possibly
+        // narrower, template parameters -- e.g. std::uint32_t). That is fine
+        // *only* because discard_number_values is exclusively set by
+        // accept() (see json.hpp), and accept() always parses through the
+        // library's own json_sax_acceptor -- never a user-supplied SAX
+        // consumer -- whose number_unsigned()/number_integer()/number_float()
+        // callbacks unconditionally discard their argument and return true.
+        // So for every caller that can reach this branch, neither the token
+        // classification below nor the eventual (possibly narrowed, and on
+        // this fast path left stale/unset) value_unsigned/value_integer is
+        // ever consulted -- an unsigned/integer token is accepted outright,
+        // and even a >18-digit token that this fast path deliberately falls
+        // through for is, once reclassified to value_float, still finite
+        // (and thus accepted) for any digit count that fits in number_unsigned_t
+        // or number_integer_t regardless of that type's width. If this
+        // function is ever taught to run with discard_number_values true for
+        // a caller that *does* read the converted value, this reasoning (and
+        // the fast path below) would need to be revisited.
+        if (discard_number_values)
+        {
+            constexpr std::size_t safe_digit_count = 18;
+            if (number_type == token_type::value_unsigned && token_buffer.size() <= safe_digit_count)
+            {
+                return token_type::value_unsigned;
+            }
+            if (number_type == token_type::value_integer && token_buffer.size() - 1 <= safe_digit_count)
+            {
+                return token_type::value_integer;
+            }
+        }
+
         char* endptr = nullptr; // NOLINT(misc-const-correctness,cppcoreguidelines-pro-type-vararg,hicpp-vararg)
         errno = 0;
 
@@ -9176,8 +9229,7 @@ scan_number_done:
     */
     char_int_type get()
     {
-        ++position.chars_read_total;
-        ++position.chars_read_current_line;
+        advance_position();
 
         if (next_unget)
         {
@@ -9189,6 +9241,23 @@ scan_number_done:
             current = ia.get_character();
         }
 
+        return track_after_read();
+    }
+
+    /// shared head of get() / get_ignoring_pending_unget(): bump the
+    /// per-character position counters (line-count-on-'\n' bookkeeping is
+    /// handled afterwards, in track_after_read(), once `current` is known)
+    void advance_position() noexcept
+    {
+        ++position.chars_read_total;
+        ++position.chars_read_current_line;
+    }
+
+    /// shared tail of get() / get_ignoring_pending_unget(): capture the
+    /// character for error messages (if needed) and update line/column
+    /// bookkeeping for the character now in `current`
+    char_int_type track_after_read()
+    {
         // seekable adapters reconstruct the token lazily on error (see
         // get_token_string), so the eager per-character copy is skipped
         capture_char(std::integral_constant<bool, lazy_token_string> {});
@@ -9202,6 +9271,29 @@ scan_number_done:
         return current;
     }
 
+    /*!
+    @brief like get(), but for call sites that can prove no unget() is pending
+
+    get() has to check the `next_unget` flag on every call, because a
+    previous token may have ended with unget() (e.g. scan_number() always
+    ungets the character that terminated the number, so the next call to
+    scan() can see it again). skip_whitespace() reads that first,
+    possibly-ungotten character via a plain get(), but every further
+    character it reads is guaranteed to be a fresh read: nothing between
+    those calls invokes unget(). This variant skips the (otherwise always
+    false) next_unget branch for those calls; it is not a general
+    replacement for get().
+    */
+    char_int_type get_ignoring_pending_unget()
+    {
+        JSON_ASSERT(!next_unget);
+
+        advance_position();
+        current = ia.get_character();
+
+        return track_after_read();
+    }
+
     /// seekable adapter: nothing to capture, the token is rebuilt on error
     void capture_char(std::true_type /*lazy*/) const noexcept {}
 
