Fix Semgrep findings in Python tooling (#5899)

- generate_docset.py: use SHA-256 instead of SHA-1 for the cache file
  name digest of downloaded images.
- generate_docset.py: suppress dynamic-urllib-use-detected; the URL
  scheme is restricted to http(s) right before the call.
- generate_natvis.py: suppress direct-use-of-jinja2; the template
  renders a local XML file with autoescape enabled.

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

.github/workflows/codeql-analysis.yml

@@ -38,14 +38,14 @@ jobs:
 
     # Initializes the CodeQL tools for scanning.
     - name: Initialize CodeQL
-      uses: github/codeql-action/init@24c54180a607b1449ed407dd24f251e4e9147c8d # v4.38.3
+      uses: github/codeql-action/init@6f2ac81f5cd81fe351a197ba0b00d91d03cd2aee # v4.39.0
       with:
         languages: c-cpp
 
     # Autobuild attempts to build any compiled languages  (C/C++, C#, or Java).
     # If this step fails, then you should remove it and run the build manually (see below)
     - name: Autobuild
-      uses: github/codeql-action/autobuild@24c54180a607b1449ed407dd24f251e4e9147c8d # v4.38.3
+      uses: github/codeql-action/autobuild@6f2ac81f5cd81fe351a197ba0b00d91d03cd2aee # v4.39.0
 
     - name: Perform CodeQL Analysis
-      uses: github/codeql-action/analyze@24c54180a607b1449ed407dd24f251e4e9147c8d # v4.38.3
+      uses: github/codeql-action/analyze@6f2ac81f5cd81fe351a197ba0b00d91d03cd2aee # v4.39.0

.github/workflows/comment_check_amalgamation.yml

@@ -37,6 +37,14 @@ jobs:
             var matchArtifact = artifacts.data.artifacts.filter((artifact) => {
               return artifact.name == "pr"
             })[0];
+            // a run that failed before its jobs started (e.g. a stale queued
+            // run cancelled long after its push) has no artifact to report on
+            if (!matchArtifact) {
+              core.notice('Run ${{ github.event.workflow_run.id }} has no "pr" artifact; nothing to comment on.');
+              core.setOutput('found', 'false');
+              return;
+            }
+            core.setOutput('found', 'true');
             var download = await github.rest.actions.downloadArtifact({
                owner: context.repo.owner,
                repo: context.repo.repo,
@@ -51,9 +59,11 @@ jobs:
       # Extract the untrusted PR artifact into a dedicated empty directory and
       # read only the two expected files by fixed path afterwards. This avoids a
       # malicious archive overwriting workspace files or escaping via ../ paths.
-      - run: unzip -o pr.zip -d ./pr_artifact
+      - if: steps.download.outputs.found == 'true'
+        run: unzip -o pr.zip -d ./pr_artifact
 
       - name: 'Comment on PR'
+        if: steps.download.outputs.found == 'true'
         uses: actions/github-script@3a2844b7e9c422d3c10d287c895573f7108da1b3 # v9.0.0
         with:
           github-token: ${{ secrets.GITHUB_TOKEN }}

.github/workflows/flawfinder.yml

@@ -47,6 +47,6 @@ jobs:
           output: 'flawfinder_results.sarif'
 
       - name: Upload analysis results to GitHub Security tab
-        uses: github/codeql-action/upload-sarif@24c54180a607b1449ed407dd24f251e4e9147c8d # v4.38.3
+        uses: github/codeql-action/upload-sarif@6f2ac81f5cd81fe351a197ba0b00d91d03cd2aee # v4.39.0
         with:
           sarif_file: ${{github.workspace}}/flawfinder_results.sarif

.github/workflows/json_view_benchmarks.yml

@@ -54,7 +54,7 @@ jobs:
     timeout-minutes: 60
     steps:
       - name: Harden Runner
-        uses: step-security/harden-runner@e14015d583714f6e62063499dc959a02595150a1 # v2.21.1
+        uses: step-security/harden-runner@ccd8616d44fd3846e67624a50d5aad6d37bf2d25 # v2.22.1
         with:
           egress-policy: audit
 
