Merge branch 'develop' into claude/fix-issue-3989-db7e45

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

.github/CONTRIBUTING.md

@@ -111,6 +111,8 @@ context which existing file needs to be extended, and only very few cases requir
 When fixing a bug, edit `unit-regression3.cpp` and add a section referencing the fixed issue.
 `unit-regression2.cpp` holds the older tests; the two files exist because a single one grew large enough for the
 MinGW linker to fail relocating it, so please keep adding to the smaller file rather than growing the larger one.
+Tests that call `sax_parse` go into `unit-sax_parse.cpp` rather than into a large test file: every call instantiates
+the parser and the binary reader that recover from errors, which grows a test file considerably.
 
 #### Exceptions
 

README.md

@@ -495,7 +495,7 @@ bool key(string_t& val);
 bool parse_error(std::size_t position, const std::string& last_token, const detail::exception& ex);
 ```
 
-The return value of each function determines whether parsing should proceed.
+The return value of each function determines whether parsing should proceed. For `parse_error`, returning `true` [recovers from the error](https://json.nlohmann.me/features/parsing/error_recovery/): the parser repairs the input and continues.
 
 To implement your own SAX handler, proceed as follows:
 
@@ -503,7 +503,7 @@ To implement your own SAX handler, proceed as follows:
 2. Create an object of your SAX interface class, e.g. `my_sax`.
 3. Call `bool json::sax_parse(input, &my_sax)`; where the first parameter can be any input like a string or an input stream and the second parameter is a pointer to your SAX interface.
 
-Note the `sax_parse` function only returns a `bool` indicating the result of the last executed SAX event. It does not return a  `json` value - it is up to you to decide what to do with the SAX events. Furthermore, no exceptions are thrown in case of a parse error -- it is up to you what to do with the exception object passed to your `parse_error` implementation. Internally, the SAX interface is used for the DOM parser (class `json_sax_dom_parser`) as well as the acceptor (`json_sax_acceptor`), see file [`json_sax.hpp`](https://github.com/nlohmann/json/blob/develop/include/nlohmann/detail/input/json_sax.hpp).
+Note the `sax_parse` function only returns a `bool` indicating whether the input was parsed without errors and no SAX event returned `false`. It does not return a  `json` value - it is up to you to decide what to do with the SAX events. Furthermore, no exceptions are thrown in case of a parse error -- it is up to you what to do with the exception object passed to your `parse_error` implementation. Internally, the SAX interface is used for the DOM parser (class `json_sax_dom_parser`) as well as the acceptor (`json_sax_acceptor`), see file [`json_sax.hpp`](https://github.com/nlohmann/json/blob/develop/include/nlohmann/detail/input/json_sax.hpp).
 
 ### STL-like access
 

docs/mkdocs/docs/api/basic_json/sax_parse.md

@@ -104,7 +104,9 @@ The SAX event lister must follow the interface of [`json_sax`](../json_sax/index
 
 ## Return value
 
-return value of the last processed SAX event
+`#!cpp true` if the input was parsed without errors and no SAX event returned `#!cpp false`; `#!cpp false` otherwise.
+In particular, the result is `#!cpp false` for input with errors, even if the SAX parser recovered from all of them
+(see [error recovery](../../features/parsing/error_recovery.md)).
 
 ## Exception safety
 
@@ -161,6 +163,7 @@ use the same file with stdio functions, but it must not wait for another thread
 - Added `ignore_trailing_commas` in version 3.13.0.
 - Added `tag_handler` in version 3.13.0.
 - Extended container support (1) to include types with lvalue-only ADL `begin`/`end` (matching `std::begin`/`std::end` semantics) in version 3.13.0.
+- Recovering from parse errors (see [`parse_error`](../json_sax/parse_error.md)) added in version 3.13.0.
 - Extended overload (2) to accept heterogeneous iterator+sentinel pairs (C++20 ranges support) in version 3.13.0.
 - `JSON_PRECISE_STREAM_POSITION` added in version 3.13.0 to optionally leave a `#!cpp std::istream` positioned right
   after the parsed value when `strict` is `#!cpp false`.

docs/mkdocs/docs/api/json_sax/index.md

@@ -7,7 +7,8 @@ struct json_sax;
 
 This class describes the SAX interface used by [sax_parse](../basic_json/sax_parse.md). Each function is called in
 different situations while the input is parsed. The boolean return value informs the parser whether to continue
-processing the input.
+processing the input; for [`parse_error`](parse_error.md), it decides whether to
+[recover from the error](../../features/parsing/error_recovery.md).
 
 For instance, parsing the JSON text `{"a": [1, true]}` triggers the following callbacks, in order:
 

docs/mkdocs/docs/api/json_sax/parse_error.md

@@ -21,11 +21,18 @@ A parse error occurred.
 
 ## Return value
 
-Whether parsing should proceed (**must return `#!cpp false`**).
+Whether to recover from the error:
+
+- `#!cpp false` stops parsing.
+- `#!cpp true` recovers from the error: the error is repaired and parsing continues. If that is not possible, which
+  happens in the binary formats when the end of the item with the error is unknown, the value read so far is completed
+  and parsing stops. See [error recovery](../../features/parsing/error_recovery.md) for how errors are repaired.
+
+Either way, [`sax_parse`](../basic_json/sax_parse.md) returns `#!cpp false`.
 
 ## Examples
 
-??? example
+??? example "Example: (1) the SAX interface"
 
     The example below shows how the SAX interface is used.
 
@@ -39,12 +46,29 @@ Whether parsing should proceed (**must return `#!cpp false`**).
     --8<-- "examples/sax_parse.output"
     ```
 
+??? example "Example: (2) recovering from errors"
+
+    The example below shows how a SAX parser recovers from errors.
+
+    ```cpp
+    --8<-- "examples/sax_parse__error_recovery.cpp"
+    ```
+    
+    Output:
+    
+    ```
+    --8<-- "examples/sax_parse__error_recovery.output"
+    ```
+
 ## See also
 
 - [sax_parse](../basic_json/sax_parse.md) - SAX parser
 - [Parsing and Exceptions](../../features/parsing/parse_exceptions.md) - the article on handling parse errors without
   exceptions
+- [Error Recovery](../../features/parsing/error_recovery.md) - the article on recovering from parse errors
 
 ## Version history
 
 - Added in version 3.2.0.
+- Returning `#!cpp true` recovers from the error since version 3.13.0; before, parsing stopped, but the result of
+  [`sax_parse`](../basic_json/sax_parse.md) could be wrong.

docs/mkdocs/docs/examples/sax_parse__error_recovery.cpp

@@ -0,0 +1,43 @@
+#include <iostream>
+#include <iomanip>
+#include <nlohmann/json.hpp>
+
+using json = nlohmann::json;
+
+// a SAX parser that creates a JSON value like json::parse does, but that
+// recovers from parse errors instead of stopping at the first one
+class recovering_parser : public nlohmann::detail::json_sax_dom_parser<json>
+{
+  public:
+    explicit recovering_parser(json& result)
+        : nlohmann::detail::json_sax_dom_parser<json>(result, false)
+    {}
+
+    bool parse_error(std::size_t position,
+                     const std::string& /*last_token*/,
+                     const json::exception& ex)
+    {
+        std::cout << "byte " << position << ": " << ex.what() << '\n';
+
+        // repair the input and continue
+        return true;
+    }
+};
+
+int main()
+{
+    // JSON text with several mistakes that ends too early
+    const std::string text = R"({
+    "name": "Hello World",
+    "tags": ["a" "b",],
+    "valid": tru,
+    "size": 1.,
+    "nested": {"x": 1)";
+
+    json result;
+    recovering_parser sax(result);
+    const bool valid = json::sax_parse(text, &sax);
+
+    std::cout << "\nvalid JSON: " << std::boolalpha << valid << '\n'
+              << std::setw(4) << result << std::endl;
+}

docs/mkdocs/docs/examples/sax_parse__error_recovery.output

@@ -0,0 +1,19 @@
+byte 49: [json.exception.parse_error.101] parse error at line 3, column 20: syntax error while parsing array - unexpected string literal; expected ']'
+byte 51: [json.exception.parse_error.101] parse error at line 3, column 22: syntax error while parsing value - unexpected ']'; expected '[', '{', or a literal
+byte 70: [json.exception.parse_error.101] parse error at line 4, column 17: syntax error while parsing value - invalid literal; last read: '"valid": tru,'
+byte 86: [json.exception.parse_error.101] parse error at line 5, column 15: syntax error while parsing value - invalid number; expected digit after '.'; last read: '1.,'
+byte 109: [json.exception.parse_error.101] parse error at line 6, column 22: syntax error while parsing object - unexpected end of input; expected '}'
+
+valid JSON: false
+{
+    "name": "Hello World",
+    "nested": {
+        "x": 1
+    },
+    "size": 1,
+    "tags": [
+        "a",
+        "b"
+    ],
+    "valid": null
+}

docs/mkdocs/docs/features/parsing/error_recovery.md

@@ -0,0 +1,122 @@
+# Error Recovery
+
+By default, parsing stops at the first error. With the [SAX interface](sax_interface.md), you can instead ask the
+parser to *recover*: to repair the error and continue, so that you get as much as possible out of malformed input, for
+instance a file that was cut off, JSON edited by hand, or the output of a language model.
+
+## Recovering from errors
+
+The SAX parser's [`parse_error`](../../api/json_sax/parse_error.md) function is called for every error. Its return value
+decides what happens next:
+
+- `#!cpp false` stops parsing. This is what the SAX parsers of the library do, so [`parse`](../../api/basic_json/parse.md)
+  and [`accept`](../../api/basic_json/accept.md) never recover.
+- `#!cpp true` repairs the error and continues parsing.
+
+When recovering, the SAX parser still receives well-formed events: every `start_object` or `start_array` is followed by
+the matching `end_object` or `end_array`, and every `key` is followed by exactly one value. A SAX parser that creates a
+JSON value, such as the one in the example below, therefore gets a complete value. Parsing always ends, and
+[`sax_parse`](../../api/basic_json/sax_parse.md) returns `#!cpp false` for input that is not valid JSON, even if every
+error was repaired. Each token is reported at most once, and the SAX parser can stop at any error by returning
+`#!cpp false`.
+
+!!! example
+
+    The example below derives a SAX parser from the library's parser for `json` values (`json_sax_dom_parser`),
+    and recovers from all errors.
+
+    ```cpp
+    --8<-- "examples/sax_parse__error_recovery.cpp"
+    ```
+    
+    Output:
+    
+    ```
+    --8<-- "examples/sax_parse__error_recovery.output"
+    ```
+
+## How errors are repaired
+
+Each error is repaired with the smallest local edit: a missing separator is inserted, a stray token is removed, what can
+be read of a broken string or number is kept, and a value that cannot be read at all becomes `#!json null`.
+
+| Mistake                   | Repair                                                                         | Example                                  | Result                     |
+|---------------------------|--------------------------------------------------------------------------------|------------------------------------------|----------------------------|
+| missing `,` or `:`        | inserted                                                                       | `#!json [1 2]`, `#!json {"a" 1}`         | `[1,2]`, `{"a":1}`         |
+| missing value             | `#!json null` for an object key or between commas in an array                  | `#!json {"a":}`, `#!json [1,,2]`         | `{"a":null}`, `[1,null,2]` |
+| trailing comma            | removed                                                                        | `#!json [1,2,]`                          | `[1,2]`                    |
+| broken string             | invalid escapes and bytes are replaced (see below); a line break ends the string | `#!json ["a\qb"]`                        | `["aqb"]`                  |
+| broken number             | the longest valid beginning is kept                                            | `#!json [1., 2e+]`                       | `[1,2]`                    |
+| unreadable value          | `#!json null`                                                                  | `#!json [1, NaN, tru]`                   | `[1,null,null]`            |
+| number too large          | passed as infinity, together with its text                                     | `#!json [1e999]`                         | infinity (see below)       |
+| stray `:`                 | removed                                                                        | `#!json ["a":1]`                         | `["a",1]`                  |
+| member without a key      | skipped up to the next `,` or `}`                                              | `#!json {1:2, "b":3}`                    | `{"b":3}`                  |
+| wrong closing bracket     | closes the innermost array or object                                           | `#!json {"a":[1,2}, "b":3}`              | `{"a":[1,2],"b":3}`        |
+| input ends too early      | all open arrays and objects are closed                                         | `#!json {"a":[1,2`                       | `{"a":[1,2]}`              |
+| text before the value     | skipped                                                                        | `#!json )]}'{"a":1}`                     | `{"a":1}`                  |
+
+In a string, an unknown escape like `\q` stands for the escaped character (`q`), as in JavaScript. An invalid `\u`
+escape, a lone surrogate, and ill-formed UTF-8 are each replaced by U+FFFD (REPLACEMENT CHARACTER), and control
+characters are kept. A string without its closing quote ends at the next line break or at the end of the input.
+
+The input after the top-level value is not repaired: as without recovery, it is reported as an error, and parsing stops.
+
+## Binary formats
+
+The binary formats ([BJData](../binary_formats/bjdata.md), [BON8](../binary_formats/bon8.md),
+[BSON](../binary_formats/bson.md), [CBOR](../binary_formats/cbor.md), [MessagePack](../binary_formats/messagepack.md),
+and [UBJSON](../binary_formats/ubjson.md)) have no delimiters to find the next value by. So what can be repaired depends
+on whether the end of the item with the error is known, a distinction that
+[RFC 8949, Section 5.3](https://www.rfc-editor.org/rfc/rfc8949.html#section-5.3) makes for CBOR, too.
+
+If the item is complete, but cannot be passed on as it is, it is replaced, and parsing continues after it:
+
+| Mistake                                                             | Formats                 | Repair                                                                  |
+|---------------------------------------------------------------------|-------------------------|-------------------------------------------------------------------------|
+| tag, if `tag_handler` is `cbor_tag_handler_t::error` (the default)  | CBOR                    | ignored                                                                 |
+| simple value other than `false`, `true`, and `null`, like undefined | CBOR                    | `#!json null`                                                           |
+| number too large for a custom `number_float_t`, like `float`        | all                     | infinity                                                                |
+| character (`C`) that is not ASCII                                   | BJData, UBJSON          | U+FFFD                                                                  |
+| invalid high-precision number (`H`)                                 | BJData, UBJSON          | the longest valid beginning is kept, as for JSON text, or `#!json null` |
+| high-precision number too large                                     | BJData, UBJSON          | passed as infinity, together with its text                              |
+| object key that is not a string                                     | BON8, CBOR, MessagePack | the member is skipped                                                   |
+| element of a type the library does not read, like ObjectId or date  | BSON                    | `#!json null`                                                           |
+| string without its terminator                                       | BSON                    | kept                                                                    |
+| document whose size does not match its content                      | BSON                    | kept                                                                    |
+
+CBOR tags and simple values are repaired as [RFC 8949, Section 6.1](https://www.rfc-editor.org/rfc/rfc8949.html#section-6.1)
+suggests for converting CBOR to JSON. By default, [`sax_parse`](../../api/basic_json/sax_parse.md) reports every tag as
+an error; when recovering, tags are then ignored like with
+[`cbor_tag_handler_t::ignore`](../../api/basic_json/cbor_tag_handler_t.md). Strings that are not valid UTF-8 are no
+error: like [`from_cbor`](../../api/basic_json/from_cbor.md) and the other functions by default, `sax_parse` passes
+them on as they are.
+
+After any other error, the end of the item is unknown: the input ended, a byte is not a valid type marker, or a size
+cannot be right. Parsing then stops, and the value read so far is completed: a key that waits for its value gets
+`#!json null`, and all open arrays and objects are closed. This keeps everything before the error of an input that was
+cut off. The exception is BSON, which stores the size of every document: an element whose end is unknown gets
+`#!json null`, the rest of its document is skipped, and parsing continues after the document.
+
+## Limitations
+
+- A repair is a guess. For example, `#!json {"a" "b": 1}` could be meant as `#!json {"a": "b"}` or as
+  `#!json {"a": null, "b": 1}`; it is repaired to the former. Treat recovered values as a best effort, and check the
+  reported errors.
+- A closing bracket always closes the innermost array or object. If a bracket is missing rather than wrong, the
+  repair differs from the intention: `#!json {"a": {"b": [1, 2}, "c": 3}` is repaired to
+  `#!json {"a": {"b": [1, 2], "c": 3}}`, although `#!json {"a": {"b": [1, 2]}, "c": 3}` may have been meant.
+- Keys without quotes, and strings in single quotes, are not supported; such members are skipped.
+- In the binary formats, a member that is skipped because its key is not a string is lost, and so are the elements of a
+  BSON document after one whose end is unknown.
+- A number that is too large for `number_float_t` is passed as positive or negative infinity. The SAX parser's
+  `number_float` also gets the number's text, but a JSON value cannot store it, and
+  [`dump`](../../api/basic_json/dump.md) serializes infinity as `#!json null`.
+- When parsing is not strict (see [`sax_parse`](../../api/basic_json/sax_parse.md)), a repair may read parts of the
+  input after the value, for instance of the next value in a stream of concatenated values.
+
+## See also
+
+- [SAX interface](sax_interface.md) - implement a custom SAX handler
+- [`parse_error`](../../api/json_sax/parse_error.md) - the SAX event for parse errors
+- [`sax_parse`](../../api/basic_json/sax_parse.md) - generate SAX events
+- [parsing and exceptions](parse_exceptions.md) - control error handling

docs/mkdocs/docs/features/parsing/index.md

@@ -130,7 +130,7 @@ See [Zero-copy JSON views](../json_view.md) for the concept and the trade-offs.
 When the input is not valid JSON, the `parse` function throws an exception by default. If exceptions are undesired or
 unavailable, the parser can instead return a discarded value, or [`accept`](../../api/basic_json/accept.md) can be used
 to only check whether an input is valid JSON. See [parsing and exceptions](parse_exceptions.md) for the available
-options.
+options. To get as much as possible out of malformed input, a SAX parser can [recover from errors](error_recovery.md).
 
 ## See also
 
@@ -142,5 +142,6 @@ options.
 - [parser callbacks](parser_callbacks.md) - influence the parsing by a callback function
 - [SAX interface](sax_interface.md) - implement a custom SAX handler
 - [parsing and exceptions](parse_exceptions.md) - control error handling
+- [error recovery](error_recovery.md) - get as much as possible out of malformed input
 - [parsing untrusted input](untrusted_input.md) - what to consider when parsing input from untrusted sources
 - [Zero-copy JSON views](../json_view.md) - parse into a document and read it through a view

docs/mkdocs/docs/features/parsing/parse_exceptions.md

@@ -64,7 +64,8 @@ bool parse_error(std::size_t position,
                  const json::exception& ex);
 ```
 
-The return value indicates whether the parsing should continue, so the function should usually return `#!cpp false`.
+The return value decides whether to stop parsing (`#!cpp false`) or to repair the error and continue
+(`#!cpp true`); see [error recovery](error_recovery.md) for the latter.
 
 ??? example "Example: report parse errors without exceptions"
 

docs/mkdocs/docs/features/parsing/sax_interface.md

@@ -60,7 +60,8 @@ bool key(string_t& val);
 bool parse_error(std::size_t position, const std::string& last_token, const json::exception& ex);
 ```
 
-The return value of each function determines whether parsing should proceed.
+The return value of each function determines whether parsing should proceed. For `parse_error`, returning
+`#!cpp true` [recovers from the error](error_recovery.md).
 
 To implement your own SAX handler, proceed as follows:
 
@@ -68,7 +69,7 @@ To implement your own SAX handler, proceed as follows:
 2. Create an object of your SAX interface class, e.g. `my_sax`.
 3. Call `#!cpp bool json::sax_parse(input, &my_sax);` where the first parameter can be any input like a string or an input stream and the second parameter is a pointer to your SAX interface.
 
-Note the `sax_parse` function only returns a `#!cpp bool` indicating the result of the last executed SAX event. It does not return `json` value - it is up to you to decide what to do with the SAX events. Furthermore, no exceptions are thrown in case of a parse error - it is up to you what to do with the exception object passed to your `parse_error` implementation. Internally, the SAX interface is used for the DOM parser (class `json_sax_dom_parser`) as well as the acceptor (`json_sax_acceptor`), see file `json_sax.hpp`.
+Note the `sax_parse` function only returns a `#!cpp bool` indicating whether the input was parsed without errors and no SAX event returned `#!cpp false`. It does not return `json` value - it is up to you to decide what to do with the SAX events. Furthermore, no exceptions are thrown in case of a parse error - it is up to you what to do with the exception object passed to your `parse_error` implementation. Internally, the SAX interface is used for the DOM parser (class `json_sax_dom_parser`) as well as the acceptor (`json_sax_acceptor`), see file `json_sax.hpp`.
 
 ## See also
 

docs/mkdocs/mkdocs.yml

@@ -134,6 +134,7 @@ nav:
       - features/performance.md
       - Parsing:
           - features/parsing/index.md
+          - features/parsing/error_recovery.md
           - features/parsing/json_lines.md
           - features/parsing/parse_exceptions.md
           - features/parsing/parser_callbacks.md

include/nlohmann/detail/input/input_adapters.hpp

@@ -796,17 +796,22 @@ class iterator_input_adapter
         return get_consumed_count() + eof_reads;
     }
 
-    // number of newlines in the consumed range [0, last_index); @a line_start
-    // is set to the index after the last of them (0 if there is none); only
-    // valid when supports_seek
-    std::size_t count_consumed_lines(std::size_t last_index, std::size_t& line_start) const
+    // number of newlines in the consumed range [first_index, last_index); if
+    // there is one, @a line_start is set to the index after the last of them,
+    // otherwise it is left unchanged; only valid when supports_seek
+    std::size_t count_consumed_lines(std::size_t first_index, std::size_t last_index, std::size_t& line_start) const
     {
-        const auto to = std::next(begin, static_cast<typename std::iterator_traits<IteratorType>::difference_type>(last_index));
+        using difference_type = typename std::iterator_traits<IteratorType>::difference_type;
+        const auto from = std::next(begin, static_cast<difference_type>(first_index));
+        const auto to = std::next(from, static_cast<difference_type>(last_index - first_index));
         const auto newline = char_traits<char_type>::to_char_type('\n');
         using reverse_it = std::reverse_iterator<IteratorType>;
-        const auto last_newline = std::find(reverse_it(to), reverse_it(begin), newline);
-        line_start = static_cast<std::size_t>(std::distance(begin, last_newline.base()));
-        return static_cast<std::size_t>(std::count(begin, last_newline.base(), newline));
+        const auto last_newline = std::find(reverse_it(to), reverse_it(from), newline);
+        if (last_newline.base() != from)
+        {
+            line_start = static_cast<std::size_t>(std::distance(begin, last_newline.base()));
+        }
+        return static_cast<std::size_t>(std::count(from, last_newline.base(), newline));
     }
 
     // append the already-consumed characters in the half-open range

include/nlohmann/detail/input/json_sax.hpp

@@ -138,7 +138,9 @@ struct json_sax
     @param[in] position    the position in the input where the error occurs
     @param[in] last_token  the last read token
     @param[in] ex          an exception object describing the error
-    @return whether parsing should proceed (must return false)
+    @return whether to recover from the error: false stops parsing; true
+            repairs the error and continues, or, if that is not possible,
+            stops after completing the value read so far
     */
     virtual bool parse_error(std::size_t position,
                              const std::string& last_token,
@@ -469,12 +471,15 @@ constructor contains the parsed value. Without @a Collect, every value is
 stored in place as it arrives, so after a parse error the value holds what was
 read up to it.
 
-@tparam BasicJsonType  the JSON type
-@tparam Collect  whether to collect the values of containers of unknown size on
-                 a stack (only for a caller that does not use the value after a
-                 parse error)
+@tparam BasicJsonType     the JSON type
+@tparam InputAdapterType  the input adapter of the lexer that can be passed to
+                          the constructor to record diagnostic positions; it
+                          does not matter if no lexer is passed
+@tparam Collect           whether to collect the values of containers of
+                          unknown size on a stack (only for a caller that does
+                          not use the value after a parse error)
 */
-template<typename BasicJsonType, typename InputAdapterType, bool Collect = false>
+template<typename BasicJsonType, typename InputAdapterType = string_input_adapter_type, bool Collect = false>
 class json_sax_dom_parser : public json_sax_dom_error_state
 {
   public:
@@ -772,6 +777,8 @@ class json_sax_dom_parser : public json_sax_dom_error_state
         JSON_HEDLEY_NEVER_INLINE
         void grow()
         {
+            // the stack starts with local_capacity frames and only grows
+            JSON_ASSERT(capacity != 0);
             const std::size_t new_capacity = capacity * 2;
             std::unique_ptr<frame[]> new_heap(new frame[new_capacity]); // NOLINT(cppcoreguidelines-avoid-c-arrays,hicpp-avoid-c-arrays,modernize-avoid-c-arrays)
             std::copy(data, data + count, new_heap.get());
@@ -1581,7 +1588,7 @@ class json_sax_dom_binary_parser : public json_sax_dom_error_state
     lexer_t* m_lexer_ref = nullptr;
 };
 
-template<typename BasicJsonType, typename InputAdapterType>
+template<typename BasicJsonType, typename InputAdapterType = string_input_adapter_type>
 class json_sax_dom_callback_parser : public json_sax_dom_error_state
 {
   public:

include/nlohmann/detail/input/lexer.hpp

@@ -10,7 +10,7 @@
 
 #include <array> // array
 #include <cstddef> // size_t
-#include <cstdint> // uint32_t
+#include <cstdint> // uint8_t, uint32_t
 #include <cstdio> // snprintf
 #include <cstring> // memcmp
 #include <initializer_list> // initializer_list
@@ -381,6 +381,20 @@ class lexer : public lexer_base<BasicJsonType>
         return codepoint;
     }
 
+    /// the code point of the surrogate pair @a high (U+D800..U+DBFF) and @a low (U+DC00..U+DFFF)
+    static constexpr int combine_surrogates(const int high, const int low) noexcept
+    {
+        return static_cast<int>(
+                   // high surrogate occupies the most significant 22 bits
+                   (static_cast<unsigned int>(high) << 10u)
+                   // low surrogate occupies the least significant 15 bits
+                   + static_cast<unsigned int>(low)
+                   // there is still the 0xD800, 0xDC00, and 0x10000 noise
+                   // in the result, so we have to subtract with:
+                   // (0xD800 << 10) + DC00 - 0x10000 = 0x35FDC00
+                   - 0x35FDC00u);
+    }
+
     /*!