@@ -9395,13 +9487,37 @@ scan_number_done:
         return true;
     }
 
+    /// whether `current` is one of the four JSON whitespace characters
+    bool current_is_whitespace() const noexcept
+    {
+        return current == ' ' || current == '\t' || current == '\n' || current == '\r';
+    }
+
     void skip_whitespace()
     {
+        // the first character may be a pending unget() left over from the
+        // previous token (see get_ignoring_pending_unget()); every
+        // subsequent character read by this loop is guaranteed fresh, since
+        // nothing below calls unget()
+        get();
+
+        if (!current_is_whitespace())
+        {
+            return;
+        }
+
+        // this is written as an if-guarded do-while (rather than a plain
+        // while loop) because that shape is what lets both GCC and Clang
+        // keep the input adapter's read pointer in a register across
+        // iterations; the equivalent while-loop measurably defeated that
+        // optimization in testing, turning long whitespace runs (e.g. the
+        // indentation of pretty-printed JSON) from a register-only loop
+        // into one that reloads the pointer from memory every character
         do
         {
-            get();
+            get_ignoring_pending_unget();
         }
-        while (current == ' ' || current == '\t' || current == '\n' || current == '\r');
+        while (current_is_whitespace());
     }
 
     token_type scan()
@@ -9537,6 +9653,13 @@ scan_number_done:
     const char_int_type decimal_point_char = '.';
     /// the position of the decimal point in the input
     std::size_t decimal_point_position = std::string::npos;
+
+    /// whether the caller (e.g. accept()/json_sax_acceptor) only needs the
+    /// token classification and never looks at the converted numeric value;
+    /// when set, scan_number() may skip strtoull()/strtoll() for
+    /// value_unsigned/value_integer tokens whose digit count guarantees they
+    /// fit into 64 bits (see scan_number())
+    const bool discard_number_values = false;
 };
 
 }  // namespace detail
@@ -14043,9 +14166,10 @@ class parser
                     parser_callback_t<BasicJsonType> cb = nullptr,
                     const bool allow_exceptions_ = true,
                     const bool ignore_comments = false,
-                    const bool ignore_trailing_commas_ = false)
+                    const bool ignore_trailing_commas_ = false,
+                    const bool discard_number_values_ = false)
         : callback(std::move(cb))
-        , m_lexer(std::move(adapter), ignore_comments)
+        , m_lexer(std::move(adapter), ignore_comments, discard_number_values_)
         , allow_exceptions(allow_exceptions_)
         , ignore_trailing_commas(ignore_trailing_commas_)
     {
@@ -21592,11 +21716,12 @@ class basic_json // NOLINT(cppcoreguidelines-special-member-functions,hicpp-spec
         detail::parser_callback_t<basic_json>cb = nullptr,
         const bool allow_exceptions = true,
         const bool ignore_comments = false,
-        const bool ignore_trailing_commas = false
+        const bool ignore_trailing_commas = false,
+        const bool discard_number_values = false
                                  )
     {
         return ::nlohmann::detail::parser<basic_json, InputAdapterType>(std::move(adapter),
-            std::move(cb), allow_exceptions, ignore_comments, ignore_trailing_commas);
+            std::move(cb), allow_exceptions, ignore_comments, ignore_trailing_commas, discard_number_values);
     }
 
   private:
@@ -25561,7 +25686,7 @@ class basic_json // NOLINT(cppcoreguidelines-special-member-functions,hicpp-spec
                        const bool ignore_comments = false,
                        const bool ignore_trailing_commas = false)
     {
-        return parser(detail::input_adapter(std::forward<InputType>(i)), nullptr, false, ignore_comments, ignore_trailing_commas).accept(true);
+        return parser(detail::input_adapter(std::forward<InputType>(i)), nullptr, false, ignore_comments, ignore_trailing_commas, true).accept(true);
     }
 
     /// @brief check if the input is valid JSON (iterator pair, or iterator+sentinel pair for C++20 ranges support)
@@ -25572,7 +25697,7 @@ class basic_json // NOLINT(cppcoreguidelines-special-member-functions,hicpp-spec
                        const bool ignore_comments = false,
                        const bool ignore_trailing_commas = false)
     {
-        return parser(detail::input_adapter(std::move(first), std::move(last)), nullptr, false, ignore_comments, ignore_trailing_commas).accept(true);
+        return parser(detail::input_adapter(std::move(first), std::move(last)), nullptr, false, ignore_comments, ignore_trailing_commas, true).accept(true);
     }
 