@@ -78,7 +78,7 @@ jobs:
       - name: Summary
         run: cat tests/benchmarks/json_view/results/*.md >> "$GITHUB_STEP_SUMMARY"
 
-      - uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
+      - uses: actions/upload-artifact@cf430e030ddbb5b0abf93d22962f4752f3646cd9 # v7.0.2
         with:
           name: json_view-benchmarks-${{ matrix.runner }}-${{ inputs.compiler || 'g++' }}
           path: tests/benchmarks/json_view/results/

.github/workflows/scorecards.yml

@@ -80,6 +80,6 @@ jobs:
 
       # Upload the results to GitHub's code scanning dashboard.
       - name: "Upload to code-scanning"
-        uses: github/codeql-action/upload-sarif@24c54180a607b1449ed407dd24f251e4e9147c8d # v4.38.3
+        uses: github/codeql-action/upload-sarif@6f2ac81f5cd81fe351a197ba0b00d91d03cd2aee # v4.39.0
         with:
           sarif_file: results.sarif

.github/workflows/semgrep.yml

@@ -65,7 +65,7 @@ jobs:
 
       # Upload SARIF file generated in previous step
       - name: Upload SARIF file
-        uses: github/codeql-action/upload-sarif@24c54180a607b1449ed407dd24f251e4e9147c8d # v4.38.3
+        uses: github/codeql-action/upload-sarif@6f2ac81f5cd81fe351a197ba0b00d91d03cd2aee # v4.39.0
         with:
           sarif_file: semgrep.sarif
         if: always()

docs/docset/generate_docset.py

@@ -341,13 +341,14 @@ def download(url, docs) -> str:
         raise ValueError(f'not an http(s) URL: {url}')
     req = urllib.request.Request(u.geturl(), headers={'User-Agent': USER_AGENT})
     # (the scheme is checked above)
+    # nosemgrep: python.lang.security.audit.dynamic-urllib-use-detected.dynamic-urllib-use-detected
     with urllib.request.urlopen(req, timeout=20) as r:  # nosec B310
         data = r.read()
         ctype = r.headers.get_content_type()
     path = urllib.parse.unquote(u.path).lstrip('/')
     ext = os.path.splitext(path)[1]
     if u.query or not ext or path.endswith('/'):
-        digest = hashlib.sha1(url.encode(), usedforsecurity=False).hexdigest()[:12]
+        digest = hashlib.sha256(url.encode()).hexdigest()[:12]
         path = os.path.join(os.path.dirname(path), digest + CONTENT_TYPE_EXT.get(ctype, ext or '.bin'))
     rel = os.path.normpath(os.path.join('assets', 'external', u.hostname, path))
     out = os.path.join(docs, rel)

docs/mkdocs/docs/home/architecture.md

@@ -339,6 +339,11 @@ The `serializer` (used by [`dump`](../api/basic_json/dump.md) and [`operator<<`]
 - `output_stream_adapter` writes to a `std::ostream`.
 - `output_string_adapter` appends to a string.
 
+The `binary_writer` does not call an adapter for every byte. The `to_*` functions that return a `std::vector` write
+into the vector through a sink (`output_vector_sink`) that is not virtual. The functions that take an output adapter
+write through a sink (`output_adapter_sink`) that collects the output in a buffer of one kibibyte and passes it to the
+adapter in blocks, so that the adapter is called (and a stream locked) once per block instead of once per value.
+
 ## Value conversion
 