     @brief check if the next byte(s) are inside a given range
 
@@ -411,6 +425,7 @@ class lexer : public lexer_base<BasicJsonType>
             else
             {
                 error_message = "invalid string: ill-formed UTF-8 byte";
+                error_cause = error_kind::string_ill_formed_utf8;
                 return false;
             }
         }
@@ -526,6 +541,7 @@ class lexer : public lexer_base<BasicJsonType>
                 case char_traits<char_type>::eof():
                 {
                     error_message = "invalid string: missing closing quote";
+                    error_cause = error_kind::string;
                     return token_type::parse_error;
                 }
 
@@ -582,6 +598,7 @@ class lexer : public lexer_base<BasicJsonType>
                             if (JSON_HEDLEY_UNLIKELY(codepoint1 == -1))
                             {
                                 error_message = "invalid string: '\\u' must be followed by 4 hex digits";
+                                error_cause = error_kind::string_invalid_escape;
                                 return token_type::parse_error;
                             }
 
@@ -589,13 +606,23 @@ class lexer : public lexer_base<BasicJsonType>
                             if (0xD800 <= codepoint1 && codepoint1 <= 0xDBFF)
                             {
                                 // expect next \uxxxx entry
-                                if (JSON_HEDLEY_LIKELY(get() == '\\' && get() == 'u'))
+                                if (JSON_HEDLEY_LIKELY(get() == '\\'))
                                 {
+                                    if (JSON_HEDLEY_UNLIKELY(get() != 'u'))
+                                    {
+                                        // current is the character escaped by the backslash
+                                        error_message = "invalid string: surrogate U+D800..U+DBFF must be followed by U+DC00..U+DFFF";
+                                        error_cause = error_kind::string_invalid_escape;
+                                        string_error_resume = resume_kind::escaped_character;
+                                        return token_type::parse_error;
+                                    }
+
                                     const int codepoint2 = get_codepoint();
 
                                     if (JSON_HEDLEY_UNLIKELY(codepoint2 == -1))
                                     {
                                         error_message = "invalid string: '\\u' must be followed by 4 hex digits";
+                                        error_cause = error_kind::string_invalid_escape;
                                         return token_type::parse_error;
                                     }
 
@@ -603,25 +630,23 @@ class lexer : public lexer_base<BasicJsonType>
                                     if (JSON_HEDLEY_LIKELY(0xDC00 <= codepoint2 && codepoint2 <= 0xDFFF))
                                     {
                                         // overwrite codepoint
-                                        codepoint = static_cast<int>(
-                                                        // high surrogate occupies the most significant 22 bits
-                                                        (static_cast<unsigned int>(codepoint1) << 10u)
-                                                        // low surrogate occupies the least significant 15 bits
-                                                        + static_cast<unsigned int>(codepoint2)
-                                                        // there is still the 0xD800, 0xDC00, and 0x10000 noise
-                                                        // in the result, so we have to subtract with:
-                                                        // (0xD800 << 10) + DC00 - 0x10000 = 0x35FDC00
-                                                        - 0x35FDC00u);
+                                        codepoint = combine_surrogates(codepoint1, codepoint2);
                                     }
                                     else
                                     {
+                                        // the second escape was read completely and is a
+                                        // code point of its own
                                         error_message = "invalid string: surrogate U+D800..U+DBFF must be followed by U+DC00..U+DFFF";
+                                        error_cause = error_kind::string_invalid_escape;
+                                        string_error_resume = resume_kind::after_escape;
+                                        string_error_codepoint = codepoint2;
                                         return token_type::parse_error;
                                     }
                                 }
                                 else
                                 {
                                     error_message = "invalid string: surrogate U+D800..U+DBFF must be followed by U+DC00..U+DFFF";
+                                    error_cause = error_kind::string_invalid_escape;
                                     return token_type::parse_error;
                                 }
                             }
@@ -629,7 +654,10 @@ class lexer : public lexer_base<BasicJsonType>
                             {
                                 if (JSON_HEDLEY_UNLIKELY(0xDC00 <= codepoint1 && codepoint1 <= 0xDFFF))
                                 {
+                                    // the escape was read completely
                                     error_message = "invalid string: surrogate U+DC00..U+DFFF must follow U+D800..U+DBFF";
+                                    error_cause = error_kind::string_invalid_escape;
+                                    string_error_resume = resume_kind::after_escape;
                                     return token_type::parse_error;
                                 }
                             }
@@ -638,10 +666,7 @@ class lexer : public lexer_base<BasicJsonType>
                             JSON_ASSERT(0x00 <= codepoint && codepoint <= 0x10FFFF);
 
                             // translate codepoint into bytes
-                            encode_utf8(static_cast<std::uint32_t>(codepoint), [this](std::uint32_t byte)
-                            {
-                                add(static_cast<char_int_type>(byte));
-                            });
+                            add_codepoint(codepoint);
 
                             break;
                         }
@@ -649,6 +674,7 @@ class lexer : public lexer_base<BasicJsonType>
                         // other characters after escape
                         default:
                             error_message = "invalid string: forbidden character after backslash";
+                            error_cause = error_kind::string;
                             return token_type::parse_error;
                     }
 
@@ -725,6 +751,7 @@ class lexer : public lexer_base<BasicJsonType>
                         }
                     };
                     error_message = control_character_messages[static_cast<std::size_t>(current)];
+                    error_cause = error_kind::string;
                     return token_type::parse_error;
                 }
 
@@ -947,6 +974,7 @@ class lexer : public lexer_base<BasicJsonType>
                 default:
                 {
                     error_message = "invalid string: ill-formed UTF-8 byte";
+                    error_cause = error_kind::string_ill_formed_utf8;
                     return token_type::parse_error;
                 }
             }
@@ -1002,6 +1030,7 @@ class lexer : public lexer_base<BasicJsonType>
                         case char_traits<char_type>::eof():
                         {
                             error_message = "invalid comment; missing closing '*/'";
+                            error_cause = error_kind::unclosed_comment;
                             return false;
                         }
 
@@ -1032,6 +1061,7 @@ class lexer : public lexer_base<BasicJsonType>
             default:
             {
                 error_message = "invalid comment; expecting '/' or '*' after '/'";
+                error_cause = error_kind::other;
                 return false;
             }
         }
@@ -1157,6 +1187,7 @@ scan_number_minus:
             default:
             {
                 error_message = "invalid number; expected digit after '-'";
+                error_cause = error_kind::number;
                 return token_type::parse_error;
             }
         }
@@ -1243,6 +1274,7 @@ scan_number_decimal1:
             default:
             {
                 error_message = "invalid number; expected digit after '.'";
+                error_cause = error_kind::number;
                 return token_type::parse_error;
             }
         }
@@ -1311,6 +1343,7 @@ scan_number_exponent:
             {
                 error_message =
                     "invalid number; expected '+', '-', or digit after exponent";
+                error_cause = error_kind::number;
                 return token_type::parse_error;
             }
         }
@@ -1337,6 +1370,7 @@ scan_number_sign:
             default:
             {
                 error_message = "invalid number; expected digit after exponent sign";
+                error_cause = error_kind::number;
                 return token_type::parse_error;
             }
         }
@@ -1764,6 +1798,7 @@ scan_number_done:
             if (JSON_HEDLEY_UNLIKELY(char_traits<char_type>::to_char_type(get()) != static_cast<char_type>(rest[i])))
             {
                 error_message = "invalid literal";
+                error_cause = error_kind::other;
                 return token_type::parse_error;
             }
         }
@@ -2154,17 +2189,25 @@ scan_number_done:
         return position.chars_read_total;
     }
 
-    /// seekable adapter: count the lines in the consumed input; O(input read
-    /// so far), but only called on errors
+    /// seekable adapter: count the lines in the consumed input. Only the
+    /// input consumed since the previous call is scanned, so a parse that
+    /// reports many errors (a SAX parser recovering from them) stays linear
+    /// in the input instead of rescanning it for every error.
     position_t get_position_impl(std::true_type /*lazy*/) const
     {
         position_t result;
         result.chars_read_total = get_chars_read_total();
         // reads at EOF count as characters but are not part of the input
         const std::size_t consumed = (std::min)(result.chars_read_total, ia.get_consumed_count());
-        std::size_t line_start = 0;
-        result.lines_read = ia.count_consumed_lines(consumed, line_start);
-        result.chars_read_current_line = result.chars_read_total - line_start;
+        if (JSON_HEDLEY_UNLIKELY(consumed < position_cache.consumed))
+        {
+            // the consumed input never shrinks; recount from the start if it did
+            position_cache = {};
+        }
+        position_cache.lines += ia.count_consumed_lines(position_cache.consumed, consumed, position_cache.line_start);
+        position_cache.consumed = consumed;
+        result.lines_read = position_cache.lines;
+        result.chars_read_current_line = result.chars_read_total - position_cache.line_start;
         return result;
     }
 
@@ -2348,6 +2391,7 @@ scan_number_done:
         if (get_chars_read_total() == 0 && !skip_bom())
         {
             error_message = "invalid BOM; must be 0xEF 0xBB 0xBF if given";
+            error_cause = error_kind::other;
             return token_type::parse_error;
         }
 
@@ -2455,10 +2499,583 @@ scan_number_done:
             // error
             default:
                 error_message = "invalid literal";
+                error_cause = error_kind::other;
                 return token_type::parse_error;
         }
     }
 
+  public:
+    /////////////////////
+    // error recovery
+    /////////////////////
+
+    /*!
+    @brief make the best of the token that scan() rejected
+
+    Called by the parser after scan() returned token_type::parse_error and the
+    SAX parser asked to recover from the error (see #3989). Keeps what can be
+    read of the token and skips the rest:
+
+    - A string keeps its characters. An unknown escape stands for the escaped
+      character itself (as in JavaScript), an invalid Unicode escape and ill-formed
+      UTF-8 become U+FFFD, and a control character is kept. A line break or the
+      end of the input ends a string that lacks its closing quote.
+    - A number keeps its longest valid prefix, e.g. `1` for `1.` or `1e+`.
+    - A block comment that is not closed runs to the end of the input.
+    - Anything else is skipped.
+
+    The rest of an invalid token is skipped up to the next delimiter
+    (whitespace, a structural character, or a quote). A delimiter that the
+    invalid token consumed is returned to the input, so that the next scan()
+    reads it.
+
+    @return token_type::value_string or a number token type if a string or a
+            number could be read, token_type::end_of_input for a block comment
+            that is not closed, token_type::uninitialized otherwise
+    */
+    token_type recover_token()
+    {
+        const resume_kind resume = string_error_resume;
+        const int codepoint = string_error_codepoint;
+        string_error_resume = resume_kind::character;
+        string_error_codepoint = -1;
+
+        const error_kind cause = error_cause;
+        error_cause = error_kind::other;
+
+        switch (cause)
+        {
+            case error_kind::string:
+            case error_kind::string_invalid_escape:
+            case error_kind::string_ill_formed_utf8:
+                return recover_string(cause, resume, codepoint);
+
+            case error_kind::number:
+                return recover_number();
+
+            case error_kind::unclosed_comment:
+                // the comment runs to the end of the input
+                return token_type::end_of_input;
+
+            case error_kind::other:
+            default:
+                skip_to_delimiter();
+                return token_type::uninitialized;
+        }
+    }
+
+    /*!
+    @brief return the token that scan() read last to the input, so that the
+           next scan() reads it again
+
+    Called by the parser when recovering from an error. The token must be a
+    single character (',', ':', '[', ']', '{', or '}') or the end of the
+    input, and scan() must have read it last.
+    */
+    void unget_token()
+    {
+        JSON_ASSERT(!next_unget);
+        unget();
+    }
+
+    /*!
+    @brief let the token string for the next error begin at the current character
+
+    The token string of an error reaches back to the beginning of the last
+    string or number. After an error, the parser calls this function so that
+    the next error does not report (and, with many errors, copy) everything
+    read since then.
+    */
+    void restart_token_string()
+    {
+        restart_token_string_impl(std::integral_constant<bool, lazy_token_string> {});
+    }
+
+  private:
+    /// how recover_string() continues after the error scan_string() reported
+    enum class resume_kind : std::uint8_t
+    {
+        /// current is the next character of the string (or the end of input)
+        character,
+        /// current is the character escaped by the preceding backslash
+        escaped_character,
+        /// current is the last character of a complete escape
+        after_escape
+    };
+
+    /// what kind of token scan() rejected, which decides how recover_token() repairs it
+    enum class error_kind : std::uint8_t
+    {
+        /// anything not listed below; skipped up to the next delimiter
+        other,
+        /// a string
+        string,
+        /// a string with an invalid Unicode escape, which becomes U+FFFD
+        string_invalid_escape,
+        /// a string with ill-formed UTF-8, which becomes U+FFFD
+        string_ill_formed_utf8,
+        /// a number
+        number,
+        /// a block comment that is not closed
+        unclosed_comment
+    };
+
+    /// whether current ends an invalid token (see recover_token())
+    bool current_is_delimiter() const noexcept
+    {
+        switch (current)
+        {
+            case ' ':
+            case '\t':
+            case '\n':
+            case '\r':
+            case '[':
+            case ']':
+            case '{':
+            case '}':
+            case ',':
+            case ':':
+            case '\"':
+#if !JSON_STRICT_NUL_HANDLING
+            case '\0':
+#endif
+            case char_traits<char_type>::eof():
+                return true;
+
+            case '/':
+                return ignore_comments;
+
+            default:
+                return false;
+        }
+    }
+
+    /// skip the rest of an invalid token and return its delimiter to the input
+    void skip_to_delimiter()
+    {
+        while (!current_is_delimiter())
+        {
+            get();
+        }
+
+        if (current != char_traits<char_type>::eof())
+        {
+            unget();
+        }
+    }
+
+    /// append U+FFFD REPLACEMENT CHARACTER to token_buffer
+    void add_replacement_character()
+    {
+        add_codepoint(0xFFFD);
+    }
+
+    /// append the UTF-8 encoding of @a codepoint (not a surrogate) to token_buffer
+    void add_codepoint(const int codepoint)
+    {
+        JSON_ASSERT(0x00 <= codepoint && codepoint <= 0x10FFFF);
+        encode_utf8(static_cast<std::uint32_t>(codepoint), [this](std::uint32_t byte)
+        {
+            add(static_cast<char_int_type>(byte));
+        });
+    }
+
+    /// append a code point read from a Unicode escape; a surrogate becomes U+FFFD
+    void add_escaped_codepoint(const int codepoint)
+    {
+        if (0xD800 <= codepoint && codepoint <= 0xDFFF)
+        {
+            add_replacement_character();
+        }
+        else
+        {
+            add_codepoint(codepoint);
+        }
+    }
+
+    /*!
+    @brief remove an incomplete UTF-8 sequence from the end of token_buffer
+
+    next_byte_in_range() adds the bytes of a sequence as it checks them, so
+    when it rejects a byte, the beginning of the sequence is already in
+    token_buffer, which otherwise holds only complete sequences.
+
+    @return whether an incomplete sequence was removed
+    */
+    bool remove_incomplete_utf8_sequence()
+    {
+        std::size_t lead = token_buffer.size();
+        std::size_t continuation_bytes = 0;
+        while (lead > 0 && continuation_bytes < 3
+                && (static_cast<unsigned char>(token_buffer[lead - 1]) & 0xC0u) == 0x80u)
+        {
+            --lead;
+            ++continuation_bytes;
+        }
+        if (lead == 0)
+        {
+            return false;
+        }
+
+        const auto lead_byte = static_cast<unsigned char>(token_buffer[lead - 1]);
+        std::size_t expected = 0;
+        if (lead_byte >= 0xF0)
+        {
+            expected = 3;
+        }
+        else if (lead_byte >= 0xE0)
+        {
+            expected = 2;
+        }
+        else if (lead_byte >= 0xC0)
+        {
+            expected = 1;
+        }
+        if (continuation_bytes >= expected)
+        {
+            return false;
+        }
+
+        token_buffer.resize(lead - 1);
+        return true;
+    }
+
+    /*!
+    @brief read the UTF-8 sequence that begins with current, which is not ASCII
+    @return whether the next character must be read; false if current still
+            needs to be handled, because it does not belong to the sequence
+    */
+    bool recover_utf8_sequence()
+    {
+        // the number of continuation bytes and the range of the first one;
+        // see the ranges in scan_string()
+        std::size_t count = 0;
+        char_int_type low = 0x80;
+        char_int_type high = 0xBF;
+        if (current >= 0xC2 && current <= 0xDF)
+        {
+            count = 1;
+        }
+        else if (current >= 0xE0 && current <= 0xEF)
+        {
+            count = 2;
+            low = (current == 0xE0) ? 0xA0 : 0x80;
+            high = (current == 0xED) ? 0x9F : 0xBF;
+        }
+        else if (current >= 0xF0 && current <= 0xF4)
+        {
+            count = 3;
+            low = (current == 0xF0) ? 0x90 : 0x80;
+            high = (current == 0xF4) ? 0x8F : 0xBF;
+        }
+        else
+        {
+            // an ill-formed byte
+            add_replacement_character();
+            return true;
+        }
+
+        const std::size_t start = token_buffer.size();
+        add(current);
+        for (std::size_t i = 0; i < count; ++i)
+        {
+            get();
+            if (current < low || current > high)
+            {
+                token_buffer.resize(start);
+                add_replacement_character();
+                return false;
+            }
+            add(current);
+            low = 0x80;
+            high = 0xBF;
+        }
+        return true;
+    }
+
+    /*!
+    @brief read the low surrogate that must follow the high surrogate @a high
+    @return whether the next character must be read; false if current still
+            needs to be handled
+    */
+    bool recover_low_surrogate(int high)
+    {
+        while (true)
+        {
+            if (get() != '\\')
+            {
+                add_replacement_character();
+                return false;
+            }
+            if (get() != 'u')
+            {
+                add_replacement_character();
+                // not 'u', so this does not come back here
+                return recover_escape();
+            }
+
+            const int low = get_codepoint();
+            if (low == -1)
+            {
+                add_replacement_character();
+                return false;
+            }
+            if (0xDC00 <= low && low <= 0xDFFF)
+            {
+                add_codepoint(combine_surrogates(high, low));
+                return true;
+            }
+
+            // high has no low surrogate
+            add_replacement_character();
+            if (low < 0xD800 || low > 0xDBFF)
+            {
+                add_codepoint(low);
+                return true;
+            }
+            // another high surrogate
+            high = low;
+        }
+    }
+
+    /*!
+    @brief read the escape whose backslash was read; current is the escaped character
+    @return whether the next character must be read; false if current still
+            needs to be handled
+    */
+    bool recover_escape()
+    {
+        switch (current)
+        {
+            case '\"':
+                add('\"');
+                return true;
+            case '\\':
+                add('\\');
+                return true;
+            case '/':
+                add('/');
+                return true;
+            case 'b':
+                add('\b');
+                return true;
+            case 'f':
+                add('\f');
+                return true;
+            case 'n':
+                add('\n');
+                return true;
+            case 'r':
+                add('\r');
+                return true;
+            case 't':
+                add('\t');
+                return true;
+
+            case 'u':
+            {
+                const int codepoint = get_codepoint();
+                if (codepoint == -1)
+                {
+                    add_replacement_character();
+                    return false;
+                }
+                if (0xD800 <= codepoint && codepoint <= 0xDBFF)
+                {
+                    return recover_low_surrogate(codepoint);
+                }
+                add_escaped_codepoint(codepoint);
+                return true;
+            }
+
+            // an unknown escape stands for the escaped character
+            default:
+                return false;
+        }
+    }
+
+    /*!
+    @brief read the rest of a string after scan_string() rejected it
+
+    token_buffer holds what scan_string() read before the error. See
+    recover_token() for how errors are repaired.
+
+    @param[in] cause      the kind of error scan_string() reported
+    @param[in] resume     how to continue, see resume_kind
+    @param[in] codepoint  for a high surrogate followed by an escape of another
+                          code point: that code point; -1 otherwise
+    */
+    token_type recover_string(const error_kind cause, const resume_kind resume, const int codepoint)
+    {
+        // whether the next character must be read before it can be handled
+        bool fetch = false;
+
+        if (cause == error_kind::string_invalid_escape
+                || (cause == error_kind::string_ill_formed_utf8 && remove_incomplete_utf8_sequence()))
+        {
+            add_replacement_character();
+        }
+
+        switch (resume)
+        {
+            case resume_kind::escaped_character:
+                fetch = recover_escape();
+                break;
+            case resume_kind::after_escape:
+                if (0xD800 <= codepoint && codepoint <= 0xDBFF)
+                {
+                    fetch = recover_low_surrogate(codepoint);
+                }
+                else
+                {
+                    if (codepoint != -1)
+                    {
+                        add_escaped_codepoint(codepoint);
+                    }
+                    fetch = true;
+                }
+                break;
+            case resume_kind::character:
+            default:
+                break;
+        }
+
+        while (true)
+        {
+            if (fetch)
+            {
+                get();
+            }
+            fetch = true;
+
+            switch (current)
+            {
+                case '\"':
+                // a line break or the end of the input ends a string that
+                // lacks its closing quote
+                case '\n':
+                case '\r':
+                case char_traits<char_type>::eof():
+                    return token_type::value_string;
+
+#if !JSON_STRICT_NUL_HANDLING
+                case '\0':
+                    // the end of the input, see scan()
+                    unget();
+                    return token_type::value_string;
+#endif
+
+                case '\\':
+                    get();
+                    fetch = recover_escape();
+                    break;
+
+                default:
+                    if (current < 0x80)
+                    {
+                        // including control characters
+                        add(current);
+                    }
+                    else
+                    {
+                        fetch = recover_utf8_sequence();
+                    }
+                    break;
+            }
+        }
+    }
+
+    /*!
+    @brief keep the longest valid prefix of a number that scan_number() rejected
+
+    token_buffer holds the characters scan_number() accepted before the error,
+    so the prefix ends at its last digit.
+    */
+    token_type recover_number()
+    {
+        // only size(), operator[], and resize() are used, which every string
+        // type the library supports provides
+        std::size_t length = token_buffer.size();
+        while (length != 0 && (token_buffer[length - 1] < '0' || token_buffer[length - 1] > '9'))
+        {
+            --length;
+        }
+        token_buffer.resize(length);
+
+        if (length == 0)
+        {
+            skip_to_delimiter();
+            return token_type::uninitialized;
+        }
+
+        if (decimal_point_position >= length)
+        {
+            decimal_point_position = std::string::npos;
+        }
+
+        std::size_t exponent = std::string::npos;
+        for (std::size_t i = 0; i < length; ++i)
+        {
+            if (token_buffer[i] == 'e' || token_buffer[i] == 'E')
+            {
+                exponent = i;
+                break;
+            }
+        }
+        const std::size_t mantissa_end = (exponent == std::string::npos) ? length : exponent;
+        token_type number_type = token_type::value_unsigned;
+        if (decimal_point_position != std::string::npos || exponent != std::string::npos)
+        {
+            number_type = token_type::value_float;
+        }
+        else if (token_buffer[0] == '-')
+        {
+            number_type = token_type::value_integer;
+        }
+
+        const token_type result = convert_number(number_type, mantissa_end);
+        skip_to_delimiter();
+        return result;
+    }
+
+    /// seekable or token-tracking adapter: the token string begins at current
+    void restart_token_string_impl(std::true_type /*lazy*/)
+    {
+        restart_token_string_lazy(std::integral_constant<bool, adapter_token> {});
+    }
+
+    /// seekable adapter: the token string begins at current, which was consumed
+    void restart_token_string_lazy(std::false_type /*adapter_token*/) noexcept
+    {
+        const std::size_t consumed = ia.get_consumed_count();
+        token_string_start = (consumed > 0 && current != char_traits<char_type>::eof()) ? consumed - 1 : consumed;
+    }
+
+    /// token-tracking adapter: the token begins with current, or with the next
+    /// character if current was put back or is the end of input
+    void restart_token_string_lazy(std::true_type /*adapter_token*/)
+    {
+        if (!next_unget && current != char_traits<char_type>::eof())
+        {
+            ia.mark_token_start(current);
+        }
+        else
+        {
+            ia.begin_token();
+        }
+    }
+
+    /// streaming adapter: the token string begins at current; a character
+    /// that was put back is copied again when it is read again
+    void restart_token_string_impl(std::false_type /*lazy*/)
+    {
+        token_string.clear();
+        if (!next_unget && current != char_traits<char_type>::eof())
+        {
+            token_string.push_back(char_traits<char_type>::to_char_type(current));
+        }
+    }
+
     /// input adapter
     InputAdapterType ia;
 
@@ -2483,6 +3100,19 @@ scan_number_done:
     /// that unget() can restore the column instead of leaving it at 0
     std::size_t chars_read_before_newline = 0;
 
+    /// seekable adapter: what get_position() counted last, so that the next
+    /// call only scans the input consumed since
+    struct position_cache_t
+    {
+        /// consumed characters scanned so far
+        std::size_t consumed = 0;
+        /// newlines among them
+        std::size_t lines = 0;
+        /// index after the last of these newlines
+        std::size_t line_start = 0;
+    };
+    mutable position_cache_t position_cache {};
+
     /// raw input token string for error messages; only populated for streaming
     /// adapters (seekable adapters reconstruct it lazily via token_string_start)
     std::vector<char_type> token_string {};
@@ -2503,6 +3133,15 @@ scan_number_done:
     /// a description of occurred lexer errors
     const char* error_message = "";
 
+    /// what kind of token scan() rejected; set together with error_message
+    error_kind error_cause = error_kind::other;
+    /// how recover_token() continues a string that scan_string() rejected;
+    /// set only on the error paths that need more than error_message
+    resume_kind string_error_resume = resume_kind::character;
+    /// the code point of the second escape when a high surrogate is followed
+    /// by an escape that is not a low surrogate; -1 otherwise
+    int string_error_codepoint = -1;
+
     // number values
     number_integer_t value_integer = 0;
     number_unsigned_t value_unsigned = 0;

include/nlohmann/detail/input/parser.hpp

@@ -153,25 +153,72 @@ class parser
     bool accept(const bool strict = true)
     {
         json_sax_acceptor<BasicJsonType> sax_acceptor;
-        return sax_parse(&sax_acceptor, strict);
+        return sax_parse_impl<false>(&sax_acceptor, strict);
     }
 
+    /*!
+    @brief public SAX interface
+
+    If the SAX parser's parse_error() returns true, the parser recovers from
+    the error: it repairs the input and continues (see #3989).
+
+    @param[in] sax     the SAX parser
+    @param[in] strict  whether to expect the last token to be EOF
+    @return whether the input was parsed without errors and no SAX event
+            returned false
+    */
     template<typename SAX>
     JSON_HEDLEY_NON_NULL(2)
     bool sax_parse(SAX* sax, const bool strict = true)
     {
-        (void)detail::is_sax_static_asserts<SAX, BasicJsonType> {};
-        const bool result = sax_parse_internal(sax);
-
-        if (result)
-        {
-            return finish_input(*sax, strict);
-        }
-
-        return false;
+        return sax_parse_impl<true>(sax, strict);
     }
 
   private:
+    /// what sax_parse_internal() does after an object key was expected
+    enum class next_step : std::uint8_t
+    {
+        /// stop parsing
+        stop,
+        /// parse a value that begins with last_token
+        parse_value,
+        /// evaluate the state of the innermost container, which reads
+        /// last_token again
+        evaluate_state
+    };
+
+    template<bool AllowRecovery, typename SAX>
+    JSON_HEDLEY_NON_NULL(2)
+    bool sax_parse_impl(SAX* sax, const bool strict)
+    {
+        (void)detail::is_sax_static_asserts<SAX, BasicJsonType> {};
+        const bool result = sax_parse_internal<AllowRecovery>(sax);
+
+        if (result)
+        {
+            if (strict)
+            {
+                // strict mode: next byte must be EOF; after recovering from an
+                // error, the end of the input may already have been read
+                if (last_token != token_type::end_of_input && get_token() != token_type::end_of_input)
+                {
+                    // the value is complete, so there is nothing to recover
+                    static_cast<void>(report_error(sax, parse_error::create(exception_id::syntax_error, m_lexer.get_position(), exception_message(token_type::end_of_input, "value"), nullptr),
+                                                   std::integral_constant<bool, AllowRecovery> {}));
+                    return false;
+                }
+            }
+            else
+            {
+                // the caller keeps using the input: position it right after
+                // the value by leaving the character that terminated it
+                m_lexer.release_lookahead();
+            }
+        }
+
+        return result && !error_reported;
+    }
+
     /// the text of the current float token for @a sax (see sax_ignores_number_text)
     template<typename SAX>
     const string_t& number_float_text(SAX* /*sax*/)
@@ -239,17 +286,26 @@ class parser
     template<typename DomSax>
     bool parse_dom(DomSax& sdp, const bool strict)
     {
-        sax_parse_internal(&sdp);
+        sax_parse_internal<false>(&sdp);
 
         static_cast<void>(finish_input(sdp, strict));
 
         return !sdp.is_errored();
     }
 
-    template<typename SAX>
+    /*!
+    @brief parse a JSON value and pass it to a SAX parser
+
+    @tparam AllowRecovery  whether to recover from an error if the SAX parser's
+                           parse_error() returns true; false for the SAX parsers
+                           of parse() and accept(), which never do, so that no
+                           code for recovering is generated for them
+    */
+    template<bool AllowRecovery, typename SAX>
     JSON_HEDLEY_NON_NULL(2)
     bool sax_parse_internal(SAX* sax)
     {
+        const std::integral_constant<bool, AllowRecovery> allow_recovery{};
         // value to avoid a goto (see comment where set to true)
         bool skip_to_state_evaluation = false;
 
@@ -288,10 +344,18 @@ class parser
                             break;
                         }
 
-                        // parse key
+                        // remember we are now inside an object
+                        states.push_back(0);
+
+                        // parse key (the steps of parse_key(), which are
+                        // repeated here and below for speed)
                         if (JSON_HEDLEY_UNLIKELY(last_token != token_type::value_string))
                         {
-                            return syntax_error(*sax, exception_message(token_type::value_string, "object key"));
+                            if (!continue_after(key_error(sax, allow_recovery, false), skip_to_state_evaluation))
+                            {
+                                return false;
+                            }
+                            continue;
                         }
                         if (JSON_HEDLEY_UNLIKELY(!sax->key(m_lexer.get_string_token())))
                         {
@@ -301,12 +365,13 @@ class parser
                         // parse separator (:)
                         if (JSON_HEDLEY_UNLIKELY(!get_token_expecting(token_type::name_separator)))
                         {
-                            return syntax_error(*sax, exception_message(token_type::name_separator, "object separator"));
+                            if (!continue_after(key_error(sax, allow_recovery, true), skip_to_state_evaluation))
+                            {
+                                return false;
+                            }
+                            continue;
                         }
 
-                        // remember we are now inside an object
-                        states.push_back(0);
-
                         // parse values
                         get_token();
                         continue;
@@ -342,9 +407,11 @@ class parser
 
                         if (JSON_HEDLEY_UNLIKELY(!std::isfinite(res)))
                         {
-                            return sax->parse_error(m_lexer.get_position(),
-                                                    m_lexer.get_token_string(),
-                                                    out_of_range::create(exception_id::number_overflow, concat("number overflow parsing '", m_lexer.get_token_string(), '\''), nullptr));
+                            if (!overflow_error(sax, res, allow_recovery))
+                            {
+                                return false;
+                            }
+                            break;
                         }
 
                         if (JSON_HEDLEY_UNLIKELY(!sax->number_float(res, number_float_text(sax))))
@@ -412,16 +479,63 @@ class parser
                     case token_type::parse_error:
                     {
                         // using "uninitialized" to avoid an "expected" message
-                        return syntax_error(*sax, exception_message(token_type::uninitialized, "value"));
+                        if (!report_error(sax, parse_error::create(exception_id::syntax_error, m_lexer.get_position(), exception_message(token_type::uninitialized, "value"), nullptr), allow_recovery))
+                        {
+                            return false;
+                        }
+
+                        // recover: keep what can be read of the token
+                        recover_token(allow_recovery);
+                        if (last_token != token_type::uninitialized)
+                        {
+                            // a string or a number
+                            continue;
+                        }
+                        if (states.empty())
+                        {
+                            // look for the value after the garbage
+                            if (!skip_to_value(allow_recovery))
+                            {
+                                return false;
+                            }
+                            continue;
+                        }
+                        // nothing could be read
+                        if (JSON_HEDLEY_UNLIKELY(!sax->null()))
+                        {
+                            return false;
+                        }
+                        break;
                     }
                     case token_type::end_of_input:
                     {
                         if (JSON_HEDLEY_UNLIKELY(m_lexer.get_chars_read_total() == 1))
                         {
-                            return syntax_error(*sax, empty_input_message());
+                            // there is nothing to recover
+                            static_cast<void>(report_error(sax, parse_error::create(exception_id::syntax_error, m_lexer.get_position(),
+                                                           empty_input_message(), nullptr), allow_recovery));
+                            return false;
                         }
 
-                        return syntax_error(*sax, exception_message(token_type::literal_or_value, "value"));
+                        if (!report_error(sax, parse_error::create(exception_id::syntax_error, m_lexer.get_position(), exception_message(token_type::literal_or_value, "value"), nullptr), allow_recovery))
+                        {
+                            return false;
+                        }
+
+                        // recover: the input ends where a value is missing
+                        if (states.empty())
+                        {
+                            // there is no value
+                            return false;
+                        }
+                        if (!recover_missing_value(sax, states, allow_recovery))
+                        {
+                            return false;
+                        }
+                        // the state evaluation reads the token again
+                        unget_token(allow_recovery);
+                        skip_to_state_evaluation = true;
+                        continue;
                     }
                     case token_type::uninitialized:
                     case token_type::end_array:
@@ -431,7 +545,35 @@ class parser
                     case token_type::literal_or_value:
                     default: // the last token was unexpected
                     {
-                        return syntax_error(*sax, exception_message(token_type::literal_or_value, "value"));
+                        if (!report_error(sax, parse_error::create(exception_id::syntax_error, m_lexer.get_position(), exception_message(token_type::literal_or_value, "value"), nullptr), allow_recovery))
+                        {
+                            return false;
+                        }
+
+                        // recover
+                        if (states.empty())
+                        {
+                            // look for the value after the garbage
+                            if (!skip_to_value(allow_recovery))
+                            {
+                                return false;
+                            }
+                            continue;
+                        }
+                        if (last_token == token_type::name_separator)
+                        {
+                            // a stray ':'; the value may follow
+                            get_token();
+                            continue;
+                        }
+                        if (!recover_missing_value(sax, states, allow_recovery))
+                        {
+                            return false;
+                        }
+                        // the state evaluation reads the token again
+                        unget_token(allow_recovery);
+                        skip_to_state_evaluation = true;
+                        continue;
                     }
                 }
             }
@@ -482,7 +624,30 @@ class parser
                     continue;
                 }
 
-                return syntax_error(*sax, exception_message(token_type::end_array, "array"));
+                if (!report_error(sax, parse_error::create(exception_id::syntax_error, m_lexer.get_position(), exception_message(token_type::end_array, "array"), nullptr), allow_recovery))
+                {
+                    return false;
+                }
+
+                // recover
+                if (last_token == token_type::end_of_input)
+                {
+                    // the input ends inside the array
+                    return close_containers(sax, states, allow_recovery);
+                }
+                if (last_token == token_type::end_object)
+                {
+                    // a wrong closing bracket closes the innermost container
+                    if (JSON_HEDLEY_UNLIKELY(!sax->end_array()))
+                    {
+                        return false;
+                    }
+                    states.pop_back();
+                    skip_to_state_evaluation = true;
+                }
+                // otherwise, a missing ',' (or a stray ':', which value
+                // parsing drops): the next value begins here
+                continue;
             }
 
             // states.back() is false -> object
@@ -499,9 +664,12 @@ class parser
                     // parse key
                     if (JSON_HEDLEY_UNLIKELY(last_token != token_type::value_string))
                     {
-                        return syntax_error(*sax, exception_message(token_type::value_string, "object key"));
+                        if (!continue_after(key_error(sax, allow_recovery, false), skip_to_state_evaluation))
+                        {
+                            return false;
+                        }
+                        continue;
                     }
-
                     if (JSON_HEDLEY_UNLIKELY(!sax->key(m_lexer.get_string_token())))
                     {
                         return false;
@@ -510,7 +678,11 @@ class parser
                     // parse separator (:)
                     if (JSON_HEDLEY_UNLIKELY(!get_token_expecting(token_type::name_separator)))
                     {
-                        return syntax_error(*sax, exception_message(token_type::name_separator, "object separator"));
+                        if (!continue_after(key_error(sax, allow_recovery, true), skip_to_state_evaluation))
+                        {
+                            return false;
+                        }
+                        continue;
                     }
 
                     // parse values
@@ -538,10 +710,546 @@ class parser
                 continue;
             }
 
-            return syntax_error(*sax, exception_message(token_type::end_object, "object"));
+            if (!report_error(sax, parse_error::create(exception_id::syntax_error, m_lexer.get_position(), exception_message(token_type::end_object, "object"), nullptr), allow_recovery))
+            {
+                return false;
+            }
+
+            // recover
+            if (last_token == token_type::end_of_input)
+            {
+                // the input ends inside the object
+                return close_containers(sax, states, allow_recovery);
+            }
+            if (last_token == token_type::end_array)
+            {
+                // a wrong closing bracket closes the innermost container
+                if (JSON_HEDLEY_UNLIKELY(!sax->end_object()))
+                {
+                    return false;
+                }
+                states.pop_back();
+                skip_to_state_evaluation = true;
+                continue;
+            }
+            if (!continue_after(recover_member(sax, allow_recovery), skip_to_state_evaluation))
+            {
+                return false;
+            }
         }
     }
 
+    /*!