     JSON_HEDLEY_WARN_UNUSED_RESULT
@@ -25581,7 +25706,7 @@ class basic_json // NOLINT(cppcoreguidelines-special-member-functions,hicpp-spec
                        const bool ignore_comments = false,
                        const bool ignore_trailing_commas = false)
     {
-        return parser(i.get(), nullptr, false, ignore_comments, ignore_trailing_commas).accept(true);
+        return parser(i.get(), nullptr, false, ignore_comments, ignore_trailing_commas, true).accept(true);
     }
 
     /// @brief generate SAX events

tests/src/skip_library_version_check.cpp

@@ -1,61 +0,0 @@
-//     __ _____ _____ _____
-//  __|  |   __|     |   | |  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
-
-// Standalone compile-and-run check for the JSON_SKIP_LIBRARY_VERSION_CHECK
-// configuration macro, which (per #5423) was never exercised anywhere in the
-// test matrix.
-//
-// include/nlohmann/detail/abi_macros.hpp normally emits a #warning if
-// NLOHMANN_JSON_VERSION_MAJOR/MINOR/PATCH are already defined (as they would
-// be by an earlier inclusion of a different version of the library) with
-// values that mismatch the version about to be defined -- unless
-// JSON_SKIP_LIBRARY_VERSION_CHECK is defined, in which case the check (and
-// that #warning) is skipped.
-//
-// This file deliberately is not named tests/src/unit-*.cpp: it is compiled
-// directly (with a modest, non-strict warning set) by the dedicated
-// ci_test_skiplibraryversioncheck target in cmake/ci.cmake, rather than being
-// folded into the library's own -Weverything/-Werror unit test matrix. That
-// is because the scenario simulated here -- mixing two different, already
-// differently-versioned inclusions of the library in one translation unit --
-// unavoidably also triggers the *compiler's own* "macro redefined" warning,
-// independent of (and unaffected by) JSON_SKIP_LIBRARY_VERSION_CHECK, which
-// only ever silences the library's own #warning. Building this file under
-// -Weverything -Werror would therefore fail for a reason unrelated to the
-// macro under test.
-#define NLOHMANN_JSON_VERSION_MAJOR 0
-#define NLOHMANN_JSON_VERSION_MINOR 0
-#define NLOHMANN_JSON_VERSION_PATCH 0
-
-#define JSON_SKIP_LIBRARY_VERSION_CHECK 1
-
-#include <nlohmann/json.hpp>
-
-int main()
-{
-    // reaching this point at all already proves that the mismatched,
-    // pre-defined version macros above did not stop compilation -- which is
-    // exactly what JSON_SKIP_LIBRARY_VERSION_CHECK is for. The library must
-    // also still be fully usable.
-    const nlohmann::json j = {{"a", 1}, {"b", {1, 2, 3}}};
-    if (j.dump() != "{\"a\":1,\"b\":[1,2,3]}")
-    {
-        return 1;
-    }
-
-    // include/nlohmann/detail/abi_macros.hpp unconditionally (re)defines the
-    // version macros to the library's real, current version right after the
-    // (here, skipped) mismatch check, regardless of the deliberately wrong
-    // stand-in values defined above.
-    if (NLOHMANN_JSON_VERSION_MAJOR == 0 && NLOHMANN_JSON_VERSION_MINOR == 0 && NLOHMANN_JSON_VERSION_PATCH == 0)
-    {
-        return 1;
-    }
-
-    return 0;
-}

tests/src/unit-byte_container_with_subtype.cpp

@@ -42,39 +42,6 @@ TEST_CASE("byte_container_with_subtype")
         CHECK(container.subtype() == static_cast<subtype_type>(-1));
     }
 