 Values are converted from and to other types with the `JSONSerializer` template parameter. The default,

docs/mkdocs/scripts/mermaid/package-lock.json

@@ -6,7 +6,7 @@
     "": {
       "name": "check-mermaid",
       "dependencies": {
-        "jsdom": "30.1.2",
+        "jsdom": "30.1.3",
         "mermaid": "11.17.2"
       }
     },
@@ -40,9 +40,9 @@
       }
     },
     "node_modules/@asamuzakjp/dom-selector": {
-      "version": "9.2.4",
-      "resolved": "https://registry.npmjs.org/@asamuzakjp/dom-selector/-/dom-selector-9.2.4.tgz",
-      "integrity": "sha512-YnVzxDxLaqL0t7Q4wfcgHZjg55Wm6y5EaO7Bhu3q9J7wq/8iw3PhsnWiOQhYekx4VexnbkgWpTT6n+QIV7kczw==",
+      "version": "10.0.3",
+      "resolved": "https://registry.npmjs.org/@asamuzakjp/dom-selector/-/dom-selector-10.0.3.tgz",
+      "integrity": "sha512-3+777rEPsOOjoy5/sRmIzHnoRDaLpm3XWY2dmWn/N7f/XwDG0bcpnFaKL3X9FYJUbNwLmpcJC1bUYexJRxsFlw==",
       "license": "MIT",
       "dependencies": {
         "bidi-js": "^1.1.0",
@@ -1202,13 +1202,13 @@
       "license": "MIT"
     },
     "node_modules/jsdom": {
-      "version": "30.1.2",
-      "resolved": "https://registry.npmjs.org/jsdom/-/jsdom-30.1.2.tgz",
-      "integrity": "sha512-0FFE/jE1rppmVfUrJUgxqXjcwolZYIGtAgj1pTussMhNZxIxi7W/PvfWaMNroGi2B36X1IKHbexpZ9DhkoOPiQ==",
+      "version": "30.1.3",
+      "resolved": "https://registry.npmjs.org/jsdom/-/jsdom-30.1.3.tgz",
+      "integrity": "sha512-hQMUPvRgN7KxsfIytDQji7bimuJeRQa6Qo5+iuiZQk5CafeRxLIA10/cG5YCBFNpD9RO3oqIiN4CO1Y44kxqXg==",
       "license": "MIT",
       "dependencies": {
         "@asamuzakjp/css-color": "^7.1.3",
-        "@asamuzakjp/dom-selector": "^9.2.4",
+        "@asamuzakjp/dom-selector": "^10.0.3",
         "@bramus/specificity": "^2.4.2",
         "@csstools/css-syntax-patches-for-csstree": "^1.1.15",
         "@exodus/bytes": "^1.16.0",
@@ -1221,7 +1221,7 @@
         "parse5": "^8.0.1",
         "saxes": "^6.0.0",
         "tough-cookie": "^6.0.2",
-        "undici": "^8.11.2",
+        "undici": "^8.12.1",
         "w3c-xmlserializer": "^6.0.0",
         "webidl-conversions": "^8.0.1",
         "whatwg-mimetype": "^5.0.0",
@@ -1521,9 +1521,9 @@
       }
     },
     "node_modules/undici": {
-      "version": "8.11.2",
-      "resolved": "https://registry.npmjs.org/undici/-/undici-8.11.2.tgz",
-      "integrity": "sha512-u4UB2/IrKdU6lFxumHmmo1a3fCQO5tzQllRorfoRS63txhrB7xTpSn1PftwC4qEHkOaqP95fCWW4lJzwErwzhQ==",
+      "version": "8.12.2",
+      "resolved": "https://registry.npmjs.org/undici/-/undici-8.12.2.tgz",
+      "integrity": "sha512-XtuDWVdqTLr3NAxm2ly9/hU6kUsJOgGh1RASzMBbsw6aWCOaQzHdW+QQDfdERjfJ2V5vL5pRPa4RetO9UfivDg==",
       "license": "MIT",
       "engines": {
         "node": ">=22.19.0"

docs/mkdocs/scripts/mermaid/package.json

@@ -4,7 +4,7 @@
   "description": "Validate the Mermaid diagrams of the documentation (see check_mermaid.mjs)",
   "type": "module",
   "dependencies": {
-    "jsdom": "30.1.2",
+    "jsdom": "30.1.3",
     "mermaid": "11.17.2"
   }
 }

