json_view: decode chains of unicode escapes in a loop

After the first \uXXXX escape of a string, slow_string() now decodes the
escapes that follow it directly, while the next six bytes are \u and four hex
digits that are not a surrogate, instead of going round its main loop for each
of them (that is how Python's json.dumps writes non-ASCII text by default).
A surrogate, a bad digit, or fewer than six bytes leave the position at the
backslash: the main loop handles them as before and reports the same error at
the same position.

twitterescaped.json (simdjson-data, 549 KB, 33,036 escapes), json_document::parse(),
A/B in one binary, 6 code layouts, time after / before:

- Apple M1 Max, Apple clang 21: 0.897
- x86-64 Haswell, GCC 16.2: 0.907
- x86-64 Haswell, Clang 23: 0.910

The other files of the benchmark set (twitter, citm_catalog, canada, gsoc-2018,
apache_builds, instruments, tree-pretty, github_events, random, update-center,
semanticscholar-corpus, twitter_timeline) stay within the noise of each machine.

New test: chains of every length from 1 to 40 of escapes that give one, two,
and three bytes of UTF-8, followed by the end of the string, text, escapes,
surrogates, bad digits, and cut in every place, in keys and in values, against
json::parse().

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

docs/mkdocs/docs/home/debugging.md

@@ -10,13 +10,6 @@ at its root, a [Natvis](https://learn.microsoft.com/en-us/visualstudio/debugger/
 file that gives `json`/`ordered_json` values a friendly, key/value debugger view instead of showing raw internal
 fields, when debugging with the MSVC debug engine (`cppvsdbg`) in Visual Studio or VS Code.
 
-The file also has visualizers for `json_document` and `json_view` (and their `ordered_json`/editable variants). They
-show the value of the document or view, and its elements as an expandable list. Because a document is only a flat index
-into the source text, the strings of the direct elements are shown, but the elements of nested arrays and objects are
-shown as index nodes (their kind, size, and numbers; strings only by length). Arrays and objects that have been edited
-in a `json_editable_document` are not expanded. The visualizers read the internal layout of the MSVC STL (`std::array`,
-`std::unique_ptr`).
-
 The checked-in visualizers cover namespace variants with zero, one, or two ABI tags. Configurations with three or more
 ABI tags are not covered and may show raw internal fields instead. To generate a Natvis file for such a configuration:
 
@@ -36,7 +29,7 @@ ABI tags are not covered and may show raw internal fields instead. To generate a
 
 4. Use the generated file instead of `nlohmann_json.natvis`.
 
-`--abi-tags` can be repeated for several configurations, and `--all` generates every combination (about 60 MB for 11
+`--abi-tags` can be repeated for several configurations, and `--all` generates every combination (about 60 MB for 12
 ABI tags, doubling with each further tag). See
 [`tools/generate_natvis`](https://github.com/nlohmann/json/tree/develop/tools/generate_natvis) for all options.
 

include/nlohmann/detail/view/builder.hpp

@@ -343,6 +343,21 @@ class builder
                 }
                 arena_ensure(w, w_end, 4);
                 w = put_utf8(w, cp);
+                // A chain of unicode escapes ("\u3042\u3044\u3046", as Python's json.dumps writes non-ASCII text) is
+                // decoded here, without going round the loop above for each of them. Anything else (a surrogate, a
+                // bad digit, fewer than six bytes) leaves p at the backslash: the loop above handles it, and reports
+                // the same error at the same position as it would without this loop.
+                while (end - p >= 6 && p[0] == '\\' && p[1] == 'u')
+                {
+                    const int next = hex_codepoint(p + 2);
+                    if (next < 0 || (next & 0xF800) == 0xD800)
+                    {
+                        break;
+                    }
+                    p += 6;
+                    arena_ensure(w, w_end, 4);
+                    w = put_utf8(w, next);
+                }
             }
             else if (NLOHMANN_VIEW_LIKELY(d < 128 && simple_escape[d] != 0))
             {