+    @brief continue sax_parse_internal() after a recovery
+    @return whether to continue parsing
+    */
+    bool continue_after(const next_step step, bool& skip_to_state_evaluation)
+    {
+        if (step == next_step::evaluate_state)
+        {
+            // the state evaluation reads the token again
+            m_lexer.unget_token();
+            skip_to_state_evaluation = true;
+        }
+        return step != next_step::stop;
+    }
+
+    /// the parser for parse() and accept() never recovers: stop parsing
+    JSON_INTERNAL_ALWAYS_INLINE
+    static std::false_type continue_after(std::false_type /*step*/, bool& /*skip_to_state_evaluation*/) noexcept
+    {
+        return {};
+    }
+
+    /*!
+    @brief parse an object key and the name separator (:) after it
+
+    last_token is the token where the key is expected. sax_parse_internal()
+    repeats these steps rather than calling this function, which is used
+    when recovering from an error.
+
+    @return next_step::parse_value if the value follows, with last_token its
+            first token; next_step::evaluate_state if the object's state is
+            to be evaluated after recovering from an error; next_step::stop
+            to stop parsing
+    */
+    template<typename SAX>
+    next_step parse_key(SAX* sax)
+    {
+        const std::true_type allow_recovery{};
+
+        if (JSON_HEDLEY_UNLIKELY(last_token != token_type::value_string))
+        {
+            return key_error(sax, allow_recovery, false);
+        }
+
+        if (JSON_HEDLEY_UNLIKELY(!sax->key(m_lexer.get_string_token())))
+        {
+            return next_step::stop;
+        }
+
+        // parse separator (:)
+        if (JSON_HEDLEY_UNLIKELY(get_token() != token_type::name_separator))
+        {
+            return key_error(sax, allow_recovery, true);
+        }
+
+        // the value begins with the next token
+        get_token();
+        return next_step::parse_value;
+    }
+
+    /*!
+    @brief report a number that is too large for number_float_t, and recover
+           from the error by passing the value on; the SAX parser gets the
+           number's text as well
+
+    This is a separate function, as reading other numbers is measurably
+    slower if the error is handled where they are read.
+
+    @param[in] sax    the SAX parser
+    @param[in] value  the value that is not finite
+    @return whether to continue parsing
+    */
+    template<typename SAX>
+    bool overflow_error(SAX* sax, const number_float_t value, std::true_type allow_recovery)
+    {
+        if (!report_error(sax, out_of_range::create(exception_id::number_overflow, concat("number overflow parsing '", m_lexer.get_token_string(), '\''), nullptr), allow_recovery))
+        {
+            return false;
+        }
+        return sax->number_float(value, number_float_text(sax));
+    }
+
+    /// @copydoc overflow_error
+    template<typename SAX>
+    JSON_INTERNAL_ALWAYS_INLINE
+    std::false_type overflow_error(SAX* sax, const number_float_t /*value*/, std::false_type allow_recovery)
+    {
+        return report_error(sax, out_of_range::create(exception_id::number_overflow, concat("number overflow parsing '", m_lexer.get_token_string(), '\''), nullptr), allow_recovery);
+    }
+
+    /*!
+    @brief report a missing key, or a missing name separator (:) after the
+           key; the parser for parse() and accept() never recovers
+
+    @param[in] key_read  whether the key was read, so that the name separator
+                         is missing
+    @return std::false_type, see report_error()
+    */
+    template<typename SAX>
+    JSON_INTERNAL_ALWAYS_INLINE
+    std::false_type key_error(SAX* sax, std::false_type allow_recovery, const bool key_read)
+    {
+        return report_error(sax, parse_error::create(exception_id::syntax_error, m_lexer.get_position(), key_read
+                            ? exception_message(token_type::name_separator, "object separator")
+                            : exception_message(token_type::value_string, "object key"), nullptr), allow_recovery);
+    }
+
+    /*!
+    @brief report a missing key, or a missing name separator (:) after the
+           key, and recover from it
+
+    @param[in] key_read  whether the key was read, so that the name separator
+                         is missing
+    */
+    template<typename SAX>
+    next_step key_error(SAX* sax, std::true_type allow_recovery, const bool key_read)
+    {
+        if (!key_read)
+        {
+            if (!report_error(sax, parse_error::create(exception_id::syntax_error, m_lexer.get_position(), exception_message(token_type::value_string, "object key"), nullptr), allow_recovery))
+            {
+                return next_step::stop;
+            }
+            return recover_key(sax);
+        }
+
+        if (!report_error(sax, parse_error::create(exception_id::syntax_error, m_lexer.get_position(), exception_message(token_type::name_separator, "object separator"), nullptr), allow_recovery))
+        {
+            return next_step::stop;
+        }
+        return recover_name_separator(sax);
+    }
+
+    /////////////////////
+    // error recovery
+    /////////////////////
+
+    /*
+    The functions below repair an error after the SAX parser's parse_error()
+    returned true (see #3989). Each mistake is repaired by the smallest local
+    edit: a missing ',' or ':' is inserted, a stray token is removed, what can
+    be read of an invalid string or number is kept (see
+    lexer::recover_token()), a missing value becomes null, a wrong closing
+    bracket closes the innermost container, and the end of the input closes
+    all of them. The events stay balanced, and every key() is followed by
+    exactly one value.
+
+    A repair hands a token to the state evaluation, by returning it to the
+    lexer (lexer::unget_token()) so that the state evaluation reads it again,
+    only if it is ',', ']', '}', or the end of the input. The state evaluation
+    hands a token to value or key parsing only if it is none of them, so a
+    token is never handed back and forth. Every other step reads a token or
+    closes a container, so parsing always ends.
+    */
+
+    /*!
+    @brief report an error to the SAX parser; the parser for parse() and
+           accept() never recovers
+
+    @return std::false_type rather than false: its value is known where the
+            function is called even if the call is not inlined, so the code
+            for recovering is not generated
+    */
+    template<typename SAX, typename Exception>
+    JSON_INTERNAL_ALWAYS_INLINE
+    std::false_type report_error(SAX* sax, const Exception& ex, std::false_type /*allow_recovery*/)
+    {
+        static_cast<void>(sax); // MSVC 2015 does not count calling a static parse_error() as using it
+        error_reported = true;
+        static_cast<void>(sax->parse_error(m_lexer.get_position(), m_lexer.get_token_string(), ex));
+        return {};
+    }
+
+    /*!
+    @brief report an error to the SAX parser
+    @return whether to recover from the error
+    */
+    template<typename SAX, typename Exception>
+    bool report_error(SAX* sax, const Exception& ex, std::true_type /*allow_recovery*/)
+    {
+        static_cast<void>(sax); // MSVC 2015 does not count calling a static parse_error() as using it
+        const std::size_t position = m_lexer.get_position().chars_read_total;
+        if (error_reported && position == last_error_position && last_token == last_error_token)
+        {
+            // a repair handed on the token of the error it repaired; the
+            // token was reported already, and the SAX parser asked to recover
+            return true;
+        }
+
+        error_reported = true;
+        last_error_position = position;
+        last_error_token = last_token;
+
+        if (!sax->parse_error(m_lexer.get_position(), m_lexer.get_token_string(), ex))
+        {
+            return false;
+        }
+
+        // the token string of the next error begins here
+        m_lexer.restart_token_string();
+        return true;
+    }
+
+    /*!
+    @brief keep what can be read of the token that the lexer rejected
+
+    The error was reported for the rejected token, so it is not reported again
+    for the token it is repaired to (see lexer::recover_token()).
+    */
+    token_type recover_token()
+    {
+        last_token = m_lexer.recover_token();
+        last_error_position = m_lexer.get_position().chars_read_total;
+        last_error_token = last_token;
+        return last_token;
+    }
+
+    // The functions below are called from sax_parse_internal() after an error
+    // was reported. The parser for parse() and accept() stopped there, so for
+    // it they are stubs: the code for recovering is then not even referenced,
+    // and unoptimized builds do not emit it either.
+
+    /// @copydoc recover_token()
+    JSON_INTERNAL_ALWAYS_INLINE
+    token_type recover_token(std::true_type /*allow_recovery*/)
+    {
+        return recover_token();
+    }
+
+    /// @copydoc recover_token()
+    JSON_INTERNAL_ALWAYS_INLINE
+    static std::false_type recover_token(std::false_type /*allow_recovery*/) noexcept
+    {
+        return {};
+    }
+
+    /// return the token that was read last to the lexer (see lexer::unget_token())
+    JSON_INTERNAL_ALWAYS_INLINE
+    void unget_token(std::true_type /*allow_recovery*/)
+    {
+        m_lexer.unget_token();
+    }
+
+    /// @copydoc unget_token(std::true_type)
+    JSON_INTERNAL_ALWAYS_INLINE
+    static void unget_token(std::false_type /*allow_recovery*/) noexcept {}
+
+    /// @copydoc skip_to_value
+    JSON_INTERNAL_ALWAYS_INLINE
+    static std::false_type skip_to_value(std::false_type /*allow_recovery*/) noexcept
+    {
+        return {};
+    }
+
+    /// @copydoc recover_missing_value
+    template<typename SAX>
+    JSON_INTERNAL_ALWAYS_INLINE
+    static std::false_type recover_missing_value(SAX* /*sax*/, const std::vector<std::uint8_t>& /*states*/, std::false_type /*allow_recovery*/) noexcept
+    {
+        return {};
+    }
+
+    /// pass the end events of all open containers
+    template<typename SAX>
+    bool close_containers(SAX* sax, std::vector<std::uint8_t>& states, std::true_type /*allow_recovery*/)
+    {
+        while (!states.empty())
+        {
+            const bool is_array = states.back() != 0;
+            states.pop_back();
+            if (JSON_HEDLEY_UNLIKELY(is_array ? !sax->end_array() : !sax->end_object()))
+            {
+                return false;
+            }
+        }
+        return true;
+    }
+
+    /// @copydoc close_containers
+    template<typename SAX>
+    JSON_INTERNAL_ALWAYS_INLINE
+    static std::false_type close_containers(SAX* /*sax*/, std::vector<std::uint8_t>& /*states*/, std::false_type /*allow_recovery*/) noexcept
+    {
+        return {};
+    }
+
+    /*!
+    @brief read tokens until one begins a value, skipping everything before
+           the top-level value
+    @return whether a value begins with last_token
+    */
+    bool skip_to_value(std::true_type /*allow_recovery*/)
+    {
+        while (true)
+        {
+            switch (get_token())
+            {
+                case token_type::begin_array:
+                case token_type::begin_object:
+                case token_type::literal_false:
+                case token_type::literal_null:
+                case token_type::literal_true:
+                case token_type::value_float:
+                case token_type::value_integer:
+                case token_type::value_string:
+                case token_type::value_unsigned:
+                    return true;
+
+                case token_type::end_of_input:
+                    return false;
+
+                case token_type::parse_error:
+                    recover_token();
+                    if (last_token != token_type::uninitialized)
+                    {
+                        return true;
+                    }
+                    break;
+
+                case token_type::uninitialized:
+                case token_type::end_array:
+                case token_type::end_object:
+                case token_type::name_separator:
+                case token_type::value_separator:
+                case token_type::literal_or_value:
+                default:
+                    break;
+            }
+        }
+    }
+
+    /*!
+    @brief skip the rest of an object member that cannot be read
+
+    Reads tokens, beginning with last_token, until a ',', '}', or ']' that is
+    not inside a container that begins in the skipped tokens, or the end of
+    the input.
+    */
+    void skip_member()
+    {
+        std::size_t depth = 0;
+        while (true)
+        {
+            switch (last_token)
+            {
+                case token_type::begin_array:
+                case token_type::begin_object:
+                    ++depth;
+                    break;
+
+                case token_type::end_array:
+                case token_type::end_object:
+                    if (depth == 0)
+                    {
+                        return;
+                    }
+                    --depth;
+                    break;
+
+                case token_type::value_separator:
+                    if (depth == 0)
+                    {
+                        return;
+                    }
+                    break;
+
+                case token_type::end_of_input:
+                    return;
+
+                case token_type::parse_error:
+                    recover_token();
+                    break;
+
+                case token_type::uninitialized:
+                case token_type::literal_true:
+                case token_type::literal_false:
+                case token_type::literal_null:
+                case token_type::value_string:
+                case token_type::value_unsigned:
+                case token_type::value_integer:
+                case token_type::value_float:
+                case token_type::name_separator:
+                case token_type::literal_or_value:
+                default:
+                    break;
+            }
+            get_token();
+        }
+    }
+
+    /*!
+    @brief pass a value where it is missing
+
+    last_token is ',', ']', '}', or the end of the input, where a value was
+    expected. In an object, the key gets null; in an array, a ',' where a
+    value is missing stands for null (as in JavaScript), while an array that
+    ends there just ends.
+    */
+    template<typename SAX>
+    bool recover_missing_value(SAX* sax, const std::vector<std::uint8_t>& states, std::true_type /*allow_recovery*/)
+    {
+        JSON_ASSERT(!states.empty());
+        if (states.back() == 0 || last_token == token_type::value_separator)
+        {
+            return sax->null();
+        }
+        return true;
+    }
+
+    /// recover from a missing key; last_token is where it was expected
+    template<typename SAX>
+    next_step recover_key(SAX* sax)
+    {
+        switch (last_token)
+        {
+            case token_type::value_separator:
+            case token_type::end_object:
+            case token_type::end_array:
+            case token_type::end_of_input:
+                // no member: the object's state handles the token
+                return next_step::evaluate_state;
+
+            case token_type::parse_error:
+                recover_token();
+                if (last_token == token_type::value_string)
+                {
+                    // a key that could be repaired
+                    return parse_key(sax);
+                }
+                skip_member();
+                return next_step::evaluate_state;
+
+            case token_type::uninitialized:
+            case token_type::literal_true:
+            case token_type::literal_false:
+            case token_type::literal_null:
+            case token_type::value_string:
+            case token_type::value_unsigned:
+            case token_type::value_integer:
+            case token_type::value_float:
+            case token_type::begin_array:
+            case token_type::begin_object:
+            case token_type::name_separator:
+            case token_type::literal_or_value:
+            default:
+                // a member without a key
+                skip_member();
+                return next_step::evaluate_state;
+        }
+    }
+
+    /// recover from a missing name separator (:) after the key; last_token
+    /// is where it was expected
+    template<typename SAX>
+    next_step recover_name_separator(SAX* sax)
+    {
+        switch (last_token)
+        {
+            case token_type::value_separator:
+            case token_type::end_object:
+            case token_type::end_array:
+            case token_type::end_of_input:
+                // the value is missing as well
+                return sax->null() ? next_step::evaluate_state : next_step::stop;
+
+            case token_type::uninitialized:
+            case token_type::literal_true:
+            case token_type::literal_false:
+            case token_type::literal_null:
+            case token_type::value_string:
+            case token_type::value_unsigned:
+            case token_type::value_integer:
+            case token_type::value_float:
+            case token_type::begin_array:
+            case token_type::begin_object:
+            case token_type::name_separator:
+            case token_type::parse_error:
+            case token_type::literal_or_value:
+            default:
+                // a missing ':'; the value begins here
+                return next_step::parse_value;
+        }
+    }
+
+    /// recover from a token after an object member that is neither ',' nor
+    /// '}' (nor ']' or the end of the input, which the caller handles)
+    template<typename SAX>
+    next_step recover_member(SAX* sax, std::true_type /*allow_recovery*/)
+    {
+        if (last_token == token_type::parse_error)
+        {
+            recover_token();
+        }
+        if (last_token == token_type::value_string)
+        {
+            // a missing ','; the next key begins here
+            return parse_key(sax);
+        }
+        skip_member();
+        return next_step::evaluate_state;
+    }
+
+    /// the parser for parse() and accept() never recovers (and does not come
+    /// here, as report_error() returned false)
+    template<typename SAX>
+    JSON_INTERNAL_ALWAYS_INLINE
+    std::false_type recover_member(SAX* /*sax*/, std::false_type /*allow_recovery*/) const noexcept
+    {
+        return {};
+    }
+
     /// get next token from lexer
     token_type get_token()
     {
@@ -612,6 +1320,12 @@ class parser
     const bool allow_exceptions = true;
     /// whether trailing commas in objects and arrays should be ignored (true) or signaled as errors (false)
     const bool ignore_trailing_commas = false;
+    /// whether an error was reported to the SAX parser
+    bool error_reported = false;
+    /// the position of the last reported error
+    std::size_t last_error_position = 0;
+    /// the token of the last reported error
+    token_type last_error_token = token_type::uninitialized;
 };
 
 }  // namespace detail

include/nlohmann/detail/macro_scope.hpp

@@ -215,6 +215,17 @@
     #define JSON_NO_UNIQUE_ADDRESS
 #endif
 
+// Inlines small functions even in unoptimized builds, so that they are not
+// emitted. The parsers for parse(), accept(), and from_*() use it for the
+// functions that stand in for the code recovering from errors (see #3989).
+// MSVC is left to decide, as it warns (C4714) where it does not inline a
+// __forceinline function.
+#if defined(_MSC_VER) && !defined(__clang__)
+    #define JSON_INTERNAL_ALWAYS_INLINE
+#else
+    #define JSON_INTERNAL_ALWAYS_INLINE JSON_HEDLEY_ALWAYS_INLINE
+#endif
+
 // Clang targeting MinGW does not survive the thread_local storage the copy
 // constructor uses to bound its descent: every test that copies a value
 // segfaults with clang 11.0.1 and clang 18.1.8, while the same tests pass with

include/nlohmann/detail/macro_unscope.hpp

@@ -19,6 +19,7 @@
 #undef NLOHMANN_CAN_CALL_STD_FUNC_IMPL
 #undef JSON_INLINE_VARIABLE
 #undef JSON_NO_UNIQUE_ADDRESS
+#undef JSON_INTERNAL_ALWAYS_INLINE
 #undef JSON_DISABLE_ENUM_SERIALIZATION
 #undef JSON_DISABLE_TUPLE_REFERENCE_CONVERSION
 #undef JSON_UNLOCKED_FILE_IO

include/nlohmann/detail/string_utils.hpp

@@ -13,6 +13,7 @@
 #include <cstddef> // size_t
 #include <cstdint> // uint8_t, uint32_t
 #include <string> // string, to_string
+#include <utility> // move
 
 #include <nlohmann/detail/abi_macros.hpp>
 #include <nlohmann/detail/macro_scope.hpp>

include/nlohmann/detail/view/errors.hpp

@@ -144,11 +144,21 @@ struct throwing_sax
 
     /// (a template, as in json_sax_dom_parser: the parser passes the concrete
     /// exception type, which is thrown as is)
+    ///
+    /// It is not [[noreturn]]: sax_parse() recovers from an error if
+    /// parse_error() returns true, and MSVC reports that code as unreachable
+    /// (C4702) for a handler that never returns. The flag hides that from it.
     template<class Exception>
-    [[noreturn]] bool parse_error(std::size_t /*unused*/, const std::string& /*unused*/, const Exception& ex)
+    bool parse_error(std::size_t /*unused*/, const std::string& /*unused*/, const Exception& ex)
     {
-        NLOHMANN_VIEW_THROW(ex);
+        if (throwing)
+        {
+            NLOHMANN_VIEW_THROW(ex);
+        }
+        return false; // LCOV_EXCL_LINE
     }
+
+    bool throwing = true;
 };
 
 /*!

include/nlohmann/detail/view/number.hpp

@@ -19,7 +19,6 @@
 #include <nlohmann/detail/view/macro_scope.hpp>
 #include <nlohmann/detail/view/node.hpp>
 #include <nlohmann/detail/view/scan.hpp>
-#include <nlohmann/detail/view/simd.hpp>
 
 NLOHMANN_JSON_NAMESPACE_BEGIN
 namespace detail
@@ -104,56 +103,12 @@ NLOHMANN_VIEW_NOINLINE FloatType float_value(const char* first, const node& n)
     return convert_float<FloatType>(first, last, dot, mantissa_end);
 }
 
-#if NLOHMANN_VIEW_VECTOR
-/*!
-@brief the value of the first 16 digits of a run of digits with a point after
-int_digits of them (or none, if int_digits >= 16)
-
-The 17 bytes at p are read at once: the bytes after the point move down by one,
-then pairs, quadruples, and octets of digits are combined by multiplications
-(after simdjson). The bytes [p, p + 32) must be readable. Bytes that are not
-digits (only in a damaged image loaded with image_check::bounds) give a wrong
-value, but no undefined behavior.
-*/
-NLOHMANN_VIEW_ALWAYS_INLINE std::uint64_t sixteen_digits(const unsigned char* p, unsigned int_digits) noexcept
-{
-#if NLOHMANN_VIEW_NEON
-    const uint8x16_t low = vld1q_u8(p);
-    const uint8x16_t index = vcombine_u8(vcreate_u8(0x0706050403020100u), vcreate_u8(0x0F0E0D0C0B0A0908u));
-    const uint8x16_t before_point = vcltq_u8(index, vdupq_n_u8(static_cast<std::uint8_t>(int_digits)));
-    const uint8x16_t digits = vsubq_u8(vbslq_u8(before_point, low, vextq_u8(low, vld1q_u8(p + 16), 1)), vdupq_n_u8('0'));
-    const uint16x8_t twos = vpaddlq_u8(vmulq_u8(digits, vreinterpretq_u8_u16(vdupq_n_u16(0x010A))));
-    const uint32x4_t fours = vpaddlq_u16(vmulq_u16(twos, vreinterpretq_u16_u32(vdupq_n_u32(0x00010064))));
-    const uint64x2_t eights = vpaddlq_u32(vmulq_u32(fours, vreinterpretq_u32_u64(vdupq_n_u64(0x0000000100002710u))));
-    return (vgetq_lane_u64(eights, 0) * 100000000u) + vgetq_lane_u64(eights, 1);
-#else
-    // NOLINTBEGIN(portability-simd-intrinsics)
-    const __m128i low = _mm_loadu_si128(static_cast<const __m128i*>(static_cast<const void*>(p))); // NOLINT(bugprone-casting-through-void)
-    const __m128i high = _mm_loadu_si128(static_cast<const __m128i*>(static_cast<const void*>(p + 16))); // NOLINT(bugprone-casting-through-void)
-    const __m128i before_point = _mm_cmpgt_epi8(_mm_set1_epi8(static_cast<char>(int_digits)), _mm_setr_epi8(0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15));
-    const __m128i moved = _mm_or_si128(_mm_srli_si128(low, 1), _mm_slli_si128(high, 15));
-    const __m128i digits = _mm_sub_epi8(_mm_or_si128(_mm_and_si128(before_point, low), _mm_andnot_si128(before_point, moved)), _mm_set1_epi8('0'));
-    // (16-bit lanes: SSE2 has no multiply-add of bytes)
-    const __m128i tens = _mm_setr_epi16(10, 1, 10, 1, 10, 1, 10, 1);
-    const __m128i twos = _mm_packs_epi32(_mm_madd_epi16(_mm_unpacklo_epi8(digits, _mm_setzero_si128()), tens),
-                                         _mm_madd_epi16(_mm_unpackhi_epi8(digits, _mm_setzero_si128()), tens));
-    const __m128i fours = _mm_madd_epi16(twos, _mm_setr_epi16(100, 1, 100, 1, 100, 1, 100, 1));
-    const __m128i eights = _mm_madd_epi16(_mm_packs_epi32(fours, fours), _mm_setr_epi16(10000, 1, 10000, 1, 10000, 1, 10000, 1));
-    const auto upper = static_cast<std::uint32_t>(_mm_cvtsi128_si32(eights));
-    const auto lower = static_cast<std::uint32_t>(_mm_cvtsi128_si32(_mm_srli_si128(eights, 4)));
-    // NOLINTEND(portability-simd-intrinsics)
-    return (std::uint64_t{upper} * 100000000u) + lower;
-#endif
-}
-#endif
-
 /*!
 @brief the digits of a float token with at most 19 digits, from its layout
 
 The digit layout recorded while parsing says where the integer digits, the
 fraction digits, and the exponent are, so the digits are read eight at a
-time without scanning (with SIMD, a token of 16 to 19 digits sixteen at a
-time).
+time without scanning.
 
 @param[in] p  first character of the token
 @param[in] e  end of the token
@@ -163,34 +118,15 @@ NLOHMANN_VIEW_ALWAYS_INLINE float_significand layout_decimal(const unsigned char
 {
     const bool negative = *p == '-';
     p += negative ? 1 : 0;
-    std::uint64_t w = 0;
+    std::uint64_t w = parse_upto19(p, int_digits, limit);
+    p += int_digits;
     std::int64_t q = 0;
-#if NLOHMANN_VIEW_VECTOR
-    const unsigned count = int_digits + frac_digits;
-    if (count >= 16 && limit - p >= 32)
+    if (frac_digits != 0)
     {
-        // (the point, if any, follows the integer digits; the digits after
-        // the first 16 are at most three)
-        w = sixteen_digits(p, int_digits);
-        for (unsigned i = 16; i < count; ++i)
-        {
-            w = (w * 10) + static_cast<std::uint64_t>(p[i + (i >= int_digits ? 1 : 0)] - '0');
-        }
-        p += count + (frac_digits != 0 ? 1 : 0);
+        w = (w * int_pow10(frac_digits)) + parse_upto19(p + 1, frac_digits, limit);
+        p += 1 + frac_digits;
         q = -static_cast<std::int64_t>(frac_digits);
     }
-    else
-#endif
-    {
-        w = parse_upto19(p, int_digits, limit);
-        p += int_digits;
-        if (frac_digits != 0)
-        {
-            w = (w * int_pow10(frac_digits)) + parse_upto19(p + 1, frac_digits, limit);
-            p += 1 + frac_digits;
-            q = -static_cast<std::int64_t>(frac_digits);
-        }
-    }
     if (p != e)
     {
         // [eE][+-]digits; huge exponents saturate (the parser rejected
@@ -225,22 +161,12 @@ NLOHMANN_VIEW_ALWAYS_INLINE float_significand layout_decimal(const unsigned char
 The result is correctly rounded by the lexer's conversion
 (detail::decimal_to_float(): Clinger's fast path where both operands are
 exact, else the Eisel-Lemire algorithm, which needs no fallback for up to 19
-digits), so it is the value parse() produces. Tokens of 16 or more digits go
-to Eisel-Lemire directly: the fast path takes few of them (none of 17 digits),
-and its check is a branch that data like GeoJSON coordinates mispredicts.
+digits), so it is the value parse() produces.
 */
 template<typename FloatType>
 NLOHMANN_VIEW_ALWAYS_INLINE FloatType layout_float(const unsigned char* p, const unsigned char* e, unsigned int_digits, unsigned frac_digits, const unsigned char* limit) noexcept
 {
-    const float_significand d = layout_decimal(p, e, int_digits, frac_digits, limit);
-    if (int_digits + frac_digits >= 16)
-    {
-        using format = ieee_binary_format<std::numeric_limits<FloatType>::digits>;
-        native_float_t<FloatType> value{};
-        float_from_bits(eisel_lemire<format>(d.exponent, d.w) | (d.negative ? (std::uint64_t{1} << format::sign_bit()) : 0u), value);
-        return value;
-    }
-    return decimal_to_float<FloatType>(d);
+    return decimal_to_float<FloatType>(layout_decimal(p, e, int_digits, frac_digits, limit));
 }
 
 /// the value of a float set by an edit: its token (the shortest round-trip
@@ -259,80 +185,6 @@ NLOHMANN_VIEW_NOINLINE FloatType edited_float(const char* token, const node& n)
     return float_value<FloatType>(token, n);
 }
 
-#if NLOHMANN_VIEW_VECTOR
-/// eisel_lemire() for the token w * 10^q of coordinate_float()
-template<typename Format>
-NLOHMANN_VIEW_ALWAYS_INLINE std::uint64_t coordinate_bits(std::int64_t q, std::uint64_t w, Format /*format*/) noexcept
-{
-    return eisel_lemire<Format>(q, w);
-}
-
-/*!