include/nlohmann/detail/input/input_adapters.hpp

@@ -8,17 +8,18 @@
 
 #pragma once
 
-#include <algorithm> // count, find, min
+#include <algorithm> // count, find, max, min
 #include <array> // array
 #include <cstddef> // size_t
 #include <cstdint> // uint32_t
 #include <cstring> // strlen
 #include <iterator> // begin, end, iterator_traits, random_access_iterator_tag, distance, next, reverse_iterator
 #include <limits> // numeric_limits
+#include <memory> // unique_ptr
 #include <streambuf> // streambuf
 #include <string> // string, char_traits
 #include <type_traits> // enable_if, is_base_of, is_pointer, is_integral, remove_pointer
-#include <utility> // pair, declval
+#include <utility> // declval, move, pair
 #include <vector> // vector
 
 #ifndef JSON_NO_IO
@@ -72,6 +73,82 @@ enum class input_format_t { json, cbor, msgpack, ubjson, bson, bjdata, bon8 };
 
 #ifndef JSON_NO_IO
 /*!
+A growable array of characters for the part of a token that an input adapter
+copied out of its window: unlike std::vector<char>::push_back(), appending a
+character takes no reference to a local, and is a few instructions when there
+is room.
+*/
+class token_carry_buffer
+{
+  public:
+    token_carry_buffer() = default;
+    token_carry_buffer(const token_carry_buffer&) = delete;
+    token_carry_buffer& operator=(const token_carry_buffer&) = delete;
+    token_carry_buffer(token_carry_buffer&& rhs) noexcept
+        : m_data(std::move(rhs.m_data)), m_end(rhs.m_end), m_capacity_end(rhs.m_capacity_end)
+    {
+        rhs.m_end = rhs.m_capacity_end = nullptr;
+    }
+    token_carry_buffer& operator=(token_carry_buffer&&) = delete;
+    ~token_carry_buffer() = default;
+
+    void clear() noexcept
+    {
+        m_end = m_data.get();
+    }
+
+    void push_back(char c)
+    {
+        if (JSON_HEDLEY_UNLIKELY(m_end == m_capacity_end))
+        {
+            grow(1);
+        }
+        *m_end++ = c;
+    }
+
+    void append(const char* first, const char* last)
+    {
+        const auto n = static_cast<std::size_t>(last - first);
+        const auto room = static_cast<std::size_t>(m_capacity_end - m_end);
+        if (n > room)
+        {
+            grow(n);
+        }
+        if (n != 0)
+        {
+            std::memcpy(m_end, first, n);
+        }
+        m_end += n;
+    }
+
+    /// replace the contents of @a out
+    void copy_to(std::vector<char>& out) const
+    {
+        out.assign(m_data.get(), m_end);
+    }
+
+  private:
+    /// make room for @a extra more characters
+    JSON_HEDLEY_NEVER_INLINE void grow(std::size_t extra)
+    {
+        const auto size = static_cast<std::size_t>(m_end - m_data.get());
+        const auto capacity = static_cast<std::size_t>(m_capacity_end - m_data.get());
+        const std::size_t new_capacity = (std::max)(capacity < 32 ? static_cast<std::size_t>(32) : 2 * capacity, size + extra);
+        std::unique_ptr<char[]> data(new char[new_capacity]); // NOLINT(cppcoreguidelines-avoid-c-arrays,hicpp-avoid-c-arrays,modernize-avoid-c-arrays)
+        if (size != 0)
+        {
+            std::memcpy(data.get(), m_data.get(), size);
+        }
+        m_data = std::move(data);
+        m_end = m_data.get() + size;
+        m_capacity_end = m_data.get() + new_capacity;
+    }
+
+    std::unique_ptr<char[]> m_data{}; // NOLINT(cppcoreguidelines-avoid-c-arrays,hicpp-avoid-c-arrays,modernize-avoid-c-arrays,readability-redundant-member-init)
+    char* m_end = nullptr;
+    char* m_capacity_end = nullptr;
+};
+/*!
 Input adapter for stdio file access. This adapter read only 1 byte and do not use any
  buffer. This adapter is a very low level adapter.
 */
@@ -209,6 +286,7 @@ class input_stream_adapter
         , token_segment(rhs.token_segment)
         , current_in_get_area(rhs.current_in_get_area)
         , tracking_token(rhs.tracking_token)
+        , no_get_area(rhs.no_get_area)
 #endif
     {
         rhs.is = nullptr;
@@ -295,37 +373,20 @@ class input_stream_adapter
     std::char_traits<char>::int_type get_character()
     {
 #if !JSON_DISABLE_STREAM_BULK_SCAN
-        // sbumpc() refills the get area if it is exhausted
-        const bool refill = streambuf_access::get_ptr(sb) == streambuf_access::get_end(sb);
-        if (JSON_HEDLEY_UNLIKELY(refill))
-        {
-            save_token_segment();
-        }
-#endif
-
-        auto res = sb->sbumpc();
-        // set eof manually, as we don't use the istream interface.
-        if (JSON_HEDLEY_UNLIKELY(res == std::char_traits<char>::eof()))
-        {
-#if !JSON_DISABLE_STREAM_BULK_SCAN
-            current_in_get_area = false;
-#endif
-            is->clear(is->rdstate() | std::ios::eofbit);
-        }
-#if !JSON_DISABLE_STREAM_BULK_SCAN
-        else if (JSON_HEDLEY_UNLIKELY(refill))
-        {
-            // the consumed character is at gptr() - 1, unless the streambuf
-            // has no get area and returned it from uflow()
+        // the character is in the get area: step over it, as sbumpc() does
         const char* const next = streambuf_access::get_ptr(sb);
-            after_refill(res, next != nullptr && next != streambuf_access::get_begin(sb));
-        }
-        else
+        if (JSON_HEDLEY_LIKELY(next != streambuf_access::get_end(sb)))
         {
             current_in_get_area = true;
+            streambuf_access::bump(sb, 1);
+            return std::char_traits<char>::to_int_type(*next);
+        }
+        if (no_get_area)
+        {
+            return get_character_no_get_area();
         }
 #endif
-        return res;
+        return get_character_slow();
     }
 #endif
 