single_include/nlohmann/json_view.hpp

@@ -1585,6 +1585,21 @@ class builder
                 }
                 arena_ensure(w, w_end, 4);
                 w = put_utf8(w, cp);
+                // A chain of unicode escapes ("\u3042\u3044\u3046", as Python's json.dumps writes non-ASCII text) is
+                // decoded here, without going round the loop above for each of them. Anything else (a surrogate, a
+                // bad digit, fewer than six bytes) leaves p at the backslash: the loop above handles it, and reports
+                // the same error at the same position as it would without this loop.
+                while (end - p >= 6 && p[0] == '\\' && p[1] == 'u')
+                {
+                    const int next = hex_codepoint(p + 2);
+                    if (next < 0 || (next & 0xF800) == 0xD800)
+                    {
+                        break;
+                    }
+                    p += 6;
+                    arena_ensure(w, w_end, 4);
+                    w = put_utf8(w, next);
+                }
             }
             else if (NLOHMANN_VIEW_LIKELY(d < 128 && simple_escape[d] != 0))
             {

tests/src/unit-json_view_builder.cpp

@@ -575,6 +575,59 @@ TEST_CASE("json_view builder: escapes between runs of text")
     }
 }
 
+TEST_CASE("json_view builder: chains of unicode escapes")
+{
+    // Consecutive \uXXXX escapes are decoded in a loop of their own. Chains of
+    // every length from 1 to 40 of escapes that give one, two, and three
+    // bytes of UTF-8, in upper and lower case, followed by what ends a chain:
+    // the end of the string, text, another escape, UTF-8, and everything that
+    // is not a valid escape (a bad digit at every position of the last
+    // escape, a lone surrogate, a surrogate pair, a cut in every place).
+    const std::array<const char*, 8> chain = {{"\\u0041", "\\u00e4", "\\u00E4", "\\u20ac", "\\u3042", "\\uFFFF", "\\u0000", "\\u07ff"}};
+    const std::array<const char*, 12> ends = {{"", "abc", "\\n", "\\u00e9", "\\ud83d\\ude00", "\\ud83d", "\\ude00", "\\ud83d\\u0041", "\\u12G4", "\\u12", "\xc3\xa4", "\\"}};
+    std::mt19937 rng(2027); // NOLINT(cert-msc32-c,cert-msc51-cpp,bugprone-random-generator-seed): reproducible
+    for (std::size_t length = 1; length <= 40; ++length)
+    {
+        for (int i = 0; i < 20; ++i)
+        {
+            std::string content = std::string(rng() % 20, 'a');
+            for (std::size_t k = 0; k < length; ++k)
+            {
+                content += chain[rng() % chain.size()];
+            }
+            const std::string whole = content;
+            for (const char* end : ends)
+            {
+                check_string(whole + end);
+                check_string(whole + end + std::string(rng() % 19, 'z'));
+            }
+            // a bad digit at every position of every escape near the end of the chain
+            for (std::size_t back = 1; back <= 6 && back <= length; ++back)
+            {
+                for (std::size_t digit = 2; digit < 6; ++digit)
+                {
+                    std::string broken = whole;
+                    broken[broken.size() - (6 * back) + digit] = "gG\"\x01 z"[rng() % 6];
+                    check_string(broken);
+                }
+            }
+            // the text cut in every place of the last three escapes
+            for (std::size_t cut = 0; cut <= 18 && cut <= whole.size(); ++cut)
+            {
+                check_string(whole.substr(0, whole.size() - cut));
+            }
+        }
+    }
+    // a chain longer than the first growth of the arena, in a key and in a value
+    std::string long_chain;
+    for (int i = 0; i < 3000; ++i)
+    {
+        long_chain += chain[static_cast<std::size_t>(i) % chain.size()];
+    }
+    check_string(long_chain);
+    check_string(long_chain + "\\ud83d\\ude00" + long_chain);
+}
+
 TEST_CASE("json_view node integer bits")
 {
     using nlohmann::detail::view::integer_bits;

tools/generate_natvis/README.md

@@ -6,11 +6,6 @@ The checked-in `nlohmann_json.natvis` covers the default namespace (no ABI tags)
 of ABI tags, with both versioned and unversioned namespace variants. Configurations that enable three or more ABI tags
 are not covered by this reduced file; use `--abi-tags` or `--all` (see below) to generate a file for them.
 