-@brief (binary64) eisel_lemire() without the cases that tokens of 16 or 17
-digits with 5 or more after the point cannot reach: a value of 0, a
-subnormal, an infinity, or a tie (q < -4)
-*/
-NLOHMANN_VIEW_ALWAYS_INLINE std::uint64_t coordinate_bits(std::int64_t q, std::uint64_t w, ieee_binary_format<53> format) noexcept
-{
-    if (NLOHMANN_VIEW_UNLIKELY(w == 0 || q > -5))
-    {
-        return eisel_lemire<ieee_binary_format<53>>(q, w);
-    }
-    static_cast<void>(format);
-    const int lz = count_leading_zeros(w);
-    const std::uint64_t normalized = w << static_cast<unsigned>(lz);
-    const auto index = static_cast<std::size_t>(2 * (q - pow5_128_smallest_power));
-    const uint128_parts product = full_multiplication(normalized, pow5_128()[index]);
-    std::uint64_t high = product.high;
-    if (NLOHMANN_VIEW_UNLIKELY((high & 0x1FFu) == 0x1FFu))
-    {
-        // (the lower bits may carry into the result)
-        const std::uint64_t second = full_multiplication(normalized, pow5_128()[index + 1]).high;
-        high += product.low + second < product.low ? 1u : 0u;
-    }
-    const auto upperbit = static_cast<unsigned>(high >> 63u);
-    std::uint64_t mantissa = high >> (upperbit + 9);
-    std::uint64_t power2 = static_cast<std::uint64_t>(((217706 * q) >> 16) + 63 + static_cast<std::int64_t>(upperbit) - lz + 1023);
-    mantissa = (mantissa + (mantissa & 1u)) >> 1u;
-    const bool carry = mantissa >= (std::uint64_t{2} << 52u);
-    mantissa = carry ? (std::uint64_t{1} << 52u) : mantissa;
-    power2 += carry ? 1u : 0u;
-    return (mantissa & ~(std::uint64_t{1} << 52u)) | (power2 << 52u);
-}
-
-/*!
-@brief (binary32 and binary64) the value of a float token of the source with
-16 or 17 digits, a point, and no exponent, the shape of most coordinates
-written with full precision; whether the token of n has this shape
-
-These tokens skip the general layout of layout_decimal(): no digits beyond
-the 17th, no exponent (and Eisel-Lemire directly, as in layout_float()).
-*/
-template<typename FloatType>
-NLOHMANN_VIEW_ALWAYS_INLINE bool coordinate_float(const document_data& d, const node& n, FloatType& value) noexcept
-{
-    const unsigned int_digits = n.extra & 0xFFu;
-    const unsigned frac_digits = n.extra >> 8u;
-    const unsigned count = int_digits + frac_digits;
-    const auto* const first = reinterpret_cast<const unsigned char*>(d.src + n.off); // NOLINT(cppcoreguidelines-pro-type-reinterpret-cast)
-    const unsigned negative = first[0] == '-' ? 1u : 0u;
-    if (count - 16u > 1u || frac_digits == 0 || n.len != negative + count + 1 || d.size - n.off < 33)
-    {
-        return false;
-    }
-    const unsigned char* const digits = first + negative;
-    const std::uint64_t w = sixteen_digits(digits, int_digits);
-    // (the 17th digit follows the point)
-    const std::uint64_t w17 = (w * 10) + static_cast<std::uint64_t>(digits[17] - '0');
-    using format = ieee_binary_format<std::numeric_limits<FloatType>::digits>;
-    native_float_t<FloatType> result{};
-    float_from_bits(coordinate_bits(-static_cast<std::int64_t>(frac_digits), count == 17 ? w17 : w, format{}) | (static_cast<std::uint64_t>(negative) << format::sign_bit()), result);
-    value = result;
-    return true;
-}
-
-#endif
-
 /// the value of the float token of a node, as parse() converts it; floats and
 /// doubles with at most 19 digits are converted from the digit layout
 template<typename FloatType>
@@ -348,13 +200,6 @@ FloatType float_value(const document_data& d, const node& n)
 template<typename FloatType>
 FloatType float_value(const document_data& d, const node& n, std::true_type /*binary32 or binary64*/)
 {
-#if NLOHMANN_VIEW_VECTOR
-    FloatType value {};
-    if (coordinate_float(d, n, value))
-    {
-        return value;
-    }
-#endif
     const unsigned int_digits = n.extra & 0xFFu;
     const unsigned frac_digits = n.extra >> 8u;
     if (NLOHMANN_VIEW_LIKELY(int_digits + frac_digits <= 19)) // (255 marks "many")

include/nlohmann/detail/view/serializer.hpp

@@ -14,9 +14,7 @@
 #include <cstddef> // size_t
 #include <cstdint> // uint8_t, uint32_t
 #include <cstring> // memcpy, memset
-#include <functional> // less
 #include <limits> // numeric_limits
-#include <memory> // unique_ptr
 #include <type_traits> // integral_constant
 #include <vector> // vector
 
@@ -386,127 +384,6 @@ struct depth_counter<false>
     }
 };
 
-/*!
-@brief where the double x nearest to a token w * 10^q lies, for 10^15 <= w <
-10^17 and -17 <= q <= -1 (Eisel-Lemire, then a fixed-point position)
-
-dump() writes the shortest decimal in the rounding interval of x (the closest
-one if there are several, see zmij::to_shortest()). Its digits are those of a
-multiple of 10^g of the token's last digit, with g = dec_exp + 1 - q for x's
-decimal exponent dec_exp: the multiple of 10^g nearest to x if it lies in the
-interval (at most one does), else the multiple of 10^(g - 1) nearest to x
-(see token_digits_1() and token_digits_2()). For g <= 0, the token itself is
-in the interval.
-
-The position is exact in a 64-bit fixed-point number with 56 bits after the
-point (of the token's last digit): x / 10^q = 2c * 5^-q / 2^t for x = c * 2^e2
-and t = q + 1 - e2, with at most 56 bits after the point here, and |x / 10^q -
-w| is at most half the spacing of the doubles, below 12 units. So only the low
-64 bits of 2c * 5^-q are needed. c and e2 come from Eisel-Lemire's product
-directly (as eisel_lemire() rounds it: no tie is possible for q < -4, and its
-second product is added where it can carry into the result).
-
-@param[out] offset  x / 10^q - w + 32, with 56 bits after the point
-@param[out] radius  half the spacing of the doubles at x, likewise (+ 1 if
-                    the interval is closed)
-@return g; 3 if the digits are not decided here (a power of two, whose
-        interval is not symmetric, or out of range)
-*/
-NLOHMANN_VIEW_ALWAYS_INLINE int locate_token(std::uint64_t w, int q, std::uint64_t& offset, std::uint64_t& radius) noexcept
-{
-    const int lz = count_leading_zeros(w);
-    const std::uint64_t normalized = w << static_cast<unsigned>(lz);
-    const auto index = static_cast<std::size_t>(2 * (q - pow5_128_smallest_power));
-    const uint128_parts product = full_multiplication(normalized, pow5_128()[index]);
-    std::uint64_t high = product.high;
-    if (NLOHMANN_VIEW_UNLIKELY((high & 0x1FFu) == 0x1FFu))
-    {
-        // (as eisel_lemire(): the lower bits may carry into the result)
-        const std::uint64_t second = full_multiplication(normalized, pow5_128()[index + 1]).high;
-        high += product.low + second < product.low ? 1u : 0u;
-    }
-    const auto upperbit = static_cast<unsigned>(high >> 63u);
-    const std::uint64_t mantissa = high >> (upperbit + 9);
-    const std::uint64_t c = (mantissa + (mantissa & 1u)) >> 1u;
-    // e2 as eisel_lemire() computes the exponent (log2(10) ~ 217706 / 2^16)
-    const int e2 = static_cast<int>((217706 * static_cast<std::int64_t>(q)) >> 16) + 63 + static_cast<int>(upperbit) - lz - 52;
-    const int t = q + 1 - e2;
-    const int g = ((e2 * 315653) >> 20) + 1 - q; // (dec_exp as in zmij::to_shortest(), for a regular x)
-    static const std::array<std::uint64_t, 17> fives = {{1u, 5u, 25u, 125u, 625u, 3125u, 15625u, 78125u, 390625u, 1953125u, 9765625u, 48828125u, 244140625u, 1220703125u, 6103515625u, 30517578125u, 152587890625u}};
-    const std::uint64_t five = fives[static_cast<std::size_t>(-q)]; // 5^-q
-    const unsigned s = static_cast<unsigned>(56 - t) & 63u;
-    offset = ((2 * c * five) << s) - (w << 56u) + (std::uint64_t{32} << 56u);
-    radius = (five << s) + ((c & 1u) == 0 ? 1u : 0u);
-    // (c = 2^52: a power of two; c = 2^53: rounded up to one. With 8 to 16
-    // fraction digits, as the batch's tokens have, t is in [15, 40] and g at
-    // most 2. x lies halfway between two candidates only if its distance from
-    // the token is a whole unit (g == 2) or a half (g == 1): such tokens are
-    // left to the full conversion)
-    const bool decided = c - ((std::uint64_t{1} << 52u) + 1) < (std::uint64_t{1} << 52u) - 1
-                         && (offset & ((std::uint64_t{1} << 55u) - 1)) != 0;
-    return decided ? g : 3;
-}
-
-/// the distance of a position (see locate_token()) from the biased point 32
-NLOHMANN_VIEW_ALWAYS_INLINE std::uint64_t biased_distance(std::uint64_t position) noexcept
-{
-    constexpr std::uint64_t center = std::uint64_t{32} << 56u;
-    return position >= center ? position - center : center - position;
-}
-
-/*!
-@brief (g == 1) the digits dump() writes for a token, from the position of its
-double (see locate_token()): the multiple of 10 nearest to x if it is in the
-interval, else w rounded to x (an exact tie between two candidates, like
-754570942.33984373, whose double is 754570942.33984375, is excluded by
-locate_token())
-
-With |x - w| at most the radius, which is below half the spacing of the
-candidates, the multiple of 10 nearest to x is w - r10 or the one above it.
-*/
-NLOHMANN_VIEW_ALWAYS_INLINE std::uint64_t token_digits_1(std::uint64_t w, std::uint64_t offset, std::uint64_t radius, std::uint64_t r10) noexcept
-{
-    constexpr std::uint64_t one = std::uint64_t{1} << 56u;
-    const std::uint64_t from = offset + (r10 * one); // x - (w - r10) (+ 32)
-    const bool up = from > 37 * one;                  // (nearer to the multiple above)
-    const bool shorter = biased_distance(from - (up ? 10 * one : 0)) < radius;
-    const std::uint64_t a = w - r10 + (up ? 10u : 0u);
-    const std::uint64_t b = w + ((offset + (one / 2)) >> 56u) - 32;                // w rounded to x
-    return shorter ? a : b;
-}
-
-/*!
-@brief (g == 2) the digits dump() writes for a token, as token_digits_1(): the
-multiple of 100 nearest to x if it is in the interval, else the multiple of 10
-nearest to x
-*/
-NLOHMANN_VIEW_ALWAYS_INLINE std::uint64_t token_digits_2(std::uint64_t w, std::uint64_t offset, std::uint64_t radius, std::uint64_t r10, std::uint64_t r100) noexcept
-{
-    constexpr std::uint64_t one = std::uint64_t{1} << 56u;
-    const std::uint64_t from = offset + (r100 * one);
-    const bool up = from > 82 * one;
-    const bool shorter = biased_distance(from - (up ? 100 * one : 0)) < radius;
-    const std::uint64_t a = w - r100 + (up ? 100u : 0u);
-    // x - (w - r10), in [-12, 21]: the multiple of 10 nearest to x is that one,
-    // or one or two above, or one below
-    const std::uint64_t y = offset + (r10 * one);
-    const std::uint64_t steps = (y > 37 * one ? 1u : 0u) + (y > 47 * one ? 1u : 0u);
-    const bool down = y < 27 * one;
-    const std::uint64_t b = w - r10 + (10 * steps) - (down ? 10u : 0u);
-    return shorter ? a : b;
-}
-
-/// the four ASCII digits of x < 10^4, the first in the lowest byte (as
-/// little-endian memory holds them)
-NLOHMANN_VIEW_ALWAYS_INLINE std::uint32_t four_digit_text(std::uint32_t x) noexcept
-{
-    const std::uint32_t ab = (x * 5243u) >> 19u; // x / 100
-    std::uint32_t v = ab | ((x - (ab * 100u)) << 16u);
-    const std::uint32_t a = ((v * 103u) >> 10u) & 0x000F000Fu; // / 10 per 16-bit lane
-    v = a | ((v - (a * 10u)) << 8u);
-    return v + 0x30303030u;
-}
-
 /// how the view's dump() writes a value
 struct dump_style
 {
@@ -834,7 +711,6 @@ class view_serializer
         std::array<frame, 32> buffer; // NOLINT(cppcoreguidelines-pro-type-member-init,hicpp-member-init): written before read
         std::vector<frame> heap;
         small_stack<frame> stack(buffer.data(), buffer.size(), heap);
-        float_batch batch;
         const char* const src = m_doc.src;
         const char* const src_end = src + m_doc.size;
         char* w = m_out.cursor();
@@ -974,7 +850,7 @@ class view_serializer
                     else if (enabled(fast_double::value))
                     {
                         room(64);
-                        w = write_double_ahead(w, src, n, batch);
+                        w = write_double_at(w, *n);
                     }
                     else
                     {
@@ -999,20 +875,6 @@ class view_serializer
                 case value_t::array:
                 {
                     const bool object = n->kind == static_cast<std::uint8_t>(value_t::object);
-                    // (read-only documents, compact output) a pair of floats,
-                    // like the coordinates of GeoJSON, without a frame
-                    if (!Editable && !Pretty && !SourceNumbers && enabled(fast_double::value) && n->next == 3 && n->len == 2
-                            && n[1].kind == static_cast<std::uint8_t>(value_t::number_float) && n[2].kind == static_cast<std::uint8_t>(value_t::number_float))
-                    {
-                        room(2 * 64);
-                        *w++ = '[';
-                        w = write_double_ahead(w, src, n + 1, batch);
-                        *w++ = ',';
-                        w = write_double_ahead(w, src, n + 2, batch);
-                        *w++ = ']';
-                        n += 2; // (the element before the next node)
-                        break;
-                    }
                     room(8);
                     if (n->len == 0)
                     {
@@ -1122,247 +984,6 @@ class view_serializer
         }
     }
 
-    /// (float_batch) the mark of digits whose rounding carried beyond the
-    /// token's last four (see mark())
-    static constexpr std::uint64_t carried() noexcept
-    {
-        return std::uint64_t{1} << 63u;
-    }
-
-    /// floats of 16 or 17 digits ahead of the walk, converted together (see
-    /// convert_ahead()), with their digits as mark() keeps them
-    struct float_batch
-    {
-        static constexpr std::size_t capacity() noexcept
-        {
-            return 256;
-        }
-        /// (on the heap, allocated by the first conversion: a frame of more
-        /// than a page made every dump of a small document slower)
-        struct entries
-        {
-            std::array<const node*, 256> at; // NOLINT(cppcoreguidelines-pro-type-member-init,hicpp-member-init): written before read
-            std::array<std::uint64_t, 256> digits; // NOLINT(cppcoreguidelines-pro-type-member-init,hicpp-member-init): written before read
-        };
-        std::unique_ptr<entries> storage{};
-        const node** at = nullptr;
-        std::uint64_t* digits = nullptr;
-        std::size_t count = 0;
-        std::size_t next = 0;
-        const node* scanned = nullptr; ///< where the last scan of the node array ended
-    };
-
-    /// (doubles) the float at n as dump() writes it, at w (64 bytes of room):
-    /// from the batch if it holds n
-    NLOHMANN_VIEW_ALWAYS_INLINE char* write_double_ahead(char* w, const char* src, const node* n, float_batch& batch)
-    {
-#if NLOHMANN_VIEW_VECTOR && NLOHMANN_VIEW_LITTLE_ENDIAN
-        // (digits 0: not written by the batch; carried(): rounding carried
-        // beyond the last four digits)
-        if (batch.next != batch.count && batch.at[batch.next] == n && batch.digits[batch.next] - 1 < carried() - 1)
-        {
-            return write_token_digits(w, src, *n, batch.digits[batch.next++]);
-        }
-        return write_double_beside(w, n, batch);
-#else
-        static_cast<void>(src);
-        static_cast<void>(batch);
-        return write_double_at(w, *n);
-#endif
-    }
-
-#if NLOHMANN_VIEW_VECTOR && NLOHMANN_VIEW_LITTLE_ENDIAN
-    /// write_double_ahead() for a float that is not the batch's next one (out
-    /// of line: the walk is faster with less code in its loop)
-    NLOHMANN_VIEW_NOINLINE char* write_double_beside(char* w, const node* n, float_batch& batch)
-    {
-        // (an editable document's walk can skip nodes of the array)
-        const std::less<const node*> before;
-        while (batch.next != batch.count && before(batch.at[batch.next], n))
-        {
-            ++batch.next;
-        }
-        if ((batch.next != batch.count && batch.at[batch.next] == n) || convert_ahead(n, batch))
-        {
-            const std::uint64_t digits = batch.digits[batch.next++];
-            if ((digits & carried()) != 0)
-            {
-                return write_carried_digits(w, *n, digits & ~carried());
-            }
-            if (digits != 0)
-            {
-                return write_token_digits(w, m_doc.src, *n, digits);
-            }
-        }
-        return write_double_at(w, *n);
-    }
-
-    /*!
-    @brief convert the floats of 16 or 17 digits among the next nodes of the
-    array at once; whether n is the first of them
-
-    Such tokens (like the coordinates of GeoJSON, written with 17 digits) are
-    written as the token with its last four digits replaced
-    (write_token_digits()). Each step (the digits, locate_token(),
-    token_digits_1/2()) is a long chain of dependent operations; one loop per
-    step over many floats lets the processor overlap the floats, which it does
-    not across the steps of one float (on Apple M1, one loop for the digits and
-    locate_token() made the dump of canada.json 12 % slower). Each node of the array
-    is scanned once; a node that the walk does not reach in order (in an
-    editable document) is written without the batch.
-    */
-    NLOHMANN_VIEW_NOINLINE bool convert_ahead(const node* n, float_batch& batch)
-    {
-        const std::less<const node*> before;
-        const node* const last = m_doc.nodes + m_doc.node_count;
-        if (before(n, batch.scanned) || before(n, m_doc.nodes) || !before(n, last))
-        {
-            return false;
-        }
-        const char* const src = m_doc.src;
-        const char* const src_end = src + m_doc.size;
-        const node* const stop = last - n > 1024 ? n + 1024 : last;
-        batch.count = 0;
-        batch.next = 0;
-        // the digits of each token (and its last four from the text)
-        std::array<std::uint64_t, 256> digits;     // NOLINT(cppcoreguidelines-pro-type-member-init,hicpp-member-init): written before read
-        std::array<int, 256> exponent;             // NOLINT(cppcoreguidelines-pro-type-member-init,hicpp-member-init): written before read
-        std::array<std::uint32_t, 256> last_four;  // NOLINT(cppcoreguidelines-pro-type-member-init,hicpp-member-init): written before read
-        const node* p = n;
-        if (!batch.storage)
-        {
-            batch.storage.reset(new typename float_batch::entries); // NOLINT(cppcoreguidelines-owning-memory,modernize-make-unique): C++11
-            batch.at = batch.storage->at.data();
-            batch.digits = batch.storage->digits.data();
-        }
-        for (; p != stop && batch.count != float_batch::capacity(); ++p)
-        {
-            // a token of the source: 16 or 17 digits, at least 8 after the
-            // point, no exponent, 32 bytes to read
-            const unsigned frac_digits = p->extra >> 8u;
-            const unsigned int_digits = p->extra & 0xFFu;
-            const unsigned count = int_digits + frac_digits;
-            if (p->kind == static_cast<std::uint8_t>(value_t::number_float) && (p->flags & node_flags::storage) == 0
-                    && count - 16u <= 1u && frac_digits >= 8 && src_end - (src + p->off) >= 33)
-            {
-                const auto* const token = reinterpret_cast<const unsigned char*>(src + p->off); // NOLINT(cppcoreguidelines-pro-type-reinterpret-cast)
-                const unsigned sign = token[0] == '-' ? 1 : 0;
-                const unsigned char* const first = token + sign;
-                const std::uint64_t w = sixteen_digits(first, int_digits);
-                // (the 17th digit follows the point; a token with an exponent
-                // gets digits outside the range of locate_token(), and is not
-                // written by the batch. A token 0.xxx with 16 significant
-                // digits is: its double is written as "0." and its digits too)
-                const std::uint64_t w17 = (w * 10) + static_cast<std::uint64_t>(first[17] - '0');
-                const std::size_t i = batch.count++;
-                batch.at[i] = p;
-                digits[i] = p->len != sign + count + 1 ? 0 : (count == 17 ? w17 : w);
-                exponent[i] = -static_cast<int>(frac_digits);
-                // (byte 0: the last digit, byte 1: the last two, bits 16-29: the last four)
-                std::uint32_t four = 0;
-                std::memcpy(&four, token + p->len - 4, 4);
-                four -= 0x30303030u;
-                const std::uint32_t pairs = ((four & 0x0F0F0F0Fu) * 2561u) >> 8u; // (byte 0, 2: the pairs)
-                last_four[i] = (four >> 24u) | ((pairs >> 16u & 0xFFu) << 8u) | ((((pairs & 0x00FF00FFu) * 6553601u) >> 16u) << 16u);
-            }
-        }
-        batch.scanned = p;
-        // the position of each double, and the floats of either kind
-        std::array<std::uint64_t, 256> offset; // NOLINT(cppcoreguidelines-pro-type-member-init,hicpp-member-init): written before read
-        std::array<std::uint64_t, 256> radius; // NOLINT(cppcoreguidelines-pro-type-member-init,hicpp-member-init): written before read
-        std::array<std::uint8_t, 256> kind_1;  // NOLINT(cppcoreguidelines-pro-type-member-init,hicpp-member-init): written before read
-        std::array<std::uint8_t, 256> kind_2;  // NOLINT(cppcoreguidelines-pro-type-member-init,hicpp-member-init): written before read
-        std::size_t count_1 = 0;
-        std::size_t count_2 = 0;
-        for (std::size_t i = 0; i != batch.count; ++i)
-        {
-            const std::uint64_t w = digits[i];
-            // (the range excludes the values of bytes that are not digits, only
-            // in a damaged image loaded with image_check::bounds)
-            const int g = w - 1000000000000000u < 99000000000000000u ? locate_token(w, exponent[i], offset[i], radius[i]) : 3;
-            batch.digits[i] = g <= 0 ? (last_four[i] >> 16u) + 1 : 0;
-            kind_1[count_1] = static_cast<std::uint8_t>(i);
-            count_1 += g == 1 ? 1 : 0;
-            kind_2[count_2] = static_cast<std::uint8_t>(i);
-            count_2 += g == 2 ? 1 : 0;
-        }
-        // the digits to write, in one loop per kind (see mark())
-        for (std::size_t k = 0; k != count_1; ++k)
-        {
-            const std::size_t i = kind_1[k];
-            const std::uint64_t s = token_digits_1(digits[i], offset[i], radius[i], last_four[i] & 0xFFu);
-            batch.digits[i] = mark(digits[i], s, last_four[i] >> 16u);
-        }
-        for (std::size_t k = 0; k != count_2; ++k)
-        {
-            const std::size_t i = kind_2[k];
-            const std::uint64_t s = token_digits_2(digits[i], offset[i], radius[i], last_four[i] & 0xFFu, (last_four[i] >> 8u) & 0xFFu);
-            batch.digits[i] = mark(digits[i], s, last_four[i] >> 16u);
-        }
-        return batch.count != 0 && batch.at[0] == n;
-    }
-
-    /*!
-    @brief a float token of 16 or 17 digits as dump() writes it, at w (64 bytes
-    of room), from the digits of its double (see locate_token()) whose
-    rounding carried beyond the token's last four digits
-
-    (like 43.42027299999999 -> 43.420273): the digits go to the writer of
-    Zmij's results as its 15 or 16 leading digits and one more, which omits
-    the trailing zeros.
-    */
-    NLOHMANN_VIEW_NOINLINE char* write_carried_digits(char* w, const node& n, std::uint64_t digits) const noexcept
-    {
-        const int exponent = -static_cast<int>(n.extra >> 8u);
-        const std::uint64_t leading = digits / 10;
-        const auto last = static_cast<unsigned char>(digits - (leading * 10));
-        const bool seventeen = digits >= 10000000000000000u;
-        *w = '-';
-        w += m_doc.src[n.off] == '-' ? 1 : 0;
-        return ::nlohmann::detail::dtoa_impl::write_shortest(w, seventeen ? ::nlohmann::detail::zmij::shortest_decimal{leading, exponent, last, last != 0}
-                : ::nlohmann::detail::zmij::shortest_decimal{digits, exponent - 1, 0, false});
-    }
-
-    /// the digits s of a token with the digits w and the last four digits
-    /// r4, as the batch keeps them: the new last four digits + 1 (only those
-    /// are written over the token's), or s marked if it differs from w beyond
-    /// them (0 stays for a float the batch does not write)
-    static std::uint64_t mark(std::uint64_t w, std::uint64_t s, std::uint64_t r4) noexcept
-    {
-        const std::uint64_t last = s - (w - r4);
-        return last < 10000 ? last + 1 : s | carried();
-    }
-
-    /*!
-    @brief a float token of 16 or 17 digits as dump() writes it, at w (64 bytes
-    of room): the token with its last four digits replaced by digits - 1 (see
-    mark(); the others are the token's), and without trailing zeros after the
-    point, but one
-    */
-    char* write_token_digits(char* w, const char* src, const node& n, std::uint64_t digits)
-    {
-        const char* const token = src + n.off;
-        std::memcpy(w, token, 32);
-        char* const end = w + n.len;
-        const std::uint32_t last = four_digit_text(static_cast<std::uint32_t>(digits - 1));
-        std::memcpy(end - 4, &last, 4);
-        // (the token's text before the new digits, from the source: reading
-        // the output back would wait for the stores)
-        std::uint64_t before = 0;
-        std::memcpy(&before, token + n.len - 12, 8);
-        const std::uint32_t last_zeros = last ^ 0x30303030u;
-        const std::uint64_t before_zeros = before ^ 0x3030303030303030u;
-        if (NLOHMANN_VIEW_UNLIKELY(last_zeros == 0 && before_zeros == 0))
-        {
-            return write_double_at(w, n); // (twelve or more trailing zeros)
-        }
-        const unsigned zeros = last_zeros != 0 ? static_cast<unsigned>(count_leading_zeros(last_zeros)) / 8 - 4
-                               : 4 + (static_cast<unsigned>(count_leading_zeros(before_zeros)) / 8);
-        const unsigned frac_digits = n.extra >> 8u;
-        return end - (zeros < frac_digits ? zeros : frac_digits - 1);
-    }
-#endif
-
     /// a float node as dump() writes it
     void write_float_node(const node& n)
     {

include/nlohmann/json.hpp

@@ -157,7 +157,7 @@ class basic_json // NOLINT(cppcoreguidelines-special-member-functions,hicpp-spec
     friend class ::nlohmann::detail::iter_impl;
     template<typename BasicJsonType, typename CharType, typename OutputSinkType>
     friend class ::nlohmann::detail::binary_writer;
-    template<typename BasicJsonType, typename InputType, typename SAX>
+    template<typename BasicJsonType, typename InputType, typename SAX, bool AllowRecovery>
     friend class ::nlohmann::detail::binary_reader;
     template<typename BasicJsonType, typename InputAdapterType, bool Collect>
     friend class ::nlohmann::detail::json_sax_dom_parser;
@@ -3133,7 +3133,8 @@ public:
         {
             return parser(std::move(ia), nullptr, true, ignore_comments, ignore_trailing_commas).sax_parse(sax, strict);
         }
-        return detail::binary_reader<basic_json, InputAdapterType, SAX>(std::move(ia), format).sax_parse(sax, strict, tag_handler);
+        // the SAX parser may recover from errors (see parse_error())
+        return detail::binary_reader<basic_json, InputAdapterType, SAX, true>(std::move(ia), format).sax_parse(sax, strict, tag_handler);
     }
 
     /// serialize @a j into a new vector; @a write receives the vector-backed writer

single_include/nlohmann/json_view.hpp

@@ -2938,11 +2938,21 @@ struct throwing_sax
 
     /// (a template, as in json_sax_dom_parser: the parser passes the concrete
     /// exception type, which is thrown as is)
+    ///
+    /// It is not [[noreturn]]: sax_parse() recovers from an error if
+    /// parse_error() returns true, and MSVC reports that code as unreachable
+    /// (C4702) for a handler that never returns. The flag hides that from it.
     template<class Exception>
-    [[noreturn]] bool parse_error(std::size_t /*unused*/, const std::string& /*unused*/, const Exception& ex)
+    bool parse_error(std::size_t /*unused*/, const std::string& /*unused*/, const Exception& ex)
     {
-        NLOHMANN_VIEW_THROW(ex);
+        if (throwing)
+        {
+            NLOHMANN_VIEW_THROW(ex);
+        }
+        return false; // LCOV_EXCL_LINE
     }
+
+    bool throwing = true;
 };
 
 /*!
@@ -4605,8 +4615,6 @@ NLOHMANN_JSON_NAMESPACE_END
 
 // #include <nlohmann/detail/view/scan.hpp>
 
-// #include <nlohmann/detail/view/simd.hpp>
-
 
 NLOHMANN_JSON_NAMESPACE_BEGIN
 namespace detail
@@ -4691,56 +4699,12 @@ NLOHMANN_VIEW_NOINLINE FloatType float_value(const char* first, const node& n)
     return convert_float<FloatType>(first, last, dot, mantissa_end);
 }
 
-#if NLOHMANN_VIEW_VECTOR
-/*!