@@ -334,6 +395,18 @@ class input_stream_adapter
     {
 #if JSON_PRECISE_STREAM_POSITION
         commit_lookahead();
+#endif
+#if !JSON_DISABLE_STREAM_BULK_SCAN
+        // the elements are in the get area: copy them, as sgetn() does, but
+        // without the virtual call
+        const char* const first = streambuf_access::get_ptr(sb);
+        const auto available = static_cast<std::size_t>(streambuf_access::get_end(sb) - first);
+        if (JSON_HEDLEY_LIKELY(count * sizeof(T) <= available && count * sizeof(T) <= static_cast<std::size_t>((std::numeric_limits<int>::max)())))
+        {
+            std::memcpy(dest, first, count * sizeof(T));
+            streambuf_access::bump(sb, static_cast<int>(count * sizeof(T)));
+            return count * sizeof(T);
+        }
 #endif
         auto res = static_cast<std::size_t>(sb->sgetn(reinterpret_cast<char*>(dest), static_cast<std::streamsize>(count * sizeof(T))));
         if (JSON_HEDLEY_UNLIKELY(res < count * sizeof(T)))
@@ -421,7 +494,7 @@ class input_stream_adapter
     // last one if @a drop_last.
     void copy_token(std::vector<char>& out, bool drop_last) const
     {
-        out = token_carry;
+        token_carry.copy_to(out);
         if (token_segment != nullptr)
         {
             out.insert(out.end(), token_segment, current_end());
@@ -434,6 +507,47 @@ class input_stream_adapter
 #endif
 