-The visualizers of `basic_json_document` and `basic_json_view` live in the same inline namespaces as `basic_json`, so the
-ABI tags apply to them in the same way, and the `view_mode` template argument is covered by a wildcard: it does not
-multiply the namespaces. Each namespace gets three more types (`basic_json_view<*,*>`, `basic_json_document<*,*>`, and
-`detail::view::node`), which roughly doubles the size of the file.
-
 The ABI tag list and the library version are parsed from
 `include/nlohmann/detail/abi_macros.hpp`, so this script must be re-run (via
 `make natvis`) whenever an `NLOHMANN_JSON_ABI_TAG_*` macro is added to that
@@ -40,7 +35,7 @@ version other than the one in `abi_macros.hpp`.
 |-------------------|-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
 | `--list-tags`     | Print every ABI tag (in namespace order) with the macro that enables it and the `NLOHMANN_JSON_ABI_TAG_*` identifier, then exit.                                                                                                                    |
 | `--abi-tags TAGS` | Add the namespaces of one configuration, given as a comma-separated list of ABI tags (`_diag,_dp,_ns`; the leading underscore is optional, and the order does not matter). Can be repeated for several configurations. Unknown tags are an error. |
-| `--all`           | Generate the namespaces for every combination of ABI tags. The file size doubles with each tag (about 60 MB for 11 tags). Cannot be combined with `--abi-tags`.                                                                                    |
+| `--all`           | Generate the namespaces for every combination of ABI tags. The file size doubles with each tag (about 60 MB for 12 tags). Cannot be combined with `--abi-tags`.                                                                                    |
 | `--output FILE`   | Write the Natvis file to `FILE` instead of `nlohmann_json.natvis` in the repository root.                                                                                                                                                           |
 
 Each configuration gets both the versioned (`json_abi_diag_dp_ns_v3_12_0`) and the unversioned (`json_abi_diag_dp_ns`,

tools/generate_natvis/generate_natvis.py

@@ -95,7 +95,7 @@ if __name__ == '__main__':
                                 'optional) of one configuration to add to the default namespaces; can be repeated')
     selection.add_argument('--all', action='store_true',
                            help='Generate the namespaces for every combination of ABI tags; the file size doubles '
-                                'with each ABI tag (about 60 MB for 11 tags)')
+                                'with each ABI tag (about 60 MB for 12 tags)')
     parser.add_argument('--list-tags', action='store_true',
                         help='List the ABI tags with the macros that enable them, and exit')
     parser.add_argument('--output', metavar='FILE',

tools/generate_natvis/nlohmann_json.natvis.j2

@@ -5,39 +5,6 @@
 <!-- * * * * * * * * AUTO-GENERATED FILE  * * * * * * * * -->
 
 <AutoVisualizer xmlns="http://schemas.microsoft.com/vstudio/debugger/natvis/2010">
-{# The children of the container node `node` (a pointer expression) of the document `doc` (a document_data pointer
-   expression). A node is only an index into the source text: strings are read from `doc`, so the strings of the
-   direct children are readable here, but the children of a nested container are shown as nodes (see the node
-   visualizer below). Containers with edits (flag 8, "moved") are not expanded. The node kinds are the numbers of
-   node_kind: 1 object, 2 array, 3 string; the float token is shown only by the node visualizer. #}
-{% macro children(node, doc=none) %}
-            <CustomListItems MaxItemsPerView="1000">
-              <Variable Name="nd" InitialValue="{{ node }}" />
-              <Variable Name="obj" InitialValue="nd != 0 &amp;&amp; (int)nd-&gt;kind == 1" />
-              <Variable Name="n" InitialValue="(nd != 0 &amp;&amp; ((int)nd-&gt;kind == 1 || (int)nd-&gt;kind == 2) &amp;&amp; (nd-&gt;flags &amp; 8) == 0) ? (int)nd-&gt;len : 0" />
-              <Variable Name="p" InitialValue="nd + 1" />
-              <Variable Name="i" InitialValue="0" />
-              <Size>obj ? 2 * n : n</Size>
-              <Loop Condition="i &lt; n">
-{% if doc %}
-                <Item Name="key" Condition="obj">{{ doc }}-&gt;base._Elems[p-&gt;flags &amp; 3] + p-&gt;off,[p-&gt;len]s8</Item>
-{% else %}
-                <Item Name="key" Condition="obj">*p</Item>
-{% endif %}
-                <Exec Condition="obj">p = p + 1</Exec>
-{% if doc %}
-                <Item Name="value" Condition="obj &amp;&amp; (int)p-&gt;kind == 3">{{ doc }}-&gt;base._Elems[p-&gt;flags &amp; 3] + p-&gt;off,[p-&gt;len]s8</Item>
-{% endif %}
-                <Item Name="value" Condition="obj{% if doc %} &amp;&amp; (int)p-&gt;kind != 3{% endif %}">*p</Item>
-{% if doc %}
-                <Item Condition="!obj &amp;&amp; (int)p-&gt;kind == 3">{{ doc }}-&gt;base._Elems[p-&gt;flags &amp; 3] + p-&gt;off,[p-&gt;len]s8</Item>
-{% endif %}
-                <Item Condition="!obj{% if doc %} &amp;&amp; (int)p-&gt;kind != 3{% endif %}">*p</Item>
-                <Exec>p = p + (((int)p-&gt;kind == 1 || (int)p-&gt;kind == 2) ? p-&gt;next : 1)</Exec>
-                <Exec>i = i + 1</Exec>
-              </Loop>
-            </CustomListItems>
-{%- endmacro %}
 {% for ns in namespaces %}
     <!-- Namespace {{ ns }} -->
     <Type Name="{{ ns }}::basic_json&lt;*&gt;">
@@ -95,44 +62,5 @@
     </Type>
 