-@brief the value of the first 16 digits of a run of digits with a point after
-int_digits of them (or none, if int_digits >= 16)
-
-The 17 bytes at p are read at once: the bytes after the point move down by one,
-then pairs, quadruples, and octets of digits are combined by multiplications
-(after simdjson). The bytes [p, p + 32) must be readable. Bytes that are not
-digits (only in a damaged image loaded with image_check::bounds) give a wrong
-value, but no undefined behavior.
-*/
-NLOHMANN_VIEW_ALWAYS_INLINE std::uint64_t sixteen_digits(const unsigned char* p, unsigned int_digits) noexcept
-{
-#if NLOHMANN_VIEW_NEON
-    const uint8x16_t low = vld1q_u8(p);
-    const uint8x16_t index = vcombine_u8(vcreate_u8(0x0706050403020100u), vcreate_u8(0x0F0E0D0C0B0A0908u));
-    const uint8x16_t before_point = vcltq_u8(index, vdupq_n_u8(static_cast<std::uint8_t>(int_digits)));
-    const uint8x16_t digits = vsubq_u8(vbslq_u8(before_point, low, vextq_u8(low, vld1q_u8(p + 16), 1)), vdupq_n_u8('0'));
-    const uint16x8_t twos = vpaddlq_u8(vmulq_u8(digits, vreinterpretq_u8_u16(vdupq_n_u16(0x010A))));
-    const uint32x4_t fours = vpaddlq_u16(vmulq_u16(twos, vreinterpretq_u16_u32(vdupq_n_u32(0x00010064))));
-    const uint64x2_t eights = vpaddlq_u32(vmulq_u32(fours, vreinterpretq_u32_u64(vdupq_n_u64(0x0000000100002710u))));
-    return (vgetq_lane_u64(eights, 0) * 100000000u) + vgetq_lane_u64(eights, 1);
-#else
-    // NOLINTBEGIN(portability-simd-intrinsics)
-    const __m128i low = _mm_loadu_si128(static_cast<const __m128i*>(static_cast<const void*>(p))); // NOLINT(bugprone-casting-through-void)
-    const __m128i high = _mm_loadu_si128(static_cast<const __m128i*>(static_cast<const void*>(p + 16))); // NOLINT(bugprone-casting-through-void)
-    const __m128i before_point = _mm_cmpgt_epi8(_mm_set1_epi8(static_cast<char>(int_digits)), _mm_setr_epi8(0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15));
-    const __m128i moved = _mm_or_si128(_mm_srli_si128(low, 1), _mm_slli_si128(high, 15));
-    const __m128i digits = _mm_sub_epi8(_mm_or_si128(_mm_and_si128(before_point, low), _mm_andnot_si128(before_point, moved)), _mm_set1_epi8('0'));
-    // (16-bit lanes: SSE2 has no multiply-add of bytes)
-    const __m128i tens = _mm_setr_epi16(10, 1, 10, 1, 10, 1, 10, 1);
-    const __m128i twos = _mm_packs_epi32(_mm_madd_epi16(_mm_unpacklo_epi8(digits, _mm_setzero_si128()), tens),
-                                         _mm_madd_epi16(_mm_unpackhi_epi8(digits, _mm_setzero_si128()), tens));
-    const __m128i fours = _mm_madd_epi16(twos, _mm_setr_epi16(100, 1, 100, 1, 100, 1, 100, 1));
-    const __m128i eights = _mm_madd_epi16(_mm_packs_epi32(fours, fours), _mm_setr_epi16(10000, 1, 10000, 1, 10000, 1, 10000, 1));
-    const auto upper = static_cast<std::uint32_t>(_mm_cvtsi128_si32(eights));
-    const auto lower = static_cast<std::uint32_t>(_mm_cvtsi128_si32(_mm_srli_si128(eights, 4)));
-    // NOLINTEND(portability-simd-intrinsics)
-    return (std::uint64_t{upper} * 100000000u) + lower;
-#endif
-}
-#endif
-
 /*!
 @brief the digits of a float token with at most 19 digits, from its layout
 
 The digit layout recorded while parsing says where the integer digits, the
 fraction digits, and the exponent are, so the digits are read eight at a
-time without scanning (with SIMD, a token of 16 to 19 digits sixteen at a
-time).
+time without scanning.
 
 @param[in] p  first character of the token
 @param[in] e  end of the token
@@ -4750,34 +4714,15 @@ NLOHMANN_VIEW_ALWAYS_INLINE float_significand layout_decimal(const unsigned char
 {
     const bool negative = *p == '-';
     p += negative ? 1 : 0;
-    std::uint64_t w = 0;
+    std::uint64_t w = parse_upto19(p, int_digits, limit);
+    p += int_digits;
     std::int64_t q = 0;
-#if NLOHMANN_VIEW_VECTOR
-    const unsigned count = int_digits + frac_digits;
-    if (count >= 16 && limit - p >= 32)
+    if (frac_digits != 0)
     {
-        // (the point, if any, follows the integer digits; the digits after
-        // the first 16 are at most three)
-        w = sixteen_digits(p, int_digits);
-        for (unsigned i = 16; i < count; ++i)
-        {
-            w = (w * 10) + static_cast<std::uint64_t>(p[i + (i >= int_digits ? 1 : 0)] - '0');
-        }
-        p += count + (frac_digits != 0 ? 1 : 0);
+        w = (w * int_pow10(frac_digits)) + parse_upto19(p + 1, frac_digits, limit);
+        p += 1 + frac_digits;
         q = -static_cast<std::int64_t>(frac_digits);
     }
-    else
-#endif
-    {
-        w = parse_upto19(p, int_digits, limit);
-        p += int_digits;
-        if (frac_digits != 0)
-        {
-            w = (w * int_pow10(frac_digits)) + parse_upto19(p + 1, frac_digits, limit);
-            p += 1 + frac_digits;
-            q = -static_cast<std::int64_t>(frac_digits);
-        }
-    }
     if (p != e)
     {
         // [eE][+-]digits; huge exponents saturate (the parser rejected
@@ -4812,22 +4757,12 @@ NLOHMANN_VIEW_ALWAYS_INLINE float_significand layout_decimal(const unsigned char
 The result is correctly rounded by the lexer's conversion
 (detail::decimal_to_float(): Clinger's fast path where both operands are
 exact, else the Eisel-Lemire algorithm, which needs no fallback for up to 19
-digits), so it is the value parse() produces. Tokens of 16 or more digits go
-to Eisel-Lemire directly: the fast path takes few of them (none of 17 digits),
-and its check is a branch that data like GeoJSON coordinates mispredicts.
+digits), so it is the value parse() produces.
 */
 template<typename FloatType>
 NLOHMANN_VIEW_ALWAYS_INLINE FloatType layout_float(const unsigned char* p, const unsigned char* e, unsigned int_digits, unsigned frac_digits, const unsigned char* limit) noexcept
 {
-    const float_significand d = layout_decimal(p, e, int_digits, frac_digits, limit);
-    if (int_digits + frac_digits >= 16)
-    {
-        using format = ieee_binary_format<std::numeric_limits<FloatType>::digits>;
-        native_float_t<FloatType> value{};
-        float_from_bits(eisel_lemire<format>(d.exponent, d.w) | (d.negative ? (std::uint64_t{1} << format::sign_bit()) : 0u), value);
-        return value;
-    }
-    return decimal_to_float<FloatType>(d);
+    return decimal_to_float<FloatType>(layout_decimal(p, e, int_digits, frac_digits, limit));
 }
 
 /// the value of a float set by an edit: its token (the shortest round-trip
@@ -4846,80 +4781,6 @@ NLOHMANN_VIEW_NOINLINE FloatType edited_float(const char* token, const node& n)
     return float_value<FloatType>(token, n);
 }
 
-#if NLOHMANN_VIEW_VECTOR
-/// eisel_lemire() for the token w * 10^q of coordinate_float()
-template<typename Format>
-NLOHMANN_VIEW_ALWAYS_INLINE std::uint64_t coordinate_bits(std::int64_t q, std::uint64_t w, Format /*format*/) noexcept
-{
-    return eisel_lemire<Format>(q, w);
-}
-
-/*!
-@brief (binary64) eisel_lemire() without the cases that tokens of 16 or 17
-digits with 5 or more after the point cannot reach: a value of 0, a
-subnormal, an infinity, or a tie (q < -4)
-*/
-NLOHMANN_VIEW_ALWAYS_INLINE std::uint64_t coordinate_bits(std::int64_t q, std::uint64_t w, ieee_binary_format<53> format) noexcept
-{
-    if (NLOHMANN_VIEW_UNLIKELY(w == 0 || q > -5))
-    {
-        return eisel_lemire<ieee_binary_format<53>>(q, w);
-    }
-    static_cast<void>(format);
-    const int lz = count_leading_zeros(w);
-    const std::uint64_t normalized = w << static_cast<unsigned>(lz);
-    const auto index = static_cast<std::size_t>(2 * (q - pow5_128_smallest_power));
-    const uint128_parts product = full_multiplication(normalized, pow5_128()[index]);
-    std::uint64_t high = product.high;
-    if (NLOHMANN_VIEW_UNLIKELY((high & 0x1FFu) == 0x1FFu))
-    {
-        // (the lower bits may carry into the result)
-        const std::uint64_t second = full_multiplication(normalized, pow5_128()[index + 1]).high;
-        high += product.low + second < product.low ? 1u : 0u;
-    }
-    const auto upperbit = static_cast<unsigned>(high >> 63u);
-    std::uint64_t mantissa = high >> (upperbit + 9);
-    std::uint64_t power2 = static_cast<std::uint64_t>(((217706 * q) >> 16) + 63 + static_cast<std::int64_t>(upperbit) - lz + 1023);
-    mantissa = (mantissa + (mantissa & 1u)) >> 1u;
-    const bool carry = mantissa >= (std::uint64_t{2} << 52u);
-    mantissa = carry ? (std::uint64_t{1} << 52u) : mantissa;
-    power2 += carry ? 1u : 0u;
-    return (mantissa & ~(std::uint64_t{1} << 52u)) | (power2 << 52u);
-}
-
-/*!
-@brief (binary32 and binary64) the value of a float token of the source with
-16 or 17 digits, a point, and no exponent, the shape of most coordinates
-written with full precision; whether the token of n has this shape
-
-These tokens skip the general layout of layout_decimal(): no digits beyond
-the 17th, no exponent (and Eisel-Lemire directly, as in layout_float()).
-*/
-template<typename FloatType>
-NLOHMANN_VIEW_ALWAYS_INLINE bool coordinate_float(const document_data& d, const node& n, FloatType& value) noexcept
-{
-    const unsigned int_digits = n.extra & 0xFFu;
-    const unsigned frac_digits = n.extra >> 8u;
-    const unsigned count = int_digits + frac_digits;
-    const auto* const first = reinterpret_cast<const unsigned char*>(d.src + n.off); // NOLINT(cppcoreguidelines-pro-type-reinterpret-cast)
-    const unsigned negative = first[0] == '-' ? 1u : 0u;
-    if (count - 16u > 1u || frac_digits == 0 || n.len != negative + count + 1 || d.size - n.off < 33)
-    {
-        return false;
-    }
-    const unsigned char* const digits = first + negative;
-    const std::uint64_t w = sixteen_digits(digits, int_digits);
-    // (the 17th digit follows the point)
-    const std::uint64_t w17 = (w * 10) + static_cast<std::uint64_t>(digits[17] - '0');
-    using format = ieee_binary_format<std::numeric_limits<FloatType>::digits>;
-    native_float_t<FloatType> result{};
-    float_from_bits(coordinate_bits(-static_cast<std::int64_t>(frac_digits), count == 17 ? w17 : w, format{}) | (static_cast<std::uint64_t>(negative) << format::sign_bit()), result);
-    value = result;
-    return true;
-}
-
-#endif
-
 /// the value of the float token of a node, as parse() converts it; floats and
 /// doubles with at most 19 digits are converted from the digit layout
 template<typename FloatType>
@@ -4935,13 +4796,6 @@ FloatType float_value(const document_data& d, const node& n)
 template<typename FloatType>
 FloatType float_value(const document_data& d, const node& n, std::true_type /*binary32 or binary64*/)
 {
-#if NLOHMANN_VIEW_VECTOR
-    FloatType value {};
-    if (coordinate_float(d, n, value))
-    {
-        return value;
-    }
-#endif
     const unsigned int_digits = n.extra & 0xFFu;
     const unsigned frac_digits = n.extra >> 8u;
     if (NLOHMANN_VIEW_LIKELY(int_digits + frac_digits <= 19)) // (255 marks "many")
@@ -6513,9 +6367,7 @@ NLOHMANN_JSON_NAMESPACE_END
 #include <cstddef> // size_t
 #include <cstdint> // uint8_t, uint32_t
 #include <cstring> // memcpy, memset
-#include <functional> // less
 #include <limits> // numeric_limits
-#include <memory> // unique_ptr
 #include <type_traits> // integral_constant
 #include <vector> // vector
 
@@ -6890,127 +6742,6 @@ struct depth_counter<false>
     }
 };
 
-/*!
-@brief where the double x nearest to a token w * 10^q lies, for 10^15 <= w <
-10^17 and -17 <= q <= -1 (Eisel-Lemire, then a fixed-point position)
-
-dump() writes the shortest decimal in the rounding interval of x (the closest
-one if there are several, see zmij::to_shortest()). Its digits are those of a
-multiple of 10^g of the token's last digit, with g = dec_exp + 1 - q for x's
-decimal exponent dec_exp: the multiple of 10^g nearest to x if it lies in the
-interval (at most one does), else the multiple of 10^(g - 1) nearest to x
-(see token_digits_1() and token_digits_2()). For g <= 0, the token itself is
-in the interval.
-
-The position is exact in a 64-bit fixed-point number with 56 bits after the
-point (of the token's last digit): x / 10^q = 2c * 5^-q / 2^t for x = c * 2^e2
-and t = q + 1 - e2, with at most 56 bits after the point here, and |x / 10^q -
-w| is at most half the spacing of the doubles, below 12 units. So only the low
-64 bits of 2c * 5^-q are needed. c and e2 come from Eisel-Lemire's product
-directly (as eisel_lemire() rounds it: no tie is possible for q < -4, and its
-second product is added where it can carry into the result).
-
-@param[out] offset  x / 10^q - w + 32, with 56 bits after the point
-@param[out] radius  half the spacing of the doubles at x, likewise (+ 1 if
-                    the interval is closed)
-@return g; 3 if the digits are not decided here (a power of two, whose
-        interval is not symmetric, or out of range)
-*/
-NLOHMANN_VIEW_ALWAYS_INLINE int locate_token(std::uint64_t w, int q, std::uint64_t& offset, std::uint64_t& radius) noexcept
-{
-    const int lz = count_leading_zeros(w);
-    const std::uint64_t normalized = w << static_cast<unsigned>(lz);
-    const auto index = static_cast<std::size_t>(2 * (q - pow5_128_smallest_power));
-    const uint128_parts product = full_multiplication(normalized, pow5_128()[index]);
-    std::uint64_t high = product.high;
-    if (NLOHMANN_VIEW_UNLIKELY((high & 0x1FFu) == 0x1FFu))
-    {
-        // (as eisel_lemire(): the lower bits may carry into the result)
-        const std::uint64_t second = full_multiplication(normalized, pow5_128()[index + 1]).high;
-        high += product.low + second < product.low ? 1u : 0u;
-    }
-    const auto upperbit = static_cast<unsigned>(high >> 63u);
-    const std::uint64_t mantissa = high >> (upperbit + 9);
-    const std::uint64_t c = (mantissa + (mantissa & 1u)) >> 1u;
-    // e2 as eisel_lemire() computes the exponent (log2(10) ~ 217706 / 2^16)
-    const int e2 = static_cast<int>((217706 * static_cast<std::int64_t>(q)) >> 16) + 63 + static_cast<int>(upperbit) - lz - 52;
-    const int t = q + 1 - e2;
-    const int g = ((e2 * 315653) >> 20) + 1 - q; // (dec_exp as in zmij::to_shortest(), for a regular x)
-    static const std::array<std::uint64_t, 17> fives = {{1u, 5u, 25u, 125u, 625u, 3125u, 15625u, 78125u, 390625u, 1953125u, 9765625u, 48828125u, 244140625u, 1220703125u, 6103515625u, 30517578125u, 152587890625u}};
-    const std::uint64_t five = fives[static_cast<std::size_t>(-q)]; // 5^-q
-    const unsigned s = static_cast<unsigned>(56 - t) & 63u;
-    offset = ((2 * c * five) << s) - (w << 56u) + (std::uint64_t{32} << 56u);
-    radius = (five << s) + ((c & 1u) == 0 ? 1u : 0u);
-    // (c = 2^52: a power of two; c = 2^53: rounded up to one. With 8 to 16
-    // fraction digits, as the batch's tokens have, t is in [15, 40] and g at
-    // most 2. x lies halfway between two candidates only if its distance from
-    // the token is a whole unit (g == 2) or a half (g == 1): such tokens are
-    // left to the full conversion)
-    const bool decided = c - ((std::uint64_t{1} << 52u) + 1) < (std::uint64_t{1} << 52u) - 1
-                         && (offset & ((std::uint64_t{1} << 55u) - 1)) != 0;
-    return decided ? g : 3;
-}
-
-/// the distance of a position (see locate_token()) from the biased point 32
-NLOHMANN_VIEW_ALWAYS_INLINE std::uint64_t biased_distance(std::uint64_t position) noexcept
-{
-    constexpr std::uint64_t center = std::uint64_t{32} << 56u;
-    return position >= center ? position - center : center - position;
-}
-
-/*!
-@brief (g == 1) the digits dump() writes for a token, from the position of its
-double (see locate_token()): the multiple of 10 nearest to x if it is in the
-interval, else w rounded to x (an exact tie between two candidates, like
-754570942.33984373, whose double is 754570942.33984375, is excluded by
-locate_token())
-
-With |x - w| at most the radius, which is below half the spacing of the
-candidates, the multiple of 10 nearest to x is w - r10 or the one above it.
-*/
-NLOHMANN_VIEW_ALWAYS_INLINE std::uint64_t token_digits_1(std::uint64_t w, std::uint64_t offset, std::uint64_t radius, std::uint64_t r10) noexcept
-{
-    constexpr std::uint64_t one = std::uint64_t{1} << 56u;
-    const std::uint64_t from = offset + (r10 * one); // x - (w - r10) (+ 32)
-    const bool up = from > 37 * one;                  // (nearer to the multiple above)
-    const bool shorter = biased_distance(from - (up ? 10 * one : 0)) < radius;
-    const std::uint64_t a = w - r10 + (up ? 10u : 0u);
-    const std::uint64_t b = w + ((offset + (one / 2)) >> 56u) - 32;                // w rounded to x
-    return shorter ? a : b;
-}
-
-/*!
-@brief (g == 2) the digits dump() writes for a token, as token_digits_1(): the
-multiple of 100 nearest to x if it is in the interval, else the multiple of 10
-nearest to x
-*/
-NLOHMANN_VIEW_ALWAYS_INLINE std::uint64_t token_digits_2(std::uint64_t w, std::uint64_t offset, std::uint64_t radius, std::uint64_t r10, std::uint64_t r100) noexcept
-{
-    constexpr std::uint64_t one = std::uint64_t{1} << 56u;
-    const std::uint64_t from = offset + (r100 * one);
-    const bool up = from > 82 * one;
-    const bool shorter = biased_distance(from - (up ? 100 * one : 0)) < radius;
-    const std::uint64_t a = w - r100 + (up ? 100u : 0u);
-    // x - (w - r10), in [-12, 21]: the multiple of 10 nearest to x is that one,
-    // or one or two above, or one below
-    const std::uint64_t y = offset + (r10 * one);
-    const std::uint64_t steps = (y > 37 * one ? 1u : 0u) + (y > 47 * one ? 1u : 0u);
-    const bool down = y < 27 * one;
-    const std::uint64_t b = w - r10 + (10 * steps) - (down ? 10u : 0u);
-    return shorter ? a : b;
-}
-
-/// the four ASCII digits of x < 10^4, the first in the lowest byte (as
-/// little-endian memory holds them)
-NLOHMANN_VIEW_ALWAYS_INLINE std::uint32_t four_digit_text(std::uint32_t x) noexcept
-{
-    const std::uint32_t ab = (x * 5243u) >> 19u; // x / 100
-    std::uint32_t v = ab | ((x - (ab * 100u)) << 16u);
-    const std::uint32_t a = ((v * 103u) >> 10u) & 0x000F000Fu; // / 10 per 16-bit lane
-    v = a | ((v - (a * 10u)) << 8u);
-    return v + 0x30303030u;
-}
-
 /// how the view's dump() writes a value
 struct dump_style
 {
@@ -7338,7 +7069,6 @@ class view_serializer
         std::array<frame, 32> buffer; // NOLINT(cppcoreguidelines-pro-type-member-init,hicpp-member-init): written before read
         std::vector<frame> heap;
         small_stack<frame> stack(buffer.data(), buffer.size(), heap);
-        float_batch batch;
         const char* const src = m_doc.src;
         const char* const src_end = src + m_doc.size;
         char* w = m_out.cursor();
@@ -7478,7 +7208,7 @@ class view_serializer
                     else if (enabled(fast_double::value))
                     {
                         room(64);
-                        w = write_double_ahead(w, src, n, batch);
+                        w = write_double_at(w, *n);
                     }
                     else
                     {
@@ -7503,20 +7233,6 @@ class view_serializer
                 case value_t::array:
                 {
                     const bool object = n->kind == static_cast<std::uint8_t>(value_t::object);
-                    // (read-only documents, compact output) a pair of floats,
-                    // like the coordinates of GeoJSON, without a frame
-                    if (!Editable && !Pretty && !SourceNumbers && enabled(fast_double::value) && n->next == 3 && n->len == 2
-                            && n[1].kind == static_cast<std::uint8_t>(value_t::number_float) && n[2].kind == static_cast<std::uint8_t>(value_t::number_float))
-                    {
-                        room(2 * 64);
-                        *w++ = '[';
-                        w = write_double_ahead(w, src, n + 1, batch);
-                        *w++ = ',';
-                        w = write_double_ahead(w, src, n + 2, batch);
-                        *w++ = ']';
-                        n += 2; // (the element before the next node)
-                        break;
-                    }
                     room(8);
                     if (n->len == 0)
                     {
@@ -7626,247 +7342,6 @@ class view_serializer
         }
     }
 
-    /// (float_batch) the mark of digits whose rounding carried beyond the
-    /// token's last four (see mark())
-    static constexpr std::uint64_t carried() noexcept
-    {
-        return std::uint64_t{1} << 63u;
-    }
-
-    /// floats of 16 or 17 digits ahead of the walk, converted together (see
-    /// convert_ahead()), with their digits as mark() keeps them
-    struct float_batch
-    {
-        static constexpr std::size_t capacity() noexcept
-        {
-            return 256;
-        }
-        /// (on the heap, allocated by the first conversion: a frame of more
-        /// than a page made every dump of a small document slower)
-        struct entries
-        {
-            std::array<const node*, 256> at; // NOLINT(cppcoreguidelines-pro-type-member-init,hicpp-member-init): written before read
-            std::array<std::uint64_t, 256> digits; // NOLINT(cppcoreguidelines-pro-type-member-init,hicpp-member-init): written before read
-        };
-        std::unique_ptr<entries> storage{};
-        const node** at = nullptr;
-        std::uint64_t* digits = nullptr;
-        std::size_t count = 0;
-        std::size_t next = 0;
-        const node* scanned = nullptr; ///< where the last scan of the node array ended
-    };
-
-    /// (doubles) the float at n as dump() writes it, at w (64 bytes of room):
-    /// from the batch if it holds n
-    NLOHMANN_VIEW_ALWAYS_INLINE char* write_double_ahead(char* w, const char* src, const node* n, float_batch& batch)
-    {
-#if NLOHMANN_VIEW_VECTOR && NLOHMANN_VIEW_LITTLE_ENDIAN
-        // (digits 0: not written by the batch; carried(): rounding carried
-        // beyond the last four digits)
-        if (batch.next != batch.count && batch.at[batch.next] == n && batch.digits[batch.next] - 1 < carried() - 1)
-        {
-            return write_token_digits(w, src, *n, batch.digits[batch.next++]);
-        }
-        return write_double_beside(w, n, batch);
-#else
-        static_cast<void>(src);
-        static_cast<void>(batch);
-        return write_double_at(w, *n);
-#endif
-    }
-
-#if NLOHMANN_VIEW_VECTOR && NLOHMANN_VIEW_LITTLE_ENDIAN
-    /// write_double_ahead() for a float that is not the batch's next one (out
-    /// of line: the walk is faster with less code in its loop)
-    NLOHMANN_VIEW_NOINLINE char* write_double_beside(char* w, const node* n, float_batch& batch)
-    {
-        // (an editable document's walk can skip nodes of the array)
-        const std::less<const node*> before;
-        while (batch.next != batch.count && before(batch.at[batch.next], n))
-        {
-            ++batch.next;
-        }
-        if ((batch.next != batch.count && batch.at[batch.next] == n) || convert_ahead(n, batch))
-        {
-            const std::uint64_t digits = batch.digits[batch.next++];
-            if ((digits & carried()) != 0)
-            {
-                return write_carried_digits(w, *n, digits & ~carried());
-            }
-            if (digits != 0)
-            {
-                return write_token_digits(w, m_doc.src, *n, digits);
-            }
-        }
-        return write_double_at(w, *n);
-    }
-
-    /*!
-    @brief convert the floats of 16 or 17 digits among the next nodes of the
-    array at once; whether n is the first of them
-
-    Such tokens (like the coordinates of GeoJSON, written with 17 digits) are
-    written as the token with its last four digits replaced
-    (write_token_digits()). Each step (the digits, locate_token(),
-    token_digits_1/2()) is a long chain of dependent operations; one loop per
-    step over many floats lets the processor overlap the floats, which it does
-    not across the steps of one float (on Apple M1, one loop for the digits and
-    locate_token() made the dump of canada.json 12 % slower). Each node of the array
-    is scanned once; a node that the walk does not reach in order (in an
-    editable document) is written without the batch.
-    */
-    NLOHMANN_VIEW_NOINLINE bool convert_ahead(const node* n, float_batch& batch)
-    {
-        const std::less<const node*> before;
-        const node* const last = m_doc.nodes + m_doc.node_count;
-        if (before(n, batch.scanned) || before(n, m_doc.nodes) || !before(n, last))
-        {
-            return false;
-        }
-        const char* const src = m_doc.src;
-        const char* const src_end = src + m_doc.size;
-        const node* const stop = last - n > 1024 ? n + 1024 : last;
-        batch.count = 0;
-        batch.next = 0;
-        // the digits of each token (and its last four from the text)
-        std::array<std::uint64_t, 256> digits;     // NOLINT(cppcoreguidelines-pro-type-member-init,hicpp-member-init): written before read
-        std::array<int, 256> exponent;             // NOLINT(cppcoreguidelines-pro-type-member-init,hicpp-member-init): written before read
-        std::array<std::uint32_t, 256> last_four;  // NOLINT(cppcoreguidelines-pro-type-member-init,hicpp-member-init): written before read
-        const node* p = n;
-        if (!batch.storage)
-        {
-            batch.storage.reset(new typename float_batch::entries); // NOLINT(cppcoreguidelines-owning-memory,modernize-make-unique): C++11
-            batch.at = batch.storage->at.data();
-            batch.digits = batch.storage->digits.data();
-        }
-        for (; p != stop && batch.count != float_batch::capacity(); ++p)
-        {
-            // a token of the source: 16 or 17 digits, at least 8 after the
-            // point, no exponent, 32 bytes to read
-            const unsigned frac_digits = p->extra >> 8u;
-            const unsigned int_digits = p->extra & 0xFFu;
-            const unsigned count = int_digits + frac_digits;
-            if (p->kind == static_cast<std::uint8_t>(value_t::number_float) && (p->flags & node_flags::storage) == 0
-                    && count - 16u <= 1u && frac_digits >= 8 && src_end - (src + p->off) >= 33)
-            {
-                const auto* const token = reinterpret_cast<const unsigned char*>(src + p->off); // NOLINT(cppcoreguidelines-pro-type-reinterpret-cast)
-                const unsigned sign = token[0] == '-' ? 1 : 0;
-                const unsigned char* const first = token + sign;
-                const std::uint64_t w = sixteen_digits(first, int_digits);
-                // (the 17th digit follows the point; a token with an exponent
-                // gets digits outside the range of locate_token(), and is not
-                // written by the batch. A token 0.xxx with 16 significant
-                // digits is: its double is written as "0." and its digits too)
-                const std::uint64_t w17 = (w * 10) + static_cast<std::uint64_t>(first[17] - '0');
-                const std::size_t i = batch.count++;
-                batch.at[i] = p;
-                digits[i] = p->len != sign + count + 1 ? 0 : (count == 17 ? w17 : w);
-                exponent[i] = -static_cast<int>(frac_digits);
-                // (byte 0: the last digit, byte 1: the last two, bits 16-29: the last four)
-                std::uint32_t four = 0;
-                std::memcpy(&four, token + p->len - 4, 4);
-                four -= 0x30303030u;
-                const std::uint32_t pairs = ((four & 0x0F0F0F0Fu) * 2561u) >> 8u; // (byte 0, 2: the pairs)
-                last_four[i] = (four >> 24u) | ((pairs >> 16u & 0xFFu) << 8u) | ((((pairs & 0x00FF00FFu) * 6553601u) >> 16u) << 16u);
-            }
-        }
-        batch.scanned = p;
-        // the position of each double, and the floats of either kind
-        std::array<std::uint64_t, 256> offset; // NOLINT(cppcoreguidelines-pro-type-member-init,hicpp-member-init): written before read
-        std::array<std::uint64_t, 256> radius; // NOLINT(cppcoreguidelines-pro-type-member-init,hicpp-member-init): written before read
-        std::array<std::uint8_t, 256> kind_1;  // NOLINT(cppcoreguidelines-pro-type-member-init,hicpp-member-init): written before read
-        std::array<std::uint8_t, 256> kind_2;  // NOLINT(cppcoreguidelines-pro-type-member-init,hicpp-member-init): written before read
-        std::size_t count_1 = 0;
-        std::size_t count_2 = 0;
-        for (std::size_t i = 0; i != batch.count; ++i)
-        {
-            const std::uint64_t w = digits[i];
-            // (the range excludes the values of bytes that are not digits, only
-            // in a damaged image loaded with image_check::bounds)
-            const int g = w - 1000000000000000u < 99000000000000000u ? locate_token(w, exponent[i], offset[i], radius[i]) : 3;
-            batch.digits[i] = g <= 0 ? (last_four[i] >> 16u) + 1 : 0;
-            kind_1[count_1] = static_cast<std::uint8_t>(i);
-            count_1 += g == 1 ? 1 : 0;
-            kind_2[count_2] = static_cast<std::uint8_t>(i);
-            count_2 += g == 2 ? 1 : 0;
-        }
-        // the digits to write, in one loop per kind (see mark())
-        for (std::size_t k = 0; k != count_1; ++k)
-        {
-            const std::size_t i = kind_1[k];
-            const std::uint64_t s = token_digits_1(digits[i], offset[i], radius[i], last_four[i] & 0xFFu);
-            batch.digits[i] = mark(digits[i], s, last_four[i] >> 16u);
-        }
-        for (std::size_t k = 0; k != count_2; ++k)
-        {
-            const std::size_t i = kind_2[k];
-            const std::uint64_t s = token_digits_2(digits[i], offset[i], radius[i], last_four[i] & 0xFFu, (last_four[i] >> 8u) & 0xFFu);
-            batch.digits[i] = mark(digits[i], s, last_four[i] >> 16u);
-        }
-        return batch.count != 0 && batch.at[0] == n;
-    }
-
-    /*!