   private:
+#if !JSON_PRECISE_STREAM_POSITION
+    // the get area is exhausted, or there is none
+#if !JSON_DISABLE_STREAM_BULK_SCAN
+    JSON_HEDLEY_NEVER_INLINE
+#endif
+    std::char_traits<char>::int_type get_character_slow()
+    {
+#if !JSON_DISABLE_STREAM_BULK_SCAN
+        // sbumpc() refills the get area if it is exhausted
+        const bool refill = streambuf_access::get_ptr(sb) == streambuf_access::get_end(sb);
+        if (JSON_HEDLEY_UNLIKELY(refill))
+        {
+            save_token_segment();
+        }
+#endif
+
+        auto res = sb->sbumpc();
+        // set eof manually, as we don't use the istream interface.
+        if (JSON_HEDLEY_UNLIKELY(res == std::char_traits<char>::eof()))
+        {
+#if !JSON_DISABLE_STREAM_BULK_SCAN
+            current_in_get_area = false;
+#endif
+            is->clear(is->rdstate() | std::ios::eofbit);
+        }
+#if !JSON_DISABLE_STREAM_BULK_SCAN
+        else if (JSON_HEDLEY_UNLIKELY(refill))
+        {
+            // the consumed character is at gptr() - 1, unless the streambuf
+            // has no get area and returned it from uflow()
+            const char* const next = streambuf_access::get_ptr(sb);
+            after_refill(res, next != nullptr && next != streambuf_access::get_begin(sb));
+        }
+        else
+        {
+            current_in_get_area = true;
+        }
+#endif
+        return res;
+    }
+#endif
 #if !JSON_DISABLE_STREAM_BULK_SCAN
     // Reads the protected get-area pointers of any std::streambuf. Naming the
     // members through a derived class makes forming the member pointers
@@ -457,6 +571,11 @@ class input_stream_adapter
         {
             (s->*(&streambuf_access::gbump))(n);
         }
+        // sbumpc() for an exhausted get area
+        static std::streambuf::int_type next_character(std::streambuf* s)
+        {
+            return (s->*(&streambuf_access::uflow))();
+        }
     };
 
     // one past the character last returned by get_character(), if that is in
@@ -477,16 +596,53 @@ class input_stream_adapter
     {
         if (token_segment != nullptr)
         {
-            token_carry.insert(token_carry.end(), token_segment, streambuf_access::get_end(sb));
+            token_carry.append(token_segment, streambuf_access::get_end(sb));
             token_segment = nullptr;
         }
     }
 
+#if !JSON_PRECISE_STREAM_POSITION
+    // The streambuf has no get area (e.g., std::cin synchronized with stdio):
+    // every character comes from uflow(), and the token is kept by value.
+    JSON_HEDLEY_NEVER_INLINE std::char_traits<char>::int_type get_character_no_get_area()
+    {
+        // the get area is exhausted, so this is what sbumpc() does
+        const auto res = streambuf_access::next_character(sb);
+        if (JSON_HEDLEY_UNLIKELY(res == std::char_traits<char>::eof() || streambuf_access::get_end(sb) != nullptr))
+        {
+            return get_character_no_get_area_other(res);
+        }
+        // current_in_get_area is false: only the first character of a get
+        // area, or the end of the input, is read here
+        if (tracking_token)
+        {
+            token_carry.push_back(std::char_traits<char>::to_char_type(res));
+        }
+        return res;
+    }
+
+    // @a res came from uflow() of a streambuf without a get area: it is the
+    // end of the input, or the streambuf has a get area now
+    JSON_HEDLEY_NEVER_INLINE std::char_traits<char>::int_type get_character_no_get_area_other(std::char_traits<char>::int_type res)
+    {
+        if (res == std::char_traits<char>::eof())
+        {
+            current_in_get_area = false;
+            is->clear(is->rdstate() | std::ios::eofbit);
+            return res;
+        }
+        // the consumed character is at gptr() - 1
+        after_refill(res, streambuf_access::get_ptr(sb) != streambuf_access::get_begin(sb));
+        return res;
+    }
+#endif
+
     // character @a c was just read into a new get area (@a in_get_area) or
     // returned without one
     void after_refill(std::char_traits<char>::int_type c, bool in_get_area)
     {
         current_in_get_area = in_get_area;
+        no_get_area = !in_get_area;
         if (!tracking_token)
         {
             return;
@@ -525,13 +681,16 @@ class input_stream_adapter
 #endif
 #if !JSON_DISABLE_STREAM_BULK_SCAN
     /// the part of the current token that was in earlier get areas
-    std::vector<char> token_carry{}; // NOLINT(readability-redundant-member-init)
+    token_carry_buffer token_carry{}; // NOLINT(readability-redundant-member-init)
     /// start of the current token's part in the get area, or nullptr
     const char* token_segment = nullptr;
     /// whether the character last returned by get_character() is in the get area
     bool current_in_get_area = false;
     /// whether a token is tracked at all; the binary readers never start one
     bool tracking_token = false;
+    /// whether the streambuf had no get area when it was last asked for a
+    /// character
+    bool no_get_area = false;
 #endif
 };
 #endif  // JSON_NO_IO