 {% endfor %}
-
-    <!-- json_view / json_document (include/nlohmann/json_view.hpp). They live in the same inline namespaces as
-         basic_json, so the ABI tags need no extra entries; the view_mode is a template argument that the wildcard
-         covers. A view is a pointer to the document's data and a pointer to a node of its flat index; the
-         strings of the source are found through document_data::base (std::array of the MSVC STL: _Elems). -->
-    <Type Name="{{ ns }}::detail::view::node">
-        <DisplayString Condition="(int)kind == 0">null</DisplayString>
-        <DisplayString Condition="(int)kind == 1">object, size={len}</DisplayString>
-        <DisplayString Condition="(int)kind == 2">array, size={len}</DisplayString>
-        <DisplayString Condition="(int)kind == 3">string, length={len}</DisplayString>
-        <DisplayString Condition="(int)kind == 4">boolean: {(flags &amp; 4) != 0}</DisplayString>
-        <DisplayString Condition="(int)kind == 5">int64: {(__int64)((unsigned __int64)len | ((unsigned __int64)next &lt;&lt; 32))}</DisplayString>
-        <DisplayString Condition="(int)kind == 6">uint64: {(unsigned __int64)len | ((unsigned __int64)next &lt;&lt; 32)}</DisplayString>
-        <DisplayString Condition="(int)kind == 7">float, length={len}</DisplayString>
-        <DisplayString>link</DisplayString>
-        <Expand>
-{{ children('this') }}
-        </Expand>
-    </Type>
-
-    <Type Name="{{ ns }}::basic_json_view&lt;*,*&gt;">
-        <DisplayString Condition="m_node == 0">discarded</DisplayString>
-        <DisplayString Condition="(int)m_node-&gt;kind == 3">{m_doc-&gt;base._Elems[m_node-&gt;flags &amp; 3] + m_node-&gt;off,[m_node-&gt;len]s8}</DisplayString>
-        <DisplayString Condition="(int)m_node-&gt;kind == 7">{m_doc-&gt;base._Elems[m_node-&gt;flags &amp; 3] + m_node-&gt;off,[m_node-&gt;len]s8b}</DisplayString>
-        <DisplayString>{*m_node}</DisplayString>
-        <Expand>
-{{ children('m_node', 'm_doc') }}
-        </Expand>
-    </Type>
-
-    <Type Name="{{ ns }}::basic_json_document&lt;*,*&gt;">
-        <DisplayString Condition="m_data._Mypair._Myval2 == 0 || m_data._Mypair._Myval2-&gt;discarded">discarded</DisplayString>
-        <DisplayString>{*m_data._Mypair._Myval2-&gt;nodes}</DisplayString>
-        <Expand>
-            <Item Name="[source]" Condition="m_data._Mypair._Myval2 != 0">m_data._Mypair._Myval2-&gt;src,[m_data._Mypair._Myval2-&gt;size]s8b</Item>
-            <Item Name="[nodes]" Condition="m_data._Mypair._Myval2 != 0">m_data._Mypair._Myval2-&gt;node_count</Item>
-{{ children('m_data._Mypair._Myval2->nodes', 'm_data._Mypair._Myval2') }}
-        </Expand>
-    </Type>
 {% endfor %}
 </AutoVisualizer>