-    SECTION("move semantics")
-    {
-        // the rvalue-reference constructor (without a subtype) must actually move
-        // the passed-in container rather than copy it; comparing the buffer address
-        // before and after is a stronger check than just observing the source is
-        // empty afterward, since a copy-then-clear could also leave it empty
-        {
-            std::vector<std::uint8_t> bytes = {{0xCA, 0xFE, 0xBA, 0xBE}};
-            const auto* const data_ptr = bytes.data();
-
-            nlohmann::byte_container_with_subtype<std::vector<std::uint8_t>> container(std::move(bytes));
-
-            CHECK(container.size() == 4);
-            CHECK(container.data() == data_ptr);
-            CHECK(!container.has_subtype());
-            CHECK(bytes.empty()); // NOLINT(bugprone-use-after-move,clang-analyzer-cplusplus.Move,hicpp-invalid-access-moved)
-        }
-
-        // same check for the rvalue-reference constructor that also takes a subtype
-        {
-            std::vector<std::uint8_t> bytes = {{0xCA, 0xFE, 0xBA, 0xBE}};
-            const auto* const data_ptr = bytes.data();
-
-            nlohmann::byte_container_with_subtype<std::vector<std::uint8_t>> container(std::move(bytes), 42);
-
-            CHECK(container.size() == 4);
-            CHECK(container.data() == data_ptr);
-            CHECK(container.has_subtype());
-            CHECK(container.subtype() == 42);
-            CHECK(bytes.empty()); // NOLINT(bugprone-use-after-move,clang-analyzer-cplusplus.Move,hicpp-invalid-access-moved)
-        }
-    }
-
     SECTION("comparisons")
     {
         std::vector<std::uint8_t> const bytes = {{0xCA, 0xFE, 0xBA, 0xBE}};

tests/src/unit-class_parser.cpp

@@ -930,6 +930,94 @@ TEST_CASE("parser class")
                 CHECK(accept_helper("+1") == false);
                 CHECK(accept_helper("+0") == false);
             }
+
+            SECTION("issue #5411 - skip conversion when accept() does not need the numeric value")
+            {
+                // lexer::scan_number() may skip strtoull()/strtoll() for
+                // value_unsigned/value_integer tokens when the caller (e.g.
+                // json::accept()) does not need the converted value, as long
+                // as the digit count alone guarantees no 64-bit overflow (see
+                // the "safe_digit_count" fast path in scan_number()). This
+                // differential test checks that json::accept() (which enables
+                // the fast path) and json::parse() (which never does) always
+                // agree, over a corpus that exercises both the fast path
+                // (<=18 digits) and the untouched, exact fallback path (>=19
+                // digits) -- including reclassification of huge digit-only
+                // integers to a (possibly non-finite) floating-point value.
+                const std::vector<std::pair<std::string, bool>> cases =
+                {
+                    // normal small/large integers, both signs
+                    {"0", true}, {"1", true}, {"-1", true}, {"42", true}, {"-42", true},
+                    {"123456789", true}, {"-123456789", true},
+
+                    // digit-count boundary around the 18-digit safe cutoff (both signs)
+                    {std::string(17, '9'), true},
+                    {std::string(18, '9'), true},
+                    {std::string(19, '9'), true},
+                    {std::string(20, '9'), true},
+                    {"-" + std::string(17, '9'), true},
+                    {"-" + std::string(18, '9'), true},
+                    {"-" + std::string(19, '9'), true},
+                    {"-" + std::string(20, '9'), true},
+
+                    // 64-bit boundaries
+                    {"9223372036854775807", true},    // INT64_MAX
+                    {"-9223372036854775808", true},   // INT64_MIN
+                    {"18446744073709551615", true},   // UINT64_MAX
+                    {"18446744073709551616", true},   // UINT64_MAX + 1 (overflows uint64_t, finite double)
+
+                    // the 28-digit example from the issue: overflows uint64_t
+                    // but is finite as a double, so the scanner reclassifies
+                    // it to value_float and it is accepted
+                    {"9999999999999999999999999999", true},
+
+                    // huge digit-only integers that overflow even a double -> rejected
+                    {std::string(309, '9'), false},
+                    {std::string(400, '9'), false},
+                    {"1" + std::string(400, '0'), false},
+
+                    // 1e999 / 1e400 style overflow -> rejected
+                    {"1e999", false},
+                    {"1e400", false},
+                    {"-1e999", false},
+                    {"1E999", false},
+
+                    // values straddling DBL_MAX
+                    {"1.7976931348623157e308", true},   // <= DBL_MAX, finite
+                    {"1.7976931348623159e308", false},  // > DBL_MAX, overflows to inf
+
+                    // a mix of other valid/invalid numeric syntax
+                    {"3.14159", true},
+                    {"-0.0", true},
+                    {"1.0e10", true},
+                    {"01", false},
+                    {"-", false},
+                    {"1.", false},
+                    {"1e", false},
+                    {"+1", false},
+                };
+
+                for (const auto& c : cases)
+                {
+                    const std::string& number = c.first;
+                    const bool expected = c.second;
+                    CAPTURE(number)
+                    CAPTURE(expected)
+
+                    // accept() takes the fast path (skips conversion when possible)
+                    CHECK(json::accept(number) == expected);
+
+                    // parse() always performs the full conversion; it must agree
+                    json j;
+                    CHECK_NOTHROW(json::parser(nlohmann::detail::input_adapter(number), nullptr, false).parse(true, j));
+                    CHECK(!j.is_discarded() == expected);
+
+                    // wrap in an array so get_token() is exercised beyond the
+                    // very first (constructor-time) scan as well
+                    const std::string wrapped = "[" + number + "," + number + "]";
+                    CHECK(json::accept(wrapped) == expected);
+                }
+            }
         }
     }
 