-    @brief a float token of 16 or 17 digits as dump() writes it, at w (64 bytes
-    of room), from the digits of its double (see locate_token()) whose
-    rounding carried beyond the token's last four digits
-
-    (like 43.42027299999999 -> 43.420273): the digits go to the writer of
-    Zmij's results as its 15 or 16 leading digits and one more, which omits
-    the trailing zeros.
-    */
-    NLOHMANN_VIEW_NOINLINE char* write_carried_digits(char* w, const node& n, std::uint64_t digits) const noexcept
-    {
-        const int exponent = -static_cast<int>(n.extra >> 8u);
-        const std::uint64_t leading = digits / 10;
-        const auto last = static_cast<unsigned char>(digits - (leading * 10));
-        const bool seventeen = digits >= 10000000000000000u;
-        *w = '-';
-        w += m_doc.src[n.off] == '-' ? 1 : 0;
-        return ::nlohmann::detail::dtoa_impl::write_shortest(w, seventeen ? ::nlohmann::detail::zmij::shortest_decimal{leading, exponent, last, last != 0}
-                : ::nlohmann::detail::zmij::shortest_decimal{digits, exponent - 1, 0, false});
-    }
-
-    /// the digits s of a token with the digits w and the last four digits
-    /// r4, as the batch keeps them: the new last four digits + 1 (only those
-    /// are written over the token's), or s marked if it differs from w beyond
-    /// them (0 stays for a float the batch does not write)
-    static std::uint64_t mark(std::uint64_t w, std::uint64_t s, std::uint64_t r4) noexcept
-    {
-        const std::uint64_t last = s - (w - r4);
-        return last < 10000 ? last + 1 : s | carried();
-    }
-
-    /*!
-    @brief a float token of 16 or 17 digits as dump() writes it, at w (64 bytes
-    of room): the token with its last four digits replaced by digits - 1 (see
-    mark(); the others are the token's), and without trailing zeros after the
-    point, but one
-    */
-    char* write_token_digits(char* w, const char* src, const node& n, std::uint64_t digits)
-    {
-        const char* const token = src + n.off;
-        std::memcpy(w, token, 32);
-        char* const end = w + n.len;
-        const std::uint32_t last = four_digit_text(static_cast<std::uint32_t>(digits - 1));
-        std::memcpy(end - 4, &last, 4);
-        // (the token's text before the new digits, from the source: reading
-        // the output back would wait for the stores)
-        std::uint64_t before = 0;
-        std::memcpy(&before, token + n.len - 12, 8);
-        const std::uint32_t last_zeros = last ^ 0x30303030u;
-        const std::uint64_t before_zeros = before ^ 0x3030303030303030u;
-        if (NLOHMANN_VIEW_UNLIKELY(last_zeros == 0 && before_zeros == 0))
-        {
-            return write_double_at(w, n); // (twelve or more trailing zeros)
-        }
-        const unsigned zeros = last_zeros != 0 ? static_cast<unsigned>(count_leading_zeros(last_zeros)) / 8 - 4
-                               : 4 + (static_cast<unsigned>(count_leading_zeros(before_zeros)) / 8);
-        const unsigned frac_digits = n.extra >> 8u;
-        return end - (zeros < frac_digits ? zeros : frac_digits - 1);
-    }
-#endif
-
     /// a float node as dump() writes it
     void write_float_node(const node& n)
     {

tests/src/fuzzer-parse_bjdata.cpp

@@ -47,6 +47,10 @@ 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.
 
+Furthermore, it reads data with a SAX parser that recovers from every error
+and checks that the events are balanced, that reading ends, and that it
+reports an error exactly when from_bjdata() fails (see #3989).
+
 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
@@ -57,6 +61,7 @@ drivers.
 */
 
 #include "fuzzer_common.hpp"
+#include "fuzzer-recovering_checker.hpp"
 
 namespace
 {
@@ -81,6 +86,9 @@ static bool is_value_stable(const json& lhs, const json& rhs)
 // see http://llvm.org/docs/LibFuzzer.html
 extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size)
 {
+    // recover from all errors, reading from memory and from a stream
+    const bool recovered_without_errors = check_recovering_parse(data, size, json::input_format_t::bjdata).errors == 0;
+
     std::vector<uint8_t> const vec1(data, data + size);
 
     // step 0: parse input without exceptions
@@ -99,6 +107,9 @@ extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size)
         // without exceptions, the same input must give the same value
         assert(!noexcept_threw && !j_noexcept.is_discarded() && same_value(j_noexcept, j1));
 
+        // the recovering parser must not have reported an error either
+        assert(recovered_without_errors);
+
         try
         {
             // step 2.1: round trip without adding size annotations to container types
@@ -132,6 +143,7 @@ extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size)
     {
         // parse errors are ok, because input may be random bytes
         assert(parsed || noexcept_threw || j_noexcept.is_discarded());
+        assert(parsed || !recovered_without_errors);
     }
     catch (const json::type_error&)
     {
@@ -142,6 +154,7 @@ extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size)
     {
         // out of range errors may happen if provided sizes are excessive
         assert(parsed || noexcept_threw || j_noexcept.is_discarded());
+        assert(parsed || !recovered_without_errors);
     }
 
     // step 3: the other input adapters must read the data alike

tests/src/fuzzer-parse_bon8.cpp

@@ -17,6 +17,10 @@ array data, it performs the following steps:
 - j2 = from_bon8(vec)
 - assert(to_bon8(j2) == vec)
 
+Furthermore, it reads data with a SAX parser that recovers from every error
+and checks that the events are balanced, that reading ends, and that it
+reports an error exactly when from_bon8() fails (see #3989).
+
 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
@@ -27,6 +31,7 @@ drivers.
 */
 
 #include "fuzzer_common.hpp"
+#include "fuzzer-recovering_checker.hpp"
 
 namespace
 {
@@ -44,6 +49,9 @@ struct read_bon8
 // see http://llvm.org/docs/LibFuzzer.html
 extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size)
 {
+    // recover from all errors, reading from memory and from a stream
+    const bool recovered_without_errors = check_recovering_parse(data, size, json::input_format_t::bon8).errors == 0;
+
     std::vector<uint8_t> const vec1(data, data + size);
 
     // step 0: parse input without exceptions
@@ -62,6 +70,9 @@ extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size)
         // without exceptions, the same input must give the same value
         assert(!noexcept_threw && !j_noexcept.is_discarded() && same_value(j_noexcept, j1));
 
+        // the recovering parser must not have reported an error either
+        assert(recovered_without_errors);
+
         try
         {
             // step 2: round trip
@@ -83,6 +94,7 @@ extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size)
     {
         // parse errors are ok, because input may be random bytes
         assert(parsed || noexcept_threw || j_noexcept.is_discarded());
+        assert(parsed || !recovered_without_errors);
     }
     catch (const json::type_error&)
     {
@@ -93,6 +105,7 @@ extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size)
     {
         // out of range errors may happen if provided sizes are excessive
         assert(parsed || noexcept_threw || j_noexcept.is_discarded());
+        assert(parsed || !recovered_without_errors);
     }
 
     // step 3: the other input adapters must read the data alike

tests/src/fuzzer-parse_bson.cpp

@@ -17,6 +17,10 @@ array data, it performs the following steps:
 - j2 = from_bson(vec)
 - assert(to_bson(j2) == vec)
 
+Furthermore, it reads data with a SAX parser that recovers from every error
+and checks that the events are balanced, that reading ends, and that it
+reports an error exactly when from_bson() fails (see #3989).
+
 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
@@ -27,6 +31,7 @@ drivers.
 */
 
 #include "fuzzer_common.hpp"
+#include "fuzzer-recovering_checker.hpp"
 
 namespace
 {
@@ -44,6 +49,9 @@ struct read_bson
 // see http://llvm.org/docs/LibFuzzer.html
 extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size)
 {
+    // recover from all errors, reading from memory and from a stream
+    const bool recovered_without_errors = check_recovering_parse(data, size, json::input_format_t::bson).errors == 0;
+
     std::vector<uint8_t> const vec1(data, data + size);
 
     // step 0: parse input without exceptions
@@ -62,6 +70,9 @@ extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size)
         // without exceptions, the same input must give the same value
         assert(!noexcept_threw && !j_noexcept.is_discarded() && same_value(j_noexcept, j1));
 
+        // the recovering parser must not have reported an error either
+        assert(recovered_without_errors);
+
         try
         {
             // step 2: round trip
@@ -83,6 +94,7 @@ extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size)
     {
         // parse errors are ok, because input may be random bytes
         assert(parsed || noexcept_threw || j_noexcept.is_discarded());
+        assert(parsed || !recovered_without_errors);
     }
     catch (const json::type_error&)
     {
@@ -93,6 +105,7 @@ extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size)
     {
         // out of range errors can occur during parsing, too
         assert(parsed || noexcept_threw || j_noexcept.is_discarded());
+        assert(parsed || !recovered_without_errors);
     }
 
     // step 3: the other input adapters must read the data alike

tests/src/fuzzer-parse_cbor.cpp

@@ -17,6 +17,10 @@ array data, it performs the following steps:
 - j2 = from_cbor(vec)
 - assert(to_cbor(j2) == vec)
 
+Furthermore, it reads data with a SAX parser that recovers from every error
+and checks that the events are balanced, that reading ends, and that it
+reports an error exactly when from_cbor() fails (see #3989).
+
 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
@@ -27,6 +31,7 @@ drivers.
 */
 
 #include "fuzzer_common.hpp"
+#include "fuzzer-recovering_checker.hpp"
 
 namespace
 {
@@ -44,6 +49,9 @@ struct read_cbor
 // see http://llvm.org/docs/LibFuzzer.html
 extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size)
 {
+    // recover from all errors, reading from memory and from a stream
+    const bool recovered_without_errors = check_recovering_parse(data, size, json::input_format_t::cbor).errors == 0;
+
     std::vector<uint8_t> const vec1(data, data + size);
 
     // step 0: parse input without exceptions
@@ -62,6 +70,9 @@ extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size)
         // without exceptions, the same input must give the same value
         assert(!noexcept_threw && !j_noexcept.is_discarded() && same_value(j_noexcept, j1));
 
+        // the recovering parser must not have reported an error either
+        assert(recovered_without_errors);
+
         try
         {
             // step 2: round trip
@@ -83,6 +94,7 @@ extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size)
     {
         // parse errors are ok, because input may be random bytes
         assert(parsed || noexcept_threw || j_noexcept.is_discarded());
+        assert(parsed || !recovered_without_errors);
     }
     catch (const json::type_error&)
     {
@@ -93,6 +105,7 @@ extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size)
     {
         // out of range errors can occur during parsing, too
         assert(parsed || noexcept_threw || j_noexcept.is_discarded());
+        assert(parsed || !recovered_without_errors);
     }
 
     // step 3: the other input adapters must read the data alike

tests/src/fuzzer-parse_json.cpp

@@ -24,11 +24,16 @@ array data, it performs the following steps:
 The pointer range takes the bulk fast paths, the other input adapters read byte
 by byte; see check_input_adapters().
 
+Furthermore, it parses data with a SAX parser that recovers from every error
+and checks that the events are balanced, that parsing ends, and that valid
+input is parsed without errors (see #3989).
+
 The provided function `LLVMFuzzerTestOneInput` can be used in different fuzzer
 drivers.
 */
 
 #include "fuzzer_common.hpp"
+#include "fuzzer-recovering_checker.hpp"
 
 namespace
 {
@@ -59,6 +64,13 @@ struct accept_json
 // see http://llvm.org/docs/LibFuzzer.html
 extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size)
 {
+    // recover from all errors, reading from memory and from a stream
+    {
+        const auto checker = check_recovering_parse(data, size, json::input_format_t::json);
+        assert(checker.events <= (4 * size) + 4);
+        assert((checker.errors == 0) == json::accept(data, data + size));
+    }
+
     // step 0: parse input without exceptions
     bool noexcept_threw = false;
     json const j_noexcept = parse_without_exceptions([&] { return json::parse(data, data + size, nullptr, false); }, noexcept_threw);

tests/src/fuzzer-parse_msgpack.cpp

@@ -17,6 +17,10 @@ array data, it performs the following steps:
 - j2 = from_msgpack(vec)
 - assert(to_msgpack(j2) == vec)
 
+Furthermore, it reads data with a SAX parser that recovers from every error
+and checks that the events are balanced, that reading ends, and that it
+reports an error exactly when from_msgpack() fails (see #3989).
+
 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
@@ -27,6 +31,7 @@ drivers.
 */
 
 #include "fuzzer_common.hpp"
+#include "fuzzer-recovering_checker.hpp"
 
 namespace
 {
@@ -44,6 +49,9 @@ struct read_msgpack
 // see http://llvm.org/docs/LibFuzzer.html
 extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size)
 {
+    // recover from all errors, reading from memory and from a stream
+    const bool recovered_without_errors = check_recovering_parse(data, size, json::input_format_t::msgpack).errors == 0;
+
     std::vector<uint8_t> const vec1(data, data + size);
 
     // step 0: parse input without exceptions
@@ -62,6 +70,9 @@ extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size)
         // without exceptions, the same input must give the same value
         assert(!noexcept_threw && !j_noexcept.is_discarded() && same_value(j_noexcept, j1));
 
+        // the recovering parser must not have reported an error either
+        assert(recovered_without_errors);
+
         try
         {
             // step 2: round trip
@@ -83,6 +94,7 @@ extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size)
     {
         // parse errors are ok, because input may be random bytes
         assert(parsed || noexcept_threw || j_noexcept.is_discarded());
+        assert(parsed || !recovered_without_errors);
     }
     catch (const json::type_error&)
     {
@@ -93,6 +105,7 @@ extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size)
     {
         // out of range errors may happen if provided sizes are excessive
         assert(parsed || noexcept_threw || j_noexcept.is_discarded());
+        assert(parsed || !recovered_without_errors);
     }
 
     // step 3: the other input adapters must read the data alike

tests/src/fuzzer-parse_ubjson.cpp

@@ -26,6 +26,10 @@ 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.
 
+Furthermore, it reads data with a SAX parser that recovers from every error
+and checks that the events are balanced, that reading ends, and that it
+reports an error exactly when from_ubjson() fails (see #3989).
+
 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
@@ -36,6 +40,7 @@ drivers.
 */
 
 #include "fuzzer_common.hpp"
+#include "fuzzer-recovering_checker.hpp"
 
 namespace
 {
@@ -53,6 +58,9 @@ struct read_ubjson
 // see http://llvm.org/docs/LibFuzzer.html
 extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size)
 {
+    // recover from all errors, reading from memory and from a stream
+    const bool recovered_without_errors = check_recovering_parse(data, size, json::input_format_t::ubjson).errors == 0;
+
     std::vector<uint8_t> const vec1(data, data + size);
 
     // step 0: parse input without exceptions
@@ -71,6 +79,9 @@ extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size)
         // without exceptions, the same input must give the same value
         assert(!noexcept_threw && !j_noexcept.is_discarded() && same_value(j_noexcept, j1));
 
+        // the recovering parser must not have reported an error either
+        assert(recovered_without_errors);
+
         try
         {
             // step 2.1: round trip without adding size annotations to container types
@@ -102,6 +113,7 @@ extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size)
     {
         // parse errors are ok, because input may be random bytes
         assert(parsed || noexcept_threw || j_noexcept.is_discarded());
+        assert(parsed || !recovered_without_errors);
     }
     catch (const json::type_error&)
     {
@@ -112,6 +124,7 @@ extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size)
     {
         // out of range errors may happen if provided sizes are excessive
         assert(parsed || noexcept_threw || j_noexcept.is_discarded());
+        assert(parsed || !recovered_without_errors);
     }
 
     // step 3: the other input adapters must read the data alike

tests/src/fuzzer-recovering_checker.hpp

@@ -0,0 +1,154 @@
+//     __ _____ _____ _____
+//  __|  |   __|     |   | |  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
+
+#pragma once
+
+#include <cassert>
+#include <cstddef>
+#include <cstdint>
+#include <sstream>
+#include <string>
+#include <vector>
+#include <nlohmann/json.hpp>
+
+namespace
+{
+// a SAX parser that recovers from every error and checks that the events are
+// balanced and that every key is followed by exactly one value
+class recovering_checker : public nlohmann::json_sax<nlohmann::json>
+{
+  public:
+    bool null() override
+    {
+        return value();
+    }
+
+    bool boolean(bool /*val*/) override
+    {
+        return value();
+    }
+
+    bool number_integer(number_integer_t /*val*/) override
+    {
+        return value();
+    }
+
+    bool number_unsigned(number_unsigned_t /*val*/) override
+    {
+        return value();
+    }
+
+    bool number_float(number_float_t /*val*/, const string_t& /*s*/) override
+    {
+        return value();
+    }
+
+    bool string(string_t& /*val*/) override
+    {
+        return value();
+    }
+
+    bool binary(binary_t& /*val*/) override
+    {
+        return value();
+    }
+
+    bool start_object(std::size_t /*elements*/) override
+    {
+        value();
+        stack.push_back('o');
+        return true;
+    }
+
+    bool key(string_t& /*val*/) override
+    {
+        ++events;
+        assert(!stack.empty() && stack.back() == 'o');
+        stack.back() = 'v';
+        return true;
+    }
+
+    bool end_object() override
+    {
+        ++events;
+        assert(!stack.empty() && stack.back() == 'o');
+        stack.pop_back();
+        return true;
+    }
+
+    bool start_array(std::size_t /*elements*/) override
+    {
+        value();
+        stack.push_back('a');
+        return true;
+    }
+
+    bool end_array() override
+    {
+        ++events;
+        assert(!stack.empty() && stack.back() == 'a');
+        stack.pop_back();
+        return true;
+    }
+
+    bool parse_error(std::size_t /*position*/, const std::string& /*last_token*/, const nlohmann::detail::exception& /*ex*/) override
+    {
+        ++errors;
+        return true;
+    }
+
+    bool complete() const
+    {
+        return stack.empty();
+    }
+
+    std::size_t events = 0;
+    std::size_t errors = 0;
+
+  private:
+    bool value()
+    {
+        ++events;
+        if (!stack.empty())
+        {
+            // an array element, or the value of a key
+            assert(stack.back() != 'o');
+            if (stack.back() == 'v')
+            {
+                stack.back() = 'o';
+            }
+        }
+        return true;
+    }
+
+    // 'a' for an array, 'o' for an object that expects a key, 'v' for an
+    // object that expects the value of a key
+    std::vector<char> stack {}; // NOLINT(readability-redundant-member-init)
+};
+
+/// parses @a data with a recovering_checker from memory and from a stream,
+/// checks that both see the same, that the events are balanced, and that the
+/// number of errors is bounded, and returns the checker (see #3989)
+inline recovering_checker check_recovering_parse(const std::uint8_t* data, const std::size_t size, const nlohmann::json::input_format_t format)
+{
+    recovering_checker checker;
+    const bool ok = nlohmann::json::sax_parse(data, data + size, &checker, format);
+    assert(checker.complete());
+    assert(checker.errors <= size + 1);
+    assert(ok == (checker.errors == 0));
+
+    std::istringstream stream(std::string(reinterpret_cast<const char*>(data), size));
+    recovering_checker stream_checker;
+    assert(nlohmann::json::sax_parse(stream, &stream_checker, format) == ok);
+    assert(stream_checker.complete());
+    assert(stream_checker.events == checker.events);
+    assert(stream_checker.errors == checker.errors);
+
+    return checker;
+}
+} // namespace

tests/src/sax_event_loggers.hpp

@@ -131,11 +131,13 @@ struct SaxEventLogger : public nlohmann::json_sax<nlohmann::json>
     {
         errored = true;
         events.push_back("parse_error(" + std::to_string(position) + ")");
-        return false;
+        return recover;
     }
 
     std::vector<std::string> events {}; // NOLINT(readability-redundant-member-init)
     bool errored = false;
+    /// whether parse_error() asks the parser to recover from the error (see #3989)
+    bool recover = false;
 };
 
 struct SaxEventLoggerExitAfterStartObject : public SaxEventLogger

tests/src/unit-alt-string.cpp

@@ -415,6 +415,46 @@ TEST_CASE("alternative string type")
         CHECK(j2.dump() == R"({"/foo/0":"bar","/foo/1":"baz"})");
     }
 
+    SECTION("error recovery")
+    {
+        // a SAX parser that recovers from every error (see #3989)
+        struct recovering_parser : nlohmann::detail::json_sax_dom_parser<alt_json>
+        {
+            explicit recovering_parser(alt_json& j)
+                : nlohmann::detail::json_sax_dom_parser<alt_json>(j, false)
+            {}
+
+            // sax_parse() calls the SAX parser's own parse_error(), so hiding
+            // the one of the base class is what recovering takes
+            // NOLINTNEXTLINE(bugprone-derived-method-shadowing-base-method)
+            bool parse_error(std::size_t /*unused*/, const std::string& /*unused*/, const nlohmann::detail::exception& /*unused*/)
+            {
+                ++errors;
+                return true;
+            }
+
+            std::size_t errors = 0;
+        };
+
+        alt_json j;
+        recovering_parser sax(j);
+        // not inside CHECK(): MSVC reads the escape in a stringized raw string
+        const std::string input = R"([1., "a\qb", tru, {"k" 2}])";
+        CHECK(!alt_json::sax_parse(input, &sax));
+        CHECK(sax.errors == 4);
+        CHECK(j.dump() == R"([1,"aqb",null,{"k":2}])");
+
+        // a UBJSON high-precision number, a CBOR key that is not a string
+        alt_json u;
+        recovering_parser ubjson_sax(u);
+        CHECK(!alt_json::sax_parse(std::vector<std::uint8_t> {'[', 'H', 'i', 2, '1', '.', ']'}, &ubjson_sax, alt_json::input_format_t::ubjson));
+        CHECK(u.dump() == "[1]");
+        alt_json c;
+        recovering_parser cbor_sax(c);
+        CHECK(!alt_json::sax_parse(std::vector<std::uint8_t> {0xA2, 0x01, 0x02, 0x61, 'a', 0x03}, &cbor_sax, alt_json::input_format_t::cbor));
+        CHECK(c.dump() == R"({"a":3})");
+    }
+
     SECTION("conversion between basic_json specializations (#2649)")
     {
         // explicit conversions are always possible

tests/src/unit-class_parser.cpp

@@ -2953,6 +2953,651 @@ TEST_CASE("diagnostic positions: value lifetime, input adapters, and SAX")
 }
 #endif
 
+namespace
+{
+/// builds a value like json::parse(), but asks the parser to recover from
+/// errors (see #3989), and checks that the events it receives are balanced
+class RecoveringDomParser
+{
+  public:
+    explicit RecoveringDomParser(json& j, std::size_t max_errors_ = static_cast<std::size_t>(-1))
+        : dom(j, false)
+        , max_errors(max_errors_)
+    {}
+
+    bool null()
+    {
+        value();
+        return dom.null();
+    }
+
+    bool boolean(bool val)
+    {
+        value();
+        return dom.boolean(val);
+    }
+
+    bool number_integer(json::number_integer_t val)
+    {
+        value();
+        return dom.number_integer(val);
+    }
+
+    bool number_unsigned(json::number_unsigned_t val)
+    {
+        value();
+        return dom.number_unsigned(val);
+    }
+
+    bool number_float(json::number_float_t val, const std::string& s)
+    {
+        value();
+        return dom.number_float(val, s);
+    }
+
+    bool string(std::string& val)
+    {
+        value();
+        return dom.string(val);
+    }
+
+    bool binary(json::binary_t& val)
+    {
+        value();
+        return dom.binary(val);
+    }
+
+    bool start_object(std::size_t elements)
+    {
+        value();
+        stack.push_back('o');
+        return dom.start_object(elements);
+    }
+
+    bool key(std::string& val)
+    {
+        ++events;
+        if (stack.empty() || stack.back() != 'o')
+        {
+            well_formed = false;
+            return false;
+        }
+        stack.back() = 'v';
+        return dom.key(val);
+    }
+
+    bool end_object()
+    {
+        ++events;
+        if (stack.empty() || stack.back() != 'o')
+        {
+            well_formed = false;
+            return false;
+        }
+        stack.pop_back();
+        return dom.end_object();
+    }
+
+    bool start_array(std::size_t elements)
+    {
+        value();
+        stack.push_back('a');
+        return dom.start_array(elements);
+    }
+
+    bool end_array()
+    {
+        ++events;
+        if (stack.empty() || stack.back() != 'a')
+        {
+            well_formed = false;
+            return false;
+        }
+        stack.pop_back();
+        return dom.end_array();
+    }
+
+    bool parse_error(std::size_t /*unused*/, const std::string& /*unused*/, const json::exception& ex)
+    {
+        errors.emplace_back(ex.what());
+        return errors.size() < max_errors;
+    }
+
+    /// whether the events were balanced and every key was followed by a value
+    bool balanced() const
+    {
+        return well_formed && stack.empty();
+    }
+
+    /// builds the value
+    nlohmann::detail::json_sax_dom_parser<json> dom;
+    std::vector<std::string> errors {}; // NOLINT(readability-redundant-member-init)
+    std::size_t events = 0;
+    /// the open containers: 'a' for an array, 'o' for an object that expects
+    /// a key, 'v' for an object that expects the value of a key
+    std::vector<char> stack {}; // NOLINT(readability-redundant-member-init)
+    bool well_formed = true;
+    std::size_t max_errors;
+
+  private:
+    /// a value is passed: it is an array element, or the value of a key
+    void value()
+    {
+        ++events;
+        if (!stack.empty())
+        {
+            if (stack.back() == 'v')
+            {
+                stack.back() = 'o';
+            }
+            else if (stack.back() == 'o')
+            {
+                // a value without a key
+                well_formed = false;
+            }
+        }
+    }
+
+};
+
+struct RecoveryResult
+{
+    json value;
+    std::vector<std::string> errors;
+    std::size_t events;
+    bool ok;
+    bool balanced;
+};
+
+template<typename InputType>
+RecoveryResult parse_recovering(InputType&& input, const bool strict = true,
+                                const bool ignore_comments = false, const bool ignore_trailing_commas = false)
+{
+    json j;
+    RecoveringDomParser sax(j);
+    const bool ok = json::sax_parse(std::forward<InputType>(input), &sax, json::input_format_t::json,
+                                    strict, ignore_comments, ignore_trailing_commas);
+    return {j, sax.errors, sax.events, ok, sax.balanced()};
+}
+
+/// stops after a number of events, but recovers from errors
+class RecoveringCountdown
+{
+  public:
+    explicit RecoveringCountdown(const int count)
+        : countdown(count)
+    {}
+
+    bool null()
+    {
+        return countdown.null();
+    }
+
+    bool boolean(bool val)
+    {
+        return countdown.boolean(val);
+    }
+