@@ -1394,6 +1482,62 @@ TEST_CASE("parser class")
         CHECK(accept_helper("\"\\uD80C\\uFFFF\"") == false);
     }
 
+    SECTION("issue #5412 - whitespace skipping bookkeeping (compact vs. pretty-printed)")
+    {
+        // lexer::skip_whitespace() reads its first character with get() (to
+        // honor a possibly pending unget() from the previous token) and every
+        // further whitespace character with get_ignoring_pending_unget() (a
+        // get() variant that skips the then-always-false next_unget check).
+        // This must not change the reported byte offset, line, or column of
+        // a syntax error, even when a long run of whitespace containing
+        // multiple newlines is skipped beforehand (as with pretty-printed
+        // input). The expected values below were captured from the
+        // unmodified do-while(get()) loop, so any regression that miscounts
+        // characters or newlines while skipping whitespace changes them.
+        const auto check_error = [](const std::string & input, std::size_t expected_byte,
+                                    const std::string & expected_what)
+        {
+            CAPTURE(input)
+            try
+            {
+                json _ = json::parse(input);
+                FAIL_CHECK("expected a parse_error, but parsing succeeded");
+            }
+            catch (const json::parse_error& e)
+            {
+                CHECK(e.byte == expected_byte);
+                CHECK(std::string(e.what()) == expected_what);
+            }
+        };
+
+        // a nested document, serialized both compactly and pretty-printed
+        // (dump(4)), each truncated right before the final closing '}' so
+        // that the parser hits EOF after skipping all of the (in the
+        // pretty-printed case, substantial) indentation whitespace
+        const json doc =
+        {
+            {"a", 1},
+            {"b", json::array({true, false, nullptr, "x"})},
+            {"c", json::object({{"d", 3.14}, {"e", json::array({1, 2, 3})}})}
+        };
+
+        const std::string compact = doc.dump();
+        const std::string pretty = doc.dump(4);
+
+        check_error(compact.substr(0, compact.size() - 1), 60,
+                    "[json.exception.parse_error.101] parse error at line 1, column 60: syntax error while parsing object - unexpected end of input; expected '}'");
+        check_error(pretty.substr(0, pretty.size() - 1), 193,
+                    "[json.exception.parse_error.101] parse error at line 17, column 1: syntax error while parsing object - unexpected end of input; expected '}'");
+
+        // an invalid token appearing after several indented, multi-line
+        // whitespace runs vs. the same document without any of that
+        // whitespace
+        check_error("{\n    \"a\": 1,\n    \"b\": [\n        true,\n        false\n    ],\n    \"c\": @\n}", 70,
+                    "[json.exception.parse_error.101] parse error at line 7, column 10: syntax error while parsing value - invalid literal; last read: '\"c\": @'");
+        check_error("{\"a\":1,\"b\":[true,false],\"c\":@}", 29,
+                    "[json.exception.parse_error.101] parse error at line 1, column 29: syntax error while parsing value - invalid literal; last read: '\"c\":@'");
+    }
+
     SECTION("tests found by mutate++")
     {
         // test case to make sure no comma precedes the first key