+    bool number_integer(json::number_integer_t val)
+    {
+        return countdown.number_integer(val);
+    }
+
+    bool number_unsigned(json::number_unsigned_t val)
+    {
+        return countdown.number_unsigned(val);
+    }
+
+    bool number_float(json::number_float_t val, const std::string& s)
+    {
+        return countdown.number_float(val, s);
+    }
+
+    bool string(std::string& val)
+    {
+        return countdown.string(val);
+    }
+
+    bool binary(json::binary_t& val)
+    {
+        return countdown.binary(val);
+    }
+
+    bool start_object(std::size_t elements)
+    {
+        return countdown.start_object(elements);
+    }
+
+    bool key(std::string& val)
+    {
+        return countdown.key(val);
+    }
+
+    bool end_object()
+    {
+        return countdown.end_object();
+    }
+
+    bool start_array(std::size_t elements)
+    {
+        return countdown.start_array(elements);
+    }
+
+    bool end_array()
+    {
+        return countdown.end_array();
+    }
+
+    bool parse_error(std::size_t /*position*/, const std::string& /*last_token*/, const json::exception& /*ex*/) // NOLINT(readability-convert-member-functions-to-static)
+    {
+        return true;
+    }
+
+  private:
+    SaxCountdown countdown;
+};
+
+/// a repaired input: the value it is repaired to, and the number of errors
+struct Repair
+{
+    const char* input;
+    const char* expected;
+    std::size_t errors;
+};
+} // namespace
+
+TEST_CASE("parser error recovery (#3989)")
+{
+    SECTION("repairs")
+    {
+        const std::vector<Repair> repairs =
+        {
+            // a missing separator is inserted
+            {"[1 2]", "[1,2]", 1},
+            {R"({"a":1 "b":2})", R"({"a":1,"b":2})", 1},
+            {R"({"a" 1})", R"({"a":1})", 1},
+            {"[1 tru 2]", "[1,null,2]", 2},
+            {R"({"a" "b": 1})", R"({"a":"b"})", 2},
+
+            // a missing value is null in an object; in an array, a ',' stands
+            // for null, while an array that ends there just ends
+            {R"({"a":})", R"({"a":null})", 1},
+            {R"({"a"})", R"({"a":null})", 1},
+            {R"({"a","b":1})", R"({"a":null,"b":1})", 1},
+            {"[1,,2]", "[1,null,2]", 1},
+            {"[,1]", "[null,1]", 1},
+            {"[1,]", "[1]", 1},
+            {"[1,2,3,]", "[1,2,3]", 1},
+            {R"({"a":1,})", R"({"a":1})", 1},
+
+            // a broken string keeps what can be read
+            {R"(["a\qb"])", R"(["aqb"])", 1},
+            {R"({"na\me":1})", R"({"name":1})", 1},
+            {"[\"\xFF\"]", R"(["\uFFFD"])", 1},
+            {"[\"a\xC3(\"]", R"(["a\uFFFD("])", 1},
+            {"[\"\xE2\x82\"]", R"(["\uFFFD"])", 1},
+            {"[\"\xC3\\\\\", 1]", R"(["\uFFFD\\",1])", 1},
+            {R"(["\u12"])", R"(["\uFFFD"])", 1},
+            {R"(["\u12G4"])", R"(["\uFFFDG4"])", 1},
+            {R"(["\uDC00x"])", R"(["\uFFFDx"])", 1},
+            {R"(["\uD800x"])", R"(["\uFFFDx"])", 1},
+            {R"(["\uD800\u0041"])", R"(["\uFFFDA"])", 1},
+            {R"(["\uD800\uD800\uDC00"])", R"(["\uFFFD\uD800\uDC00"])", 1},
+            {R"(["\uD800\uD800\uD800x"])", R"(["\uFFFD\uFFFD\uFFFDx"])", 1},
+            {
+                R"(["\uD800\"x", 1])", R"(["\uFFFD\"x",1])", 1
+            },
+            {R"(["\uD800\q"])", R"(["\uFFFDq"])", 1},
+            {"[\"a\tb\"]", R"(["a\tb"])", 1},
+            {R"(["a\qb\u0041\x"])", R"(["aqbAx"])", 1},
+
+            // a broken number keeps its longest valid prefix
+            {"[1.]", "[1]", 1},
+            {"[-2.]", "[-2]", 1},
+            {"[1.5e]", "[1.5]", 1},
+            {"[1e+]", "[1]", 1},
+            {"[1.x2, 3]", "[1,3]", 1},
+
+            // what cannot be read at all is null
+            {"[1,NaN,3]", "[1,null,3]", 1},
+            {"[tru]", "[null]", 1},
+            {"[-]", "[null]", 1},
+            {R"({"a":Infinity})", R"({"a":null})", 1},
+
+            // a stray token is dropped
+            {"[:1]", "[1]", 1},
+            {R"(["a":1])", R"(["a",1])", 1},
+            {R"({"a"::1})", R"({"a":1})", 1},
+
+            // a member that cannot be read is skipped
+            {R"({1:2,"b":3})", R"({"b":3})", 1},
+            {R"({"a":1 2})", R"({"a":1})", 1},
+            {R"({,"a":1})", R"({"a":1})", 1},
+            {R"({"a":1,,"b":2})", R"({"a":1,"b":2})", 1},
+            {"{a:1}", "{}", 1},
+            {R"({"a":1 [1,{"b":2}], "c":3})", R"({"a":1,"c":3})", 1},
+            {R"([{1}, "a"])", R"([{},"a"])", 1},
+
+            // a wrong closing bracket closes the innermost container
+            {R"({"a":[1,2}, "b":3})", R"({"a":[1,2],"b":3})", 1},
+            {R"([{"a":1], 2])", R"([{"a":1},2])", 1},
+            {"{]", "{}", 1},
+            {"[}", "[]", 1},
+
+            // the end of the input closes all containers
+            {R"({"a":[1,2)", R"({"a":[1,2]})", 1},
+            {"[", "[]", 1},
+            {"{", "{}", 1},
+            {R"({"a")", R"({"a":null})", 1},
+            {R"({"a":)", R"({"a":null})", 1},
+            {"[1,", "[1]", 1},
+            {"[[[1", "[[[1]]]", 1},
+            {
+                R"(["abc)", R"(["abc"])", 2
+            },
+            {"[1,tr", "[1,null]", 2},
+            {"\"abc", "\"abc\"", 1},
+            {"[\"ab\ncd\"]", R"(["ab",null,"]"])", 4},
+
+            // what comes before the top-level value is skipped
+            {")]}'\n{\"a\":1}", R"({"a":1})", 1},
+            {R"(data: {"a":1})", R"({"a":1})", 1},
+            {"\xEF\xBB[1]", "[1]", 1},
+
+            // what comes after it is an error that ends parsing
+            {R"({"a":1}})", R"({"a":1})", 1},
+            {"[1}]", "[1]", 2},
+            {"[1] [2]", "[1]", 1},
+        };
+
+        for (const auto& repair : repairs)
+        {
+            CAPTURE(repair.input)
+            const auto result = parse_recovering(std::string(repair.input));
+            CHECK(!result.ok);
+            CHECK(result.balanced);
+            CHECK(result.value == json::parse(repair.expected));
+            CHECK(result.errors.size() == repair.errors);
+        }
+    }
+
+    SECTION("number overflow")
+    {
+        const auto result = parse_recovering(std::string("[1e999,-1e999]"));
+        CHECK(!result.ok);
+        CHECK(result.balanced);
+        CHECK(result.errors.size() == 2);
+        CHECK(result.errors[0] == "[json.exception.out_of_range.406] number overflow parsing '1e999'");
+        REQUIRE(result.value.size() == 2);
+        CHECK(result.value[0].is_number_float());
+        CHECK(result.value[0].get<double>() == std::numeric_limits<double>::infinity());
+        CHECK(result.value[1].get<double>() == -std::numeric_limits<double>::infinity());
+
+        // the SAX parser gets the number's text
+        SaxEventLogger logger;
+        logger.recover = true;
+        CHECK(!json::sax_parse("1e999", &logger));
+        CHECK(logger.events == std::vector<std::string>({"parse_error(5)", "number_float(1e999)"}));
+    }
+
+    SECTION("nothing to recover")
+    {
+        for (const std::string s :
+                {
+                    "", "   ", "]", "tru", "NaN", ",:", "/* comment"
+                })
+        {
+            CAPTURE(s)
+            const auto result = parse_recovering(s, true, true);
+            CHECK(!result.ok);
+            CHECK(result.balanced);
+            CHECK(result.events == 0);
+            CHECK(result.value == nullptr);
+            CHECK(result.errors.size() == 1);
+        }
+    }
+
+    SECTION("error messages")
+    {
+        // the first error is reported as without recovery
+        for (const std::string s :
+                {
+                    "[1 2]", R"({"a":1 "b":2})", R"({"a" 1})", R"({"a":})", "[1,]", "[1.]",
+                    R"(["a\qb"])", "[1e999]", "{1:2}", R"({"a":[1,2}})", "[1,", "[1] [2]", "{a:1}"
+                })
+        {
+            CAPTURE(s)
+            const auto result = parse_recovering(s);
+            REQUIRE(!result.errors.empty());
+            json _;
+            CHECK_THROWS_WITH_STD_STR(_ = json::parse(s), result.errors.front());
+        }
+
+        // the token of an error begins where the previous error was
+        const auto result = parse_recovering(std::string("[tru, fals, nul]"));
+        CHECK(result.errors == std::vector<std::string>(
+        {
+            "[json.exception.parse_error.101] parse error at line 1, column 5: syntax error while parsing value - invalid literal; last read: '[tru,'",
+            "[json.exception.parse_error.101] parse error at line 1, column 11: syntax error while parsing value - invalid literal; last read: ', fals,'",
+            "[json.exception.parse_error.101] parse error at line 1, column 16: syntax error while parsing value - invalid literal; last read: ', nul]'"
+        }));
+        CHECK(result.value == json::parse("[null,null,null]"));
+    }
+
+    SECTION("events")
+    {
+        // see #4522
+        SaxEventLogger logger;
+        logger.recover = true;
+        CHECK(!json::sax_parse(R"([{1}, "a"])", &logger));
+        CHECK(logger.events == std::vector<std::string>(
+        {
+            "start_array()", "start_object()", "parse_error(3)", "end_object()", "string(a)", "end_array()"
+        }));
+    }
+
+    SECTION("options")
+    {
+        SECTION("strict")
+        {
+            const auto result = parse_recovering(std::string("[1 2] [3]"), false);
+            CHECK(!result.ok);
+            CHECK(result.value == json::parse("[1,2]"));
+            CHECK(result.errors.size() == 1);
+        }
+
+        SECTION("ignore_trailing_commas")
+        {
+            for (const std::string s :
+                    {
+                        "[1,]", R"({"a":1,})", "[[1,],]"
+                    })
+            {
+                CAPTURE(s)
+                const auto result = parse_recovering(s, true, false, true);
+                CHECK(result.ok);
+                CHECK(result.errors.empty());
+            }
+
+            auto result = parse_recovering(std::string("[1,,]"), true, false, true);
+            CHECK(result.value == json::parse("[1,null]"));
+            CHECK(result.errors.size() == 1);
+
+            result = parse_recovering(std::string(R"({"a":1,,})"), true, false, true);
+            CHECK(result.value == json::parse(R"({"a":1})"));
+            CHECK(result.errors.size() == 1);
+        }
+
+        SECTION("ignore_comments")
+        {
+            auto result = parse_recovering(std::string("[1 /* one */ 2]"), true, true);
+            CHECK(result.value == json::parse("[1,2]"));
+            CHECK(result.errors.size() == 1);
+
+            // a comment that is not closed runs to the end of the input, which
+            // is not reported again
+            result = parse_recovering(std::string("[1, 2 /* unterminated"), true, true);
+            CHECK(result.balanced);
+            CHECK(result.value == json::parse("[1,2]"));
+            CHECK(result.errors.size() == 1);
+
+            // a '/' that does not begin a comment is garbage
+            result = parse_recovering(std::string("[1, /x, 2]"), true, true);
+            CHECK(result.balanced);
+            CHECK(result.value == json::parse("[1,null,2]"));
+            CHECK(result.errors.size() == 1);
+        }
+    }
+
+    SECTION("null bytes")
+    {
+        // a null byte ends the input, unless JSON_STRICT_NUL_HANDLING is set
+        const auto result = parse_recovering(std::string("[1,\0x", 5));
+        CHECK(result.balanced);
+        CHECK(!result.ok);
+#ifdef JSON_TEST_STRICT_NUL_HANDLING_ENABLED
+        CHECK(result.value == json::parse("[1,null]"));
+#else
+        CHECK(result.value == json::parse("[1]"));
+        CHECK(result.errors.size() == 1);
+#endif
+
+        const auto in_string = parse_recovering(std::string("[\"a\0b\"]", 7));
+        CHECK(in_string.balanced);
+#ifdef JSON_TEST_STRICT_NUL_HANDLING_ENABLED
+        CHECK(in_string.value == json::array({std::string("a\0b", 3)}));
+#else
+        CHECK(in_string.value == json::parse(R"(["a"])"));
+#endif
+    }
+
+    SECTION("the SAX parser stops recovering")
+    {
+        json j;
+        RecoveringDomParser sax(j, 2);
+        CHECK(!json::sax_parse("[1 2 3 4 5]", &sax));
+        CHECK(sax.errors.size() == 2);
+
+        // an error at a delimiter that an invalid token consumed is reported
+        // to the SAX parser, too
+        json j2;
+        RecoveringDomParser sax2(j2, 2);
+        CHECK(!json::sax_parse("[tru}, 1]", &sax2));
+        CHECK(sax2.errors.size() == 2);
+    }
+
+    SECTION("an event stops parsing during a repair")
+    {
+        // start_object() and key() are passed, then null() for the missing
+        // value returns false
+        RecoveringCountdown countdown(2);
+        CHECK(!json::sax_parse(R"({"a":})", &countdown));
+
+        // the end of the input: end_array() for the second array returns false
+        RecoveringCountdown countdown2(4);
+        CHECK(!json::sax_parse("[[1", &countdown2));
+    }
+
+    SECTION("input adapters")
+    {
+        // the lexer reads contiguous and streaming input differently, and it
+        // puts back a character that ended an invalid token
+        for (const std::string s :
+                {
+                    "[1 2]", "[tru}, 1]", R"({"a" "b\q", "c":[1.x, 2}})", "[\"\xFF\xC3(\", -, 1e+]", "{a:1,\"b\":2", ")]}' [1]"
+                })
+        {
+            CAPTURE(s)
+            const auto reference = parse_recovering(s);
+            CHECK(reference.balanced);
+
+            const auto from_c_string = parse_recovering(s.c_str());
+            CHECK(from_c_string.value == reference.value);
+            CHECK(from_c_string.errors == reference.errors);
+
+            const std::list<char> l(s.begin(), s.end());
+            json j;
+            RecoveringDomParser sax(j);
+            CHECK(!json::sax_parse(l.begin(), l.end(), &sax));
+            CHECK(j == reference.value);
+            CHECK(sax.errors == reference.errors);
+
+            std::istringstream ss(s);
+            const auto from_stream = parse_recovering(ss);
+            CHECK(from_stream.value == reference.value);
+            CHECK(from_stream.errors == reference.errors);
+        }
+    }
+
+    SECTION("long runs of errors")
+    {
+        // no error may copy all the input read before it
+        const auto closing = parse_recovering("[" + std::string(100000, '}'));
+        CHECK(closing.balanced);
+        CHECK(closing.value == json::array());
+
+        const auto garbage = parse_recovering("[" + std::string(100000, 'x') + "]");
+        CHECK(garbage.balanced);
+        CHECK(garbage.errors.size() == 1);
+
+        const auto commas = parse_recovering("{" + std::string(100000, ',') + "}");
+        CHECK(commas.balanced);
+        CHECK(commas.value == json::object());
+    }
+
+    SECTION("mutations of valid input")
+    {
+        // whatever the input, the events are balanced, every error is reported
+        // at most once, and valid input is parsed as usual
+        const std::vector<std::string> documents =
+        {
+            R"({"name": "value", "list": [1, -2.5, true, null, {"x": [[]]}], "e": "\u00e9"})",
+            R"([{"a": [1, 2, {"b": "c"}]}, [], {}, "\ud83d\ude00", 1e10])",
+            "{\"\xC3\xA9\": \"\xF0\x9F\x98\x80\"}",
+            R"(  {"k" : [ "v" , 0 ] }  )",
+        };
+        // each character that can be inserted, including a null byte
+        const std::string insertions("[]{},:\"x\\\0\xFF", 11);
+
+        std::vector<std::string> inputs;
+        for (const auto& doc : documents)
+        {
+            for (std::size_t i = 0; i <= doc.size(); ++i)
+            {
+                inputs.push_back(doc.substr(0, i));
+                if (i < doc.size())
+                {
+                    inputs.push_back(doc.substr(0, i) + doc.substr(i + 1));
+                }
+                for (const char c : insertions)
+                {
+                    inputs.push_back(doc.substr(0, i) + c + doc.substr(i));
+                }
+            }
+        }
+
+        for (const auto& s : inputs)
+        {
+            CAPTURE(s)
+            const auto result = parse_recovering(s);
+            CHECK(result.balanced);
+            CHECK(result.errors.size() <= s.size() + 1);
+            CHECK(result.events <= (4 * s.size()) + 4);
+            if (json::accept(s))
+            {
+                CHECK(result.ok);
+                CHECK(result.errors.empty());
+                CHECK(result.value == json::parse(s));
+            }
+            else
+            {
+                CHECK(!result.ok);
+                CHECK(!result.errors.empty());
+            }
+        }
+    }
+}
+
 TEST_CASE("accept() agrees with parse() on floats around the overflow limit")
 {
     // accept() does not convert a float that cannot overflow; the tokens

tests/src/unit-json_view_dump.cpp

@@ -150,71 +150,6 @@ TEST_CASE("json_view dump")
             CHECK(many_doc.root().dump() == json::parse(many_tokens).dump());
         }
 
-        // tokens of 16 and 17 digits without an exponent (like GeoJSON
-        // coordinates), which dump() converts in batches: doubles exactly
-        // halfway between two candidates for the shortest digits, random
-        // digits, and digits near a shorter decimal (...99999997,
-        // ...00000003), whose rounding carries; in pairs and among other
-        // values, so that the batches are interrupted; the last token near the
-        // end of the text
-        {
-            std::mt19937_64 tokens(5627); // NOLINT(cert-msc32-c,cert-msc51-cpp,bugprone-random-generator-seed)
-            const auto draw = [&tokens](std::size_t n)
-            {
-                const std::uint64_t r = tokens() % n;
-                return static_cast<std::size_t>(r);
-            };
-            std::string coordinates = "[754570942.33984373,257769445.56835938,858106901.94921875,97944964.876953125,"
-                                      "[0.5,1e-7],[1.5,-0.0,2.5],[1E2,12.5,0.1,3.25,4.5,5.75,6.125,7.0,8.5],[],[1.5],[1,2.5],[2.5,\"x\"],[[1.5],2.5],";
-            for (int i = 0; i < 20000; ++i)
-            {
-                const std::size_t length = 16 + draw(2);
-                std::string digits(1, static_cast<char>('1' + draw(9)));
-                const std::size_t kept = draw(3) == 0 ? length : 2 + draw(length - 3);
-                for (std::size_t k = 1; k < kept; ++k)
-                {
-                    digits += static_cast<char>('0' + draw(10));
-                }
-                digits += std::string(length - kept, draw(2) == 0 ? '9' : '0');
-                digits.back() = static_cast<char>('0' + draw(10));
-                const std::size_t point = 1 + draw(9);
-                const std::string token = (draw(2) == 0 ? "-" : "") + digits.substr(0, point) + "." + digits.substr(point);
-                coordinates += (i != 0 ? "," : "") + (draw(8) == 0 ? "{\"x\":" + token + ",\"s\":\"a\"}" : "[" + token + "," + token + "]");
-            }
-            coordinates += ",-65.613616999999977]";
-            const json_document parsed = json_document::parse(coordinates);
-            const ordered_json j = ordered_json::parse(coordinates);
-            CHECK(parsed.root().dump() == j.dump());
-            CHECK(parsed.root().dump(1) == j.dump(1));
-            CHECK(parsed.root().dump(-1, ' ', true) == j.dump(-1, ' ', true));
-            CHECK(parsed.root().back().get<double>() == j.back().get<double>());
-            CHECK(parsed.root().back().dump() == "-65.61361699999998");
-            std::size_t i = 0;
-            for (const json_view v : parsed.root())
-            {
-                const ordered_json& e = j[i++];
-                CHECK(v.dump() == e.dump());
-                if (v.is_array() && v.size() != 0 && v[0].is_number())
-                {
-                    CHECK(v[0].get<double>() == e[0].get<double>());
-                }
-            }
-
-            // an editable document: edits move some arrays, so that the walk
-            // leaves the order of the parsed nodes
-            ordered_json_editable_document editable = ordered_json_editable_document::parse(coordinates);
-            ordered_json expected = j;
-            for (std::size_t k = 0; k < 20000; k += 97)
-            {
-                if (expected[k].is_array())
-                {
-                    editable.push_back(editable.root()[k], ordered_json(1.5));
-                    expected[k].push_back(1.5);
-                }
-            }
-            CHECK(editable.root().dump() == expected.dump());
-        }
-
         // random doubles, written as parse() and dump() would
         std::mt19937_64 rng(1170); // NOLINT(cert-msc32-c,cert-msc51-cpp,bugprone-random-generator-seed)
         std::string many = "[";

tests/src/unit-msgpack.cpp

@@ -31,8 +31,6 @@ using nlohmann::json;
 #include "make_test_data_available.hpp"
 #include "round_trip_corpus.hpp"
 #include "test_utils.hpp"
-#include "sax_countdown.hpp"
-using utils::SaxCountdown;
 
 
 TEST_CASE("MessagePack")
@@ -1645,30 +1643,6 @@ TEST_CASE("MessagePack")
             }
         }
     }
-
-    SECTION("SAX aborts")
-    {
-        SECTION("start_array(len)")
-        {
-            std::vector<uint8_t> const v = {0x93, 0x01, 0x02, 0x03};
-            SaxCountdown scp(0);
-            CHECK(!json::sax_parse(v, &scp, json::input_format_t::msgpack));
-        }
-
-        SECTION("start_object(len)")
-        {
-            std::vector<uint8_t> const v = {0x81, 0xa3, 0x66, 0x6F, 0x6F, 0xc2};
-            SaxCountdown scp(0);
-            CHECK(!json::sax_parse(v, &scp, json::input_format_t::msgpack));
-        }
-
-        SECTION("key()")
-        {
-            std::vector<uint8_t> const v = {0x81, 0xa3, 0x66, 0x6F, 0x6F, 0xc2};
-            SaxCountdown scp(1);
-            CHECK(!json::sax_parse(v, &scp, json::input_format_t::msgpack));
-        }
-    }
 }
 
 TEST_CASE("issue #5405 - array reserve for definite-length MessagePack arrays")
@@ -1739,21 +1713,6 @@ TEST_CASE("issue #5405 - array reserve for definite-length MessagePack arrays")
             CHECK(json::from_msgpack(packed) == j);
         }
     }
-
-    SECTION("a user-defined SAX consumer is unaffected by the internal DOM reserve optimization")
-    {
-        // the reserve() call is local to json_sax_dom_parser / json_sax_dom_callback_parser;
-        // a custom SAX consumer that does not touch a DOM array sees identical events
-        json j = json::array();
-        for (int i = 0; i < 100; ++i)
-        {
-            j.push_back(i);
-        }
-        const auto packed = json::to_msgpack(j);
-
-        SaxCountdown scp(1000000); // large enough to never trigger an abort
-        CHECK(json::sax_parse(packed, &scp, json::input_format_t::msgpack));
-    }
 }
 
 TEST_CASE("regression test - MessagePack ext type rejects a subtype that doesn't fit a single byte")
@@ -1792,15 +1751,6 @@ TEST_CASE("MessagePack nesting does not consume the call stack")
         CHECK(json::from_msgpack(input, true, false).is_discarded());
     }
 
-    SECTION("a well-formed deep value is read through the SAX interface")
-    {
-        std::vector<uint8_t> input(300000, 0x91);
-        input.push_back(0x01); // innermost value
-
-        SaxCountdown accept_all(600001);
-        CHECK(json::sax_parse(input, &accept_all, json::input_format_t::msgpack));
-    }
-
     SECTION("a well-formed deep value is read into a value")
     {
         const std::size_t depth = 10000;
@@ -1848,31 +1798,6 @@ TEST_CASE("MessagePack input that cannot be read is discarded by every overload"
 #endif
 }
 
-TEST_CASE("MessagePack SAX parsing stops at every event")
-{
-    // Containers are opened and closed by the loop that reads them; a SAX
-    // handler that rejects any event - including the end of a nested
-    // container - must stop the parse right there.
-    const auto count_events = [](const std::vector<std::uint8_t>& input)
-    {
-        int events = 0;
-        while (true)
-        {
-            SaxCountdown scp(events);
-            if (json::sax_parse(input, &scp, json::input_format_t::msgpack))
-            {
-                return events;
-            }
-            ++events;
-            REQUIRE(events < 1000);
-        }
-    };
-
-    // 20 events: every container kind closes inside another one
-    const json j = json::parse(R"({"a": [1, {"b": []}], "c": {"d": [[2]]}})");
-    CHECK(count_events(json::to_msgpack(j)) == 20);
-}
-
 TEST_CASE("single MessagePack roundtrip")
 {
     SECTION("sample.json")

tests/src/unit-regression3.cpp

@@ -22,14 +22,6 @@
 // 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
-// capture whether JSON_DELETE_DEPRECATED_FUNCTIONS was enabled on the command
-// line *before* including json.hpp, since the library #undefs it once the header
-// has been fully processed (see include/nlohmann/detail/macro_unscope.hpp); the
-// tests of deprecated functions are skipped if these functions are deleted
-#if defined(JSON_DELETE_DEPRECATED_FUNCTIONS) && (JSON_DELETE_DEPRECATED_FUNCTIONS == 1)
-    #define JSON_TEST_DEPRECATED_FUNCTIONS_DELETED
-#endif
-
 #if defined(JSON_DISABLE_ENUM_SERIALIZATION) && (JSON_DISABLE_ENUM_SERIALIZATION == 1)
     #define SKIP_TESTS_FOR_ENUM_SERIALIZATION
 #endif
@@ -810,50 +802,6 @@ TEST_CASE("issue #5338 - truncated CBOR tagged binary subtype is rejected")
     }
 }
 
-TEST_CASE("issue #5676 - SAX parsing of CBOR tags")
-{
-    const json expected = json::binary({1, 2, 3}, 42);
-    const auto cbor = json::to_cbor(expected);
-
-    nlohmann::detail::json_sax_acceptor<json> acceptor;
-    CHECK_FALSE(json::sax_parse(cbor, &acceptor, json::input_format_t::cbor));
-    CHECK_FALSE(json::sax_parse(cbor, &acceptor, json::input_format_t::cbor,
-                                true, false, false, json::cbor_tag_handler_t::error));
-
-    CHECK(json::sax_parse(cbor, &acceptor, json::input_format_t::cbor,
-                          true, false, false, json::cbor_tag_handler_t::ignore));
-
-    json parsed;
-    nlohmann::detail::json_sax_dom_parser<json, nlohmann::detail::string_input_adapter_type> sax(parsed);
-    CHECK(json::sax_parse(cbor, &sax, json::input_format_t::cbor,
-                          true, false, false, json::cbor_tag_handler_t::store));
-    CHECK(parsed == expected);
-
-    json iterator_parsed;
-    nlohmann::detail::json_sax_dom_parser<json, nlohmann::detail::string_input_adapter_type> iterator_sax(iterator_parsed);
-    CHECK(json::sax_parse(cbor.begin(), cbor.end(), &iterator_sax, json::input_format_t::cbor,
-                          true, false, false, json::cbor_tag_handler_t::store));
-    CHECK(iterator_parsed == expected);
-
-#ifndef JSON_TEST_DEPRECATED_FUNCTIONS_DELETED
-    json span_parsed;
-    nlohmann::detail::json_sax_dom_parser<json, nlohmann::detail::string_input_adapter_type> span_sax(span_parsed);
-    CHECK(json::sax_parse(nlohmann::detail::span_input_adapter(cbor.data(), cbor.size()), &span_sax,
-                          json::input_format_t::cbor, true, false, false, json::cbor_tag_handler_t::store));
-    CHECK(span_parsed == expected);
-#endif
-
-    const std::string text = "null";
-    CHECK(json::sax_parse(text, &acceptor, json::input_format_t::json,
-                          true, false, false, json::cbor_tag_handler_t::store));
-    CHECK(json::sax_parse(text.begin(), text.end(), &acceptor, json::input_format_t::json,
-                          true, false, false, json::cbor_tag_handler_t::store));
-#ifndef JSON_TEST_DEPRECATED_FUNCTIONS_DELETED
-    CHECK(json::sax_parse(nlohmann::detail::span_input_adapter(text.data(), text.size()), &acceptor,
-                          json::input_format_t::json, true, false, false, json::cbor_tag_handler_t::store));
-#endif
-}
-
 TEST_CASE("issue #5402 - update(merge_objects=true) overwrites a primitive with an object")
 {
     json t = {{"k", 1}};

tests/src/unit-sax_parse.cpp

@@ -0,0 +1,768 @@
+//     __ _____ _____ _____
+//  __|  |   __|     |   | |  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
+
+/////////////////////////////////////////////////////////////////////
+// Tests that call basic_json::sax_parse have a file of their own: every
+// sax_parse call instantiates the parser and binary reader that recover from
+// errors (see #3989), and in unit-regression2.cpp, unit-regression3.cpp, and
+// unit-msgpack.cpp this made the objects too large for the MinGW linker to
+// relocate (see #5511).