tests/src/unit-conversions.cpp

@@ -1761,33 +1761,6 @@ TEST_CASE("std::filesystem::path")
 }
 #endif
 
-// the ADL to_json overload for std::u8string only exists under the same guard
-// as std::filesystem::path support (it is otherwise only reached indirectly,
-// via std::filesystem::path::u8string()) -- mirror both #if conditions from
-// include/nlohmann/detail/conversions/to_json.hpp exactly
-#if JSON_HAS_FILESYSTEM || JSON_HAS_EXPERIMENTAL_FILESYSTEM
-#if defined(__cpp_lib_char8_t)
-TEST_CASE("std::u8string")
-{
-    SECTION("ascii")
-    {
-        const std::u8string s = u8"Path";
-        json const j = s;
-
-        CHECK(j.template get<std::string>() == "Path");
-    }
-
-    SECTION("utf-8")
-    {
-        const std::u8string s = u8"P\xc4\x9b\xc5\xa1ina";
-        json const j = s;
-
-        CHECK(j.template get<std::string>() == "P\xc4\x9b\xc5\xa1ina");
-    }
-}
-#endif
-#endif
-
 TEST_CASE("std::optional")
 {
     SECTION("null")

tests/src/unit-json_patch.cpp

@@ -672,94 +672,6 @@ TEST_CASE("JSON patch")
         }
     }
 
-    SECTION("patch_inplace")
-    {
-        SECTION("happy path: patch_inplace mirrors patch() on success")
-        {
-            // mirrors "A.5. Replacing a Value" above, but applies the patch with
-            // patch_inplace() to a mutable copy instead of using patch()'s
-            // returned copy
-            json doc = R"(
-                    {
-                        "baz": "qux",
-                        "foo": "bar"
-                    }
-                )"_json;
-
-            json const patch = R"(
-                    [
-                        { "op": "replace", "path": "/baz", "value": "boo" }
-                    ]
-                )"_json;
-
-            json const expected = R"(
-                    {
-                        "baz": "boo",
-                        "foo": "bar"
-                    }
-                )"_json;
-
-            doc.patch_inplace(patch);
-            CHECK(doc == expected);
-        }
-
-        // this test relies on the "test" operation actually throwing so the
-        // partial-application state can be observed right after the throw
-        // point; under JSON_NOEXCEPTION, JSON_THROW() calls std::abort()
-        // instead (there is no C++ exception to throw), and doctest's
-        // CHECK_THROWS_AS() is compiled out to a no-op that never even
-        // invokes the given expression (see doctest's "--no-throw" test
-        // filter, which ci_test_noexceptions passes) -- so patch()/
-        // patch_inplace() would never be called at all and the follow-up
-        // state assertions below would fail against the untouched original
-#if !defined(JSON_NOEXCEPTION)
-        SECTION("distinguishing contract vs patch(): partial application on failure")
-        {
-            // Unlike patch(), which is all-or-nothing because it applies the
-            // patch to an internal copy that is simply discarded when an
-            // exception is thrown (leaving the original untouched no matter
-            // what), patch_inplace() mutates the document it is called on
-            // directly and immediately, operation by operation. So if a JSON
-            // Patch fails partway through, whatever operations already
-            // succeeded remain applied -- the document is left in a partially
-            // patched state. This is empirically verified current behavior,
-            // not just documented intent, and is pinned here as such.
-            json const original = R"(
-                    {
-                        "baz": "qux",
-                        "foo": "bar"
-                    }
-                )"_json;
-
-            // the first operation ("replace") succeeds; the second ("test")
-            // fails because the value at "/baz" no longer (and never did)
-            // equal "not boo"
-            json const patch = R"(
-                    [
-                        { "op": "replace", "path": "/baz", "value": "boo" },
-                        { "op": "test", "path": "/baz", "value": "not boo" }
-                    ]
-                )"_json;
-
-            // patch() never modifies the object it is called on -- it always
-            // operates on (and returns) a separate copy, so the original is
-            // left completely untouched, regardless of success or failure
-            json copy_for_patch = original;
-            CHECK_THROWS_AS(copy_for_patch.patch(patch), json::other_error&);
-            CHECK(copy_for_patch == original);
-
-            // patch_inplace(), in contrast, already applied the successful
-            // "replace" operation to the document before the "test" operation
-            // threw -- that change is not rolled back
-            json doc = original;
-            CHECK_THROWS_AS(doc.patch_inplace(patch), json::other_error&);
-            CHECK(doc != original);
-            CHECK(doc.at("baz") == "boo");
-            CHECK(doc.at("foo") == "bar");
-        }
-#endif // !defined(JSON_NOEXCEPTION)
-    }
-
     SECTION("errors")
     {
         SECTION("unknown operation")

tests/src/unit-no-mem-leak-on-adl-serialize.cpp

@@ -70,7 +70,7 @@ TEST_CASE("check_for_mem_leak_on_adl_to_json-2")
     }
 }
 
-TEST_CASE("check_for_mem_leak_on_adl_to_json-3")
+TEST_CASE("check_for_mem_leak_on_adl_to_json-2")
 {
     try
     {

tests/src/unit-no_io_and_user_exceptions.cpp

@@ -1,69 +0,0 @@
-//     __ _____ _____ _____
-//  __|  |   __|     |   | |  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
-
-// This translation unit is a dedicated, small compile-and-run check for two
-// configuration macros that (per #5423) were never exercised anywhere in the
-// test matrix:
-// - JSON_NO_IO, which removes the library's <istream>/<ostream> support
-//   (operator<<, operator>>, and the stream-based overloads of dump()/parse())
-// - the JSON_THROW_USER / JSON_TRY_USER / JSON_CATCH_USER trio, which lets a
-//   user replace the library's internal exception handling
-//
-// Both macros are about excluding/replacing a facility the library would
-// otherwise pull in on its own, and defining one has no bearing on the other,
-// so -- to keep the test matrix small -- they are exercised together in a
-// single dedicated file instead of two.
-//
-// JSON_NO_IO requires this file itself to never rely on <iostream>/<sstream>;
-// only string-based parsing/dumping is used below.
-#define JSON_NO_IO 1
-
-// The user-supplied exception macros below are a *conforming* replacement:
-// they simply forward to the real throw/try/catch keywords (via a counter so
-// the test can assert each macro was actually invoked, not just defined), so
-// every exception-related behavior the library relies on internally --
-// including rethrowing std::out_of_range as json::out_of_range in at() --
-// keeps working exactly as it would with the library's own default macros.
-static int json_throw_user_call_count = 0; // NOLINT(cppcoreguidelines-avoid-non-const-global-variables)
-
-#define JSON_THROW_USER(exception) do { ++json_throw_user_call_count; throw exception; } while (false) // NOLINT(cppcoreguidelines-macro-usage)
-#define JSON_TRY_USER try // NOLINT(cppcoreguidelines-macro-usage)
-#define JSON_CATCH_USER(exception) catch (exception) // NOLINT(cppcoreguidelines-macro-usage)
-
-#include "doctest_compatibility.h"
-
-#include <nlohmann/json.hpp>
-using json = nlohmann::json;
-
-TEST_CASE("JSON_NO_IO")
-{
-    // everything that does not touch <istream>/<ostream> must keep working:
-    // parsing from and dumping to std::string
-    const json j = json::parse("{\"a\":[1,2,3],\"b\":true}");
-    CHECK(j.dump() == "{\"a\":[1,2,3],\"b\":true}");
-    CHECK(j.at("a").size() == 3);
-    CHECK(j.at("b").get<bool>() == true);
-}
-
-TEST_CASE("JSON_THROW_USER, JSON_TRY_USER, JSON_CATCH_USER")
-{
-    json_throw_user_call_count = 0;
-
-    // a parse error goes through JSON_THROW directly, i.e., through our
-    // JSON_THROW_USER override
-    CHECK_THROWS_AS(json::parse("this is not JSON"), json::parse_error&);
-    CHECK(json_throw_user_call_count > 0);
-
-    // at() on an out-of-range array index internally catches std::out_of_range
-    // (JSON_TRY_USER/JSON_CATCH_USER) and rethrows it as json::out_of_range
-    // (JSON_THROW_USER again), so this exercises all three macros together
-    const int count_before = json_throw_user_call_count;
-    const json arr = json::array({1, 2, 3});
-    CHECK_THROWS_AS(arr.at(10), json::out_of_range&);
-    CHECK(json_throw_user_call_count > count_before);
-}