+/////////////////////////////////////////////////////////////////////
+
+#include "doctest_compatibility.h"
+
+// capture whether JSON_DELETE_DEPRECATED_FUNCTIONS was enabled on the command
+// line *before* including json.hpp, since the library #undefs it once the header
+// has been fully processed (see include/nlohmann/detail/macro_unscope.hpp); the
+// tests of deprecated functions are skipped if these functions are deleted
+#if defined(JSON_DELETE_DEPRECATED_FUNCTIONS) && (JSON_DELETE_DEPRECATED_FUNCTIONS == 1)
+    #define JSON_TEST_DEPRECATED_FUNCTIONS_DELETED
+#endif
+
+#include <nlohmann/json.hpp>
+using json = nlohmann::json;
+
+#include <cmath>
+#include <cstddef>
+#include <cstdint>
+#include <initializer_list>
+#include <map>
+#include <string>
+#include <utility>
+#include <vector>
+
+#include "sax_countdown.hpp"
+using utils::SaxCountdown;
+
+// a narrow number_float_t, so that a double read from binary input can
+// overflow it
+using float_json = nlohmann::basic_json<std::map, std::vector, std::string, bool, std::int64_t, std::uint64_t, float>;
+
+DOCTEST_CLANG_SUPPRESS_WARNING_PUSH
+DOCTEST_CLANG_SUPPRESS_WARNING("-Wexit-time-destructors")
+
+namespace
+{
+/// builds a value from SAX events, asks the parser to recover from its first
+/// 100 errors, and checks that the events are balanced (see #3989)
+template<typename BasicJsonType>
+class BasicRecoveringParser
+{
+  public:
+    explicit BasicRecoveringParser(BasicJsonType& j)
+        : dom(j, false)
+    {}
+
+    bool null()
+    {
+        value();
+        return dom.null();
+    }
+
+    bool boolean(bool val)
+    {
+        value();
+        return dom.boolean(val);
+    }
+
+    bool number_integer(typename BasicJsonType::number_integer_t val)
+    {
+        value();
+        return dom.number_integer(val);
+    }
+
+    bool number_unsigned(typename BasicJsonType::number_unsigned_t val)
+    {
+        value();
+        return dom.number_unsigned(val);
+    }
+
+    bool number_float(typename BasicJsonType::number_float_t val, const std::string& s)
+    {
+        value();
+        return dom.number_float(val, s);
+    }
+
+    bool string(std::string& val)
+    {
+        value();
+        return dom.string(val);
+    }
+
+    bool binary(typename BasicJsonType::binary_t& val)
+    {
+        value();
+        return dom.binary(val);
+    }
+
+    bool start_object(std::size_t elements)
+    {
+        value();
+        stack.push_back('o');
+        return dom.start_object(elements);
+    }
+
+    bool key(std::string& val)
+    {
+        if (stack.empty() || stack.back() != 'o')
+        {
+            well_formed = false;
+            return false;
+        }
+        stack.back() = 'v';
+        return dom.key(val);
+    }
+
+    bool end_object()
+    {
+        if (stack.empty() || stack.back() != 'o')
+        {
+            well_formed = false;
+            return false;
+        }
+        stack.pop_back();
+        return dom.end_object();
+    }
+
+    bool start_array(std::size_t elements)
+    {
+        value();
+        stack.push_back('a');
+        return dom.start_array(elements);
+    }
+
+    bool end_array()
+    {
+        if (stack.empty() || stack.back() != 'a')
+        {
+            well_formed = false;
+            return false;
+        }
+        stack.pop_back();
+        return dom.end_array();
+    }
+
+    bool parse_error(std::size_t /*unused*/, const std::string& /*unused*/, const json::exception& ex)
+    {
+        messages.emplace_back(ex.what());
+        // a limit, so that a reader that does not stop fails the test
+        // instead of making it hang
+        return ++errors < 100;
+    }
+
+    /// whether the events were balanced and every key was followed by a value
+    bool balanced() const
+    {
+        return well_formed && stack.empty();
+    }
+
+    /// builds the value
+    nlohmann::detail::json_sax_dom_parser<BasicJsonType> dom;
+    std::size_t errors = 0;
+    std::vector<std::string> messages {}; // NOLINT(readability-redundant-member-init)
+    std::vector<char> stack {}; // NOLINT(readability-redundant-member-init)
+    bool well_formed = true;
+
+  private:
+    void value()
+    {
+        if (!stack.empty())
+        {
+            if (stack.back() == 'v')
+            {
+                stack.back() = 'o';
+            }
+            else if (stack.back() == 'o')
+            {
+                well_formed = false;
+            }
+        }
+    }
+
+};
+
+using RecoveringParser = BasicRecoveringParser<json>;
+
+struct BinaryParseResult
+{
+    json value;
+    std::size_t errors;
+    std::vector<std::string> messages;
+    bool ok;
+    bool balanced;
+};
+
+BinaryParseResult parse_binary_recovering(const std::vector<std::uint8_t>& input, const json::input_format_t format)
+{
+    json j;
+    RecoveringParser sax(j);
+    const bool ok = json::sax_parse(input, &sax, format);
+    return {j, sax.errors, sax.messages, ok, sax.balanced()};
+}
+
+#if !defined(JSON_NOEXCEPTION)
+/// the message of the exception that reading @a input into a JSON value
+/// throws, or an empty string if reading succeeds
+std::string binary_error_message(const std::vector<std::uint8_t>& input, const json::input_format_t format)
+{
+    try
+    {
+        json _;
+        switch (format)
+        {
+            case json::input_format_t::cbor:
+                _ = json::from_cbor(input);
+                break;
+            case json::input_format_t::msgpack:
+                _ = json::from_msgpack(input);
+                break;
+            case json::input_format_t::ubjson:
+                _ = json::from_ubjson(input);
+                break;
+            case json::input_format_t::bjdata:
+                _ = json::from_bjdata(input);
+                break;
+            case json::input_format_t::bson:
+                _ = json::from_bson(input);
+                break;
+            case json::input_format_t::bon8:
+                _ = json::from_bon8(input);
+                break;
+            case json::input_format_t::json:
+            default:
+                break;
+        }
+    }
+    catch (const json::exception& e)
+    {
+        return e.what();
+    }
+    return "";
+}
+#endif
+
+/// a BSON element: its type, its name, and its value
+std::vector<std::uint8_t> bson_element(const std::uint8_t type, const std::string& name, const std::vector<std::uint8_t>& value)
+{
+    std::vector<std::uint8_t> result = {type};
+    result.insert(result.end(), name.begin(), name.end());
+    result.push_back(0x00);
+    result.insert(result.end(), value.begin(), value.end());
+    return result;
+}
+
+/// a BSON document of the given elements; @a size_offset is added to the
+/// size it declares
+std::vector<std::uint8_t> bson_document(const std::vector<std::vector<std::uint8_t>>& elements, const int size_offset = 0)
+{
+    std::vector<std::uint8_t> body;
+    for (const auto& element : elements)
+    {
+        body.insert(body.end(), element.begin(), element.end());
+    }
+    const auto size = static_cast<std::uint32_t>(static_cast<int>(body.size()) + 5 + size_offset);
+    std::vector<std::uint8_t> result = {static_cast<std::uint8_t>(size & 0xFFu), static_cast<std::uint8_t>((size >> 8u) & 0xFFu),
+                                        static_cast<std::uint8_t>((size >> 16u) & 0xFFu), static_cast<std::uint8_t>((size >> 24u) & 0xFFu)
+                                       };
+    result.insert(result.end(), body.begin(), body.end());
+    result.push_back(0x00);
+    return result;
+}
+
+/// a BSON int32 value
+std::vector<std::uint8_t> bson_int32(const std::int32_t value)
+{
+    const auto u = static_cast<std::uint32_t>(value);
+    return {static_cast<std::uint8_t>(u & 0xFFu), static_cast<std::uint8_t>((u >> 8u) & 0xFFu),
+            static_cast<std::uint8_t>((u >> 16u) & 0xFFu), static_cast<std::uint8_t>((u >> 24u) & 0xFFu)};
+}
+
+/// a BSON string value, whose length is @a length_offset off
+std::vector<std::uint8_t> bson_string(const std::string& value, const std::int32_t length_offset = 0)
+{
+    auto result = bson_int32(static_cast<std::int32_t>(value.size() + 1) + length_offset);
+    result.insert(result.end(), value.begin(), value.end());
+    result.push_back(0x00);
+    return result;
+}
+
+/// @a count bytes of value 0xAB
+std::vector<std::uint8_t> bytes(const std::size_t count)
+{
+    return std::vector<std::uint8_t>(count, 0xAB);
+}
+
+/// the bytes of @a parts, one after the other
+std::vector<std::uint8_t> concatenated(std::initializer_list<std::vector<std::uint8_t>> parts)
+{
+    std::vector<std::uint8_t> result;
+    for (const auto& part : parts)
+    {
+        result.insert(result.end(), part.begin(), part.end());
+    }
+    return result;
+}
+
+/// U+FFFD REPLACEMENT CHARACTER
+std::string replacement_character()
+{
+    return "\xEF\xBF\xBD";
+}
+}  // namespace
+
+TEST_CASE("regression test - #3989 SAX parse_error() returning true")
+{
+    SECTION("binary formats complete what was read before the input ends")
+    {
+        const json j = {{"a", {1, -2, {{"b", "c"}}, json::array()}}, {"d", {{"e", nullptr}, {"f", true}}}, {"g", 1.5}, {"h", json::binary({1, 2, 3})}};
+
+        const std::vector<std::pair<json::input_format_t, std::vector<std::uint8_t>>> encodings =
+        {
+            {json::input_format_t::cbor, json::to_cbor(j)},
+            {json::input_format_t::msgpack, json::to_msgpack(j)},
+            {json::input_format_t::ubjson, json::to_ubjson(j)},
+            {json::input_format_t::ubjson, json::to_ubjson(j, true, true)},
+            {json::input_format_t::bjdata, json::to_bjdata(j)},
+            {json::input_format_t::bjdata, json::to_bjdata(j, true, true)},
+            {json::input_format_t::bson, json::to_bson(j)},
+            {json::input_format_t::bon8, json::to_bon8(j)},
+        };
+
+        for (const auto& encoding : encodings)
+        {
+            const auto format = encoding.first;
+            const auto& bytes = encoding.second;
+            CAPTURE(format)
+
+            // every prefix is truncated input
+            for (std::size_t length = 0; length < bytes.size(); ++length)
+            {
+                CAPTURE(length)
+                const auto result = parse_binary_recovering(std::vector<std::uint8_t>(bytes.begin(), bytes.begin() + static_cast<std::ptrdiff_t>(length)), format);
+                CHECK(!result.ok);
+                CHECK(result.errors == 1);
+                CHECK(result.balanced);
+            }
+
+            // the complete input is read as usual (binary values do not
+            // round-trip through every format, so compare with a plain parse)
+            json expected;
+            nlohmann::detail::json_sax_dom_parser<json> dom(expected);
+            CHECK(json::sax_parse(bytes, &dom, format));
+            const auto complete = parse_binary_recovering(bytes, format);
+            CHECK(complete.ok);
+            CHECK(complete.errors == 0);
+            CHECK(complete.value == expected);
+
+            // a byte after the value
+            auto trailing_bytes = bytes;
+            trailing_bytes.push_back(0x01);
+            const auto trailing = parse_binary_recovering(trailing_bytes, format);
+            CHECK(!trailing.ok);
+            CHECK(trailing.errors == 1);
+            CHECK(trailing.value == expected);
+        }
+    }
+
+    SECTION("containers without an end")
+    {
+        // these made the readers loop, or read on, after the error
+        const auto cbor_array = parse_binary_recovering({0x9F}, json::input_format_t::cbor);
+        CHECK(cbor_array.errors == 1);
+        CHECK(cbor_array.value == json::array());
+
+        const auto cbor_map = parse_binary_recovering({0xBF, 0x61, 'a'}, json::input_format_t::cbor);
+        CHECK(cbor_map.errors == 1);
+        CHECK(cbor_map.value == json({{"a", nullptr}}));
+
+        const auto msgpack_array = parse_binary_recovering({0xDD, 0xFF, 0xFF, 0xFF, 0xFF}, json::input_format_t::msgpack);
+        CHECK(msgpack_array.errors == 1);
+        CHECK(msgpack_array.value == json::array());
+
+        const auto msgpack_map = parse_binary_recovering({0x81, 0xA1, 'a', 0x92, 0x01}, json::input_format_t::msgpack);
+        CHECK(msgpack_map.errors == 1);
+        CHECK(msgpack_map.value == json({{"a", {1}}}));
+    }
+
+    SECTION("BJData ndarray")
+    {
+        // a 2x3 int8 array with two of its six elements; the annotated array
+        // format opens an object and two arrays of its own
+        const auto result = parse_binary_recovering({'[', '$', 'i', '#', '[', '$', 'i', '#', 'i', 2, 2, 3, 1, 2}, json::input_format_t::bjdata);
+        CHECK(result.errors == 1);
+        CHECK(result.balanced);
+        CHECK(result.value == json({{"_ArrayType_", "int8"}, {"_ArraySize_", {2, 3}}, {"_ArrayData_", {1, 2}}}));
+    }
+
+    SECTION("binary formats repair items whose end is known")
+    {
+        struct Repair
+        {
+            json::input_format_t format;
+            std::vector<std::uint8_t> input;
+            json expected;
+            std::size_t errors;
+        };
+
+        const std::vector<Repair> repairs =
+        {
+            // CBOR: tags are ignored (here tag 1 and the self-describe tag 55799)
+            {json::input_format_t::cbor, {0x82, 0xC1, 0x05, 0xD9, 0xD9, 0xF7, 0x06}, {5, 6}, 2},
+            // CBOR: undefined and other simple values become null
+            {json::input_format_t::cbor, {0x84, 0xF7, 0xE0, 0xF8, 0x20, 0x01}, {nullptr, nullptr, nullptr, 1}, 3},
+            // CBOR: members whose key is not a string are skipped, whatever their key and value
+            {json::input_format_t::cbor, {0xA4, 0x01, 0x02, 0x82, 0x01, 0x02, 0xA1, 0x61, 'x', 0x9F, 0xFF, 0xC1, 0x01, 0x5F, 0x41, 0x00, 0xFF, 0x61, 'a', 0x03}, {{"a", 3}}, 3},
+            {json::input_format_t::cbor, {0xBF, 0xF5, 0xBF, 0x61, 'x', 0x7F, 0x61, 'y', 0xFF, 0xFF, 0x61, 'a', 0x03, 0xFF}, {{"a", 3}}, 1},
+            // MessagePack: members whose key is not a string are skipped
+            {json::input_format_t::msgpack, {0x84, 0x01, 0x02, 0x81, 0xA1, 'x', 0x01, 0x92, 0x01, 0x02, 0xD4, 0x01, 0x02, 0xC0, 0xA1, 'a', 0x04}, {{"a", 4}}, 3},
+            // UBJSON: a char that is not ASCII becomes U+FFFD
+            {json::input_format_t::ubjson, {'[', 'C', 0x80, 'C', 'A', ']'}, {replacement_character(), "A"}, 1},
+            // UBJSON: the longest beginning of a high-precision number is kept
+            {json::input_format_t::ubjson, {'[', 'H', 'i', 5, '1', '2', 'a', 'b', 'c', 'H', 'i', 2, '1', '.', 'H', 'i', 3, 'a', 'b', 'c', 'H', 'i', 3, '4', '.', '5', ']'}, {12, 1, nullptr, 4.5}, 3},
+            // BJData, too
+            {json::input_format_t::bjdata, {'[', 'C', 0xFF, 'H', 'i', 2, '-', '1', 'H', 'i', 2, '-', 'x', ']'}, {replacement_character(), -1, nullptr}, 2},
+            // a NUL ends a high-precision number, as it ends JSON text
+            {json::input_format_t::ubjson, {'[', 'H', 'i', 4, '1', '2', 0, '9', 'H', 'i', 2, 0, '1', 'i', 3, ']'}, {12, nullptr, 3}, 2},
+            // BON8: members whose key is not a string are skipped
+            {json::input_format_t::bon8, {0x89, 0x91, 0x92, 0xC9, 0x40, 0x82, 0x91, 0x92, 0x61, 0x93}, {{"a", 3}}, 2},
+            {json::input_format_t::bon8, {0x8B, 0x91, 0x85, 0x91, 0xFE, 0xFA, 0x8B, 'x', 0x91, 0xFE, 0x61, 0x93, 0xFE}, {{"a", 3}}, 2},
+            // BSON: elements of types the library does not read become null
+            {
+                json::input_format_t::bson, bson_document(
+                {
+                    bson_element(0x07, "_id", bytes(12)),                         // ObjectId
+                    bson_element(0x09, "date", bytes(8)),                         // UTC datetime
+                    bson_element(0x13, "decimal", bytes(16)),                     // 128-bit decimal
+                    bson_element(0x0B, "regex", {'a', '+', 0, 'i', 0}),           // regular expression
+                    bson_element(0x0D, "code", bson_string("f()")),               // JavaScript code
+                    bson_element(0x0E, "symbol", bson_string("s")),               // symbol
+                    bson_element(0x0C, "pointer", concatenated({bson_string("c"), bytes(12)})), // DBPointer
+                    bson_element(0x0F, "scope", concatenated({bson_int32(15), bson_string("g"), bson_document({})})), // code with scope
+                    bson_element(0x06, "undefined", {}),                          // undefined
+                    bson_element(0xFF, "min", {}),                                // min key
+                    bson_element(0x7F, "max", {}),                                // max key
+                    bson_element(0x10, "z", bson_int32(7)),
+                }),
+                {{"_id", nullptr}, {"date", nullptr}, {"decimal", nullptr}, {"regex", nullptr}, {"code", nullptr}, {"symbol", nullptr}, {"pointer", nullptr}, {"scope", nullptr}, {"undefined", nullptr}, {"min", nullptr}, {"max", nullptr}, {"z", 7}},
+                11
+            },
+            // BSON: an element of an unknown type becomes null, and the rest of its document is skipped
+            {
+                json::input_format_t::bson, bson_document(
+                {
+                    bson_element(0x03, "inner", bson_document({bson_element(0x10, "a", bson_int32(1)), bson_element(0x42, "x", bytes(3)), bson_element(0x10, "b", bson_int32(2))})),
+                    bson_element(0x04, "array", bson_document({bson_element(0x10, "0", bson_int32(1)), bson_element(0x42, "1", bytes(3))})),
+                    bson_element(0x10, "after", bson_int32(3)),
+                }),
+                {{"inner", {{"a", 1}, {"x", nullptr}}}, {"array", {1, nullptr}}, {"after", 3}},
+                2
+            },
+            // BSON: so does a string or byte array whose length cannot be right
+            {
+                json::input_format_t::bson, bson_document(
+                {
+                    bson_element(0x03, "inner", bson_document({bson_element(0x02, "s", bson_string("abc", -10)), bson_element(0x10, "b", bson_int32(2))})),
+                    bson_element(0x03, "bin", bson_document({bson_element(0x05, "b", concatenated({bson_int32(-1), bytes(1)})), bson_element(0x10, "b", bson_int32(2))})),
+                    bson_element(0x10, "after", bson_int32(3)),
+                }),
+                {{"inner", {{"s", nullptr}}}, {"bin", {{"b", nullptr}}}, {"after", 3}},
+                2
+            },
+            // BSON: a string without its terminator, and a document whose size does not match, are kept
+            {
+                json::input_format_t::bson, bson_document(
+                {
+                    bson_element(0x02, "s", {2, 0, 0, 0, 'a', 'X'}),
+                    bson_element(0x03, "inner", bson_document({bson_element(0x10, "a", bson_int32(1))}, 1)),
+                }),
+                {{"s", "a"}, {"inner", {{"a", 1}}}},
+                2
+            },
+        };
+
+        for (const auto& repair : repairs)
+        {
+            CAPTURE(repair.format)
+            CAPTURE(repair.input)
+            const auto result = parse_binary_recovering(repair.input, repair.format);
+            CHECK(!result.ok);
+            CHECK(result.balanced);
+            CHECK(result.errors == repair.errors);
+            CHECK(result.value == repair.expected);
+            REQUIRE(!result.messages.empty());
+#if !defined(JSON_NOEXCEPTION)
+            // the first error is the one reported without recovering; under
+            // JSON_NOEXCEPTION, reading without recovering aborts instead of
+            // throwing, so there is no message to compare with
+            CHECK(result.messages.front() == binary_error_message(repair.input, repair.format));
+#endif
+        }
+    }
+
+    SECTION("binary formats repair numbers that are out of range")
+    {
+        // CBOR: a double too large for a float number_float_t
+        float_json cbor;
+        BasicRecoveringParser<float_json> sax(cbor);
+        const std::vector<std::uint8_t> cbor_input = {0x82, 0xFB, 0x7E, 0x37, 0xE4, 0x3C, 0x88, 0x00, 0x75, 0x9C, 0x01}; // [1e300, 1]
+        CHECK(!float_json::sax_parse(cbor_input, &sax, float_json::input_format_t::cbor));
+        CHECK(sax.errors == 1);
+        CHECK(sax.messages.front() == "[json.exception.out_of_range.406] syntax error while parsing CBOR value: number overflow");
+        REQUIRE(cbor.size() == 2);
+        CHECK(std::isinf(cbor[0].get<float>()));
+        CHECK(cbor[1] == 1);
+
+        // UBJSON: a high-precision number too large for number_float_t
+        const auto ubjson = parse_binary_recovering({'H', 'i', 5, '1', 'e', '9', '9', '9'}, json::input_format_t::ubjson);
+        CHECK(ubjson.errors == 1);
+        CHECK(ubjson.value.is_number_float());
+        CHECK(std::isinf(ubjson.value.get<double>()));
+    }
+
+    SECTION("binary formats stop where the end of an item is not known")
+    {
+        // a byte that begins no item
+        const auto cbor = parse_binary_recovering({0x82, 0x01, 0x1C, 0x02}, json::input_format_t::cbor);
+        CHECK(cbor.errors == 1);
+        CHECK(cbor.value == json({1}));
+
+        // a key that is no item: the unused MessagePack byte, a CBOR break
+        // in a map of known size, and the end of a BON8 container
+        const auto msgpack = parse_binary_recovering({0x82, 0xA1, 'a', 0x01, 0xC1, 0x02}, json::input_format_t::msgpack);
+        CHECK(msgpack.errors == 1);
+        CHECK(msgpack.value == json({{"a", 1}}));
+        const auto cbor_break = parse_binary_recovering({0xA2, 0x61, 'a', 0x01, 0xFF, 0x02}, json::input_format_t::cbor);
+        CHECK(cbor_break.errors == 1);
+        CHECK(cbor_break.value == json({{"a", 1}}));
+        const auto bon8 = parse_binary_recovering({0x88, 0x61, 0x91, 0xFE}, json::input_format_t::bon8);
+        CHECK(bon8.errors == 1);
+        CHECK(bon8.value == json({{"a", 1}}));
+
+        // an indefinite-length string inside an indefinite-length string
+        const auto nested = parse_binary_recovering({0x82, 0x01, 0x7F, 0x7F, 0x61, 'a', 0xFF, 0xFF}, json::input_format_t::cbor);
+        CHECK(nested.errors == 1);
+        CHECK(nested.value == json({1}));
+
+        // a BJData ndarray whose element type has no name: the object that
+        // holds the ndarray was already begun
+        const auto ndarray = parse_binary_recovering({'[', '$', 0x01, '#', '[', '$', 'i', '#', 'i', 2, 2, 3}, json::input_format_t::bjdata);
+        CHECK(ndarray.errors == 1);
+        CHECK(ndarray.balanced);
+        CHECK(ndarray.value == json::object());
+
+        // a skipped member that the input ends in
+        const auto truncated = parse_binary_recovering({0xA2, 0x01, 0x82, 0x01}, json::input_format_t::cbor);
+        CHECK(truncated.errors == 2);
+        CHECK(truncated.balanced);
+        CHECK(truncated.value == json::object());
+
+        // a BSON element of an unknown type in a document whose size cannot be right
+        const auto bson = parse_binary_recovering(bson_document({bson_element(0x10, "a", bson_int32(1)), bson_element(0x42, "x", bytes(3))}, -10), json::input_format_t::bson);
+        CHECK(bson.errors == 1);
+        CHECK(bson.value == json({{"a", 1}, {"x", nullptr}}));
+    }
+
+    SECTION("changed bytes in binary input")
+    {
+        const json j = {{"a", {1, -2, {{"b", "c"}}, json::array()}}, {"d", {{"e", nullptr}, {"f", true}}}, {"g", 1.5}, {"h", json::binary({1, 2, 3})}, {"i", "\xC3\xA4"}};
+
+        const std::vector<std::pair<json::input_format_t, std::vector<std::uint8_t>>> encodings =
+        {
+            {json::input_format_t::cbor, json::to_cbor(j)},
+            {json::input_format_t::msgpack, json::to_msgpack(j)},
+            {json::input_format_t::ubjson, json::to_ubjson(j)},
+            {json::input_format_t::ubjson, json::to_ubjson(j, true, true)},
+            {json::input_format_t::bjdata, json::to_bjdata(j)},
+            {json::input_format_t::bjdata, json::to_bjdata(j, true, true)},
+            {json::input_format_t::bson, json::to_bson(j)},
+            {json::input_format_t::bon8, json::to_bon8(j)},
+        };
+        const std::vector<std::uint8_t> replacements = {0x00, 0x01, 0x7F, 0x80, 0xC1, 0xD9, 0xE0, 0xF7, 0xFE, 0xFF};
+
+        for (const auto& encoding : encodings)
+        {
+            const auto format = encoding.first;
+            const auto& original = encoding.second;
+            CAPTURE(format)
+
+            std::vector<std::vector<std::uint8_t>> inputs;
+            for (std::size_t position = 0; position < original.size(); ++position)
+            {
+                for (const auto replacement : replacements)
+                {
+                    auto changed = original;
+                    changed[position] = replacement;
+                    inputs.push_back(changed);
+                }
+                auto removed = original;
+                removed.erase(removed.begin() + static_cast<std::ptrdiff_t>(position));
+                inputs.push_back(removed);
+            }
+
+            for (const auto& input : inputs)
+            {
+                CAPTURE(input)
+                const auto result = parse_binary_recovering(input, format);
+                CHECK(result.balanced);
+                CHECK(result.errors <= input.size() + 1);
+#if !defined(JSON_NOEXCEPTION)
+                // an error is reported exactly if reading into a JSON value
+                // fails, and the first one is the same (under JSON_NOEXCEPTION,
+                // that reading aborts instead of throwing)
+                const auto message = binary_error_message(input, format);
+                CHECK(result.ok == message.empty());
+                if (!result.ok && result.errors < 100)
+                {
+                    CHECK(result.messages.front() == message);
+                }
+#endif
+            }
+        }
+    }
+
+    SECTION("JSON text")
+    {
+        // the parser stopped, but reported success
+        json j;
+        RecoveringParser sax(j);
+        CHECK(!json::sax_parse("[1,2,3,]", &sax));
+        CHECK(sax.errors == 1);
+        CHECK(j == json({1, 2, 3}));
+    }
+
+    SECTION("the SAX parsers of the library stop")
+    {
+        json _;
+        CHECK(json::from_cbor(std::vector<std::uint8_t> {0x9F}, true, false).is_discarded());
+        CHECK_THROWS_WITH_AS(_ = json::from_cbor(std::vector<std::uint8_t> {0x9F}), "[json.exception.parse_error.110] parse error at byte 2: syntax error while parsing CBOR value: unexpected end of input", json::parse_error&);
+        CHECK(json::parse("[1,2,3,]", nullptr, false).is_discarded());
+        CHECK(!json::accept("[1,2,3,]"));
+    }
+}
+
+
+TEST_CASE("issue #5676 - SAX parsing of CBOR tags")
+{
+    const json expected = json::binary({1, 2, 3}, 42);
+    const auto cbor = json::to_cbor(expected);
+
+    nlohmann::detail::json_sax_acceptor<json> acceptor;
+    CHECK_FALSE(json::sax_parse(cbor, &acceptor, json::input_format_t::cbor));
+    CHECK_FALSE(json::sax_parse(cbor, &acceptor, json::input_format_t::cbor,
+                                true, false, false, json::cbor_tag_handler_t::error));
+
+    CHECK(json::sax_parse(cbor, &acceptor, json::input_format_t::cbor,
+                          true, false, false, json::cbor_tag_handler_t::ignore));
+
+    json parsed;
+    nlohmann::detail::json_sax_dom_parser<json, nlohmann::detail::string_input_adapter_type> sax(parsed);
+    CHECK(json::sax_parse(cbor, &sax, json::input_format_t::cbor,
+                          true, false, false, json::cbor_tag_handler_t::store));
+    CHECK(parsed == expected);
+
+    json iterator_parsed;
+    nlohmann::detail::json_sax_dom_parser<json, nlohmann::detail::string_input_adapter_type> iterator_sax(iterator_parsed);
+    CHECK(json::sax_parse(cbor.begin(), cbor.end(), &iterator_sax, json::input_format_t::cbor,
+                          true, false, false, json::cbor_tag_handler_t::store));
+    CHECK(iterator_parsed == expected);
+
+#ifndef JSON_TEST_DEPRECATED_FUNCTIONS_DELETED
+    json span_parsed;
+    nlohmann::detail::json_sax_dom_parser<json, nlohmann::detail::string_input_adapter_type> span_sax(span_parsed);
+    CHECK(json::sax_parse(nlohmann::detail::span_input_adapter(cbor.data(), cbor.size()), &span_sax,
+                          json::input_format_t::cbor, true, false, false, json::cbor_tag_handler_t::store));
+    CHECK(span_parsed == expected);
+#endif
+
+    const std::string text = "null";
+    CHECK(json::sax_parse(text, &acceptor, json::input_format_t::json,
+                          true, false, false, json::cbor_tag_handler_t::store));
+    CHECK(json::sax_parse(text.begin(), text.end(), &acceptor, json::input_format_t::json,
+                          true, false, false, json::cbor_tag_handler_t::store));
+#ifndef JSON_TEST_DEPRECATED_FUNCTIONS_DELETED
+    CHECK(json::sax_parse(nlohmann::detail::span_input_adapter(text.data(), text.size()), &acceptor,
+                          json::input_format_t::json, true, false, false, json::cbor_tag_handler_t::store));
+#endif
+}
+
+
+TEST_CASE("MessagePack SAX aborts")
+{
+    SECTION("start_array(len)")
+    {
+        std::vector<uint8_t> const v = {0x93, 0x01, 0x02, 0x03};
+        SaxCountdown scp(0);
+        CHECK(!json::sax_parse(v, &scp, json::input_format_t::msgpack));
+    }
+
+    SECTION("start_object(len)")
+    {
+        std::vector<uint8_t> const v = {0x81, 0xa3, 0x66, 0x6F, 0x6F, 0xc2};
+        SaxCountdown scp(0);
+        CHECK(!json::sax_parse(v, &scp, json::input_format_t::msgpack));
+    }
+
+    SECTION("key()")
+    {
+        std::vector<uint8_t> const v = {0x81, 0xa3, 0x66, 0x6F, 0x6F, 0xc2};
+        SaxCountdown scp(1);
+        CHECK(!json::sax_parse(v, &scp, json::input_format_t::msgpack));
+    }
+}
+
+TEST_CASE("issue #5405 - a user-defined SAX consumer is unaffected by the internal DOM reserve optimization")
+{
+    // the reserve() call is local to json_sax_dom_parser / json_sax_dom_callback_parser;
+    // a custom SAX consumer that does not touch a DOM array sees identical events
+    json j = json::array();
+    for (int i = 0; i < 100; ++i)
+    {
+        j.push_back(i);
+    }
+    const auto packed = json::to_msgpack(j);
+
+    SaxCountdown scp(1000000); // large enough to never trigger an abort
+    CHECK(json::sax_parse(packed, &scp, json::input_format_t::msgpack));
+}
+
+TEST_CASE("MessagePack nesting does not consume the call stack - SAX interface")
+{
+    // see the test case of the same name in unit-msgpack.cpp (#5104)
+    std::vector<uint8_t> input(300000, 0x91);
+    input.push_back(0x01); // innermost value
+
+    SaxCountdown accept_all(600001);
+    CHECK(json::sax_parse(input, &accept_all, json::input_format_t::msgpack));
+}
+
+TEST_CASE("MessagePack SAX parsing stops at every event")
+{
+    // Containers are opened and closed by the loop that reads them; a SAX
+    // handler that rejects any event - including the end of a nested
+    // container - must stop the parse right there.
+    const auto count_events = [](const std::vector<std::uint8_t>& input)
+    {
+        int events = 0;
+        while (true)
+        {
+            SaxCountdown scp(events);
+            if (json::sax_parse(input, &scp, json::input_format_t::msgpack))
+            {
+                return events;
+            }
+            ++events;
+            REQUIRE(events < 1000);
+        }
+    };
+
+    // 20 events: every container kind closes inside another one
+    const json j = json::parse(R"({"a": [1, {"b": []}], "c": {"d": [[2]]}})");
+    CHECK(count_events(json::to_msgpack(j)) == 20);
+}
+
+DOCTEST_CLANG_SUPPRESS_WARNING_POP