tests/src/unit-regression2.cpp

@@ -18,14 +18,6 @@
 // for some reason including this after the json header leads to linker errors with VS 2017...
 #include <locale>
 
-// skip tests if JSON_DisableEnumSerialization=ON (#4384): std::byte is a
-// scoped enum, so get<std::byte>() (needed below to get<std::vector<std::byte>>()
-// from a plain JSON array, not just from an already-binary value) relies on
-// enum serialization being enabled
-#if defined(JSON_DISABLE_ENUM_SERIALIZATION) && (JSON_DISABLE_ENUM_SERIALIZATION == 1)
-    #define SKIP_TESTS_FOR_ENUM_SERIALIZATION
-#endif
-
 #define JSON_TESTS_PRIVATE
 #include <nlohmann/json.hpp>
 using json = nlohmann::json;
@@ -1144,7 +1136,6 @@ TEST_CASE("regression tests 2")
         CHECK((decoded == json_4804::array()));
     }
 
-#ifndef SKIP_TESTS_FOR_ENUM_SERIALIZATION
     SECTION("discussion #4209 - custom BinaryType direct assignment and round-tripping")
     {
         // Test that assigning a custom BinaryType directly creates a binary value, not an array
@@ -1178,7 +1169,6 @@ TEST_CASE("regression tests 2")
         CHECK(extracted[1] == std::byte{2});
         CHECK(extracted[2] == std::byte{3});
     }
-#endif
 
     SECTION("issue #5046 - implicit conversion of return json to std::optional no longer implicit")
     {

tests/src/unit-std-format.cpp

@@ -52,23 +52,6 @@ TEST_CASE("std::formatter<nlohmann::json>")
         CHECK(std::format("{:2}", j) == j.dump(2));
         CHECK(std::format("{:#2}", j) == j.dump(2));
         CHECK(std::format("{:8}", j) == j.dump(8));
-        // multi-digit widths must accumulate every digit, not just the first
-        CHECK(std::format("{:12}", j) == j.dump(12));
-        CHECK(std::format("{:#12}", j) == j.dump(12));
-        CHECK(std::format("{:10}", j) == j.dump(10));
-    }
-
-    SECTION("bare alignment with no fill character defaults to a space indent character")
-    {
-        const json j = {{"foo", 1}, {"bar", {1, 2, 3}}};
-        // without a preceding fill character, the alignment character itself must not
-        // be mistaken for the indent character -- the default space is kept
-        CHECK(std::format("{:<}", j) == j.dump());
-        CHECK(std::format("{:>}", j) == j.dump());
-        CHECK(std::format("{:^}", j) == j.dump());
-        CHECK(std::format("{:<3}", j) == j.dump(3, ' '));
-        CHECK(std::format("{:>3}", j) == j.dump(3, ' '));
-        CHECK(std::format("{:^3}", j) == j.dump(3, ' '));
     }
 
     SECTION("fill-and-align sets the indent character, like dump(indent, indent_char)")