include/nlohmann/detail/json_pointer.hpp

@@ -8,14 +8,14 @@
 
 #pragma once
 
-#include <algorithm> // all_of
+#include <algorithm> // all_of, copy
 #include <array> // array
 #include <cctype> // isdigit
 #include <cerrno> // errno, ERANGE
 #include <cstddef> // size_t
 #include <cstdint> // uint64_t
 #include <cstdlib> // strtoull
-#include <cstring> // memcmp, memcpy
+#include <cstring> // memchr, memcmp, memcpy
 #ifndef JSON_NO_IO
     #include <iosfwd> // ostream
 #endif  // JSON_NO_IO
@@ -80,14 +80,38 @@ class json_pointer
     /// @sa https://json.nlohmann.me/api/json_pointer/to_string/
     string_t to_string() const
     {
+        if (reference_tokens.empty())
+        {
+            return string_t();
+        }
+
         // one buffer: every token adds its escaped form (at least its length) and a slash
         std::size_t length = 0;
+        bool plain = true;
         for (const auto& reference_token : reference_tokens)
         {
             length += reference_token.size() + 1;
+            const char* const data = reference_token.data();
+            for (std::size_t i = 0, size = reference_token.size(); plain && i < size; ++i)
+            {
+                plain = data[i] != '~' && data[i] != '/';
+            }
         }
 
         string_t result;
+        if (JSON_HEDLEY_LIKELY(plain))
+        {
+            // nothing to escape: the tokens are copied into the string
+            result.resize(length);
+            char* out = &result[0];
+            for (const auto& reference_token : reference_tokens)
+            {
+                *out++ = '/';
+                out = std::copy(reference_token.data(), reference_token.data() + reference_token.size(), out);
+            }
+            return result;
+        }
+
         result.reserve(length);
         for (const auto& reference_token : reference_tokens)
         {
@@ -303,6 +327,12 @@ class json_pointer
         exceeds_size_type  ///< @a s converts to an integer that exceeds size_type
     };
 
+    /// whether a reference token is "-" (a plain comparison with a string literal is not always inlined)
+    static bool is_dash(const string_t& s) noexcept
+    {
+        return s.size() == 1 && s[0] == '-';
+    }
+
     /*!
     @param[in] s    reference token to be converted into an array index
     @param[out] idx the integer representation of @a s if @ref array_index_status::ok
@@ -328,7 +358,7 @@ class json_pointer
         // error condition (cf. RFC 6901, Sect. 4); this also covers single-
         // character tokens, so "/x" and "/xy" fail alike; "-" and the empty
         // token are left to the conversion below and are reported as unresolved
-        if (JSON_HEDLEY_UNLIKELY(!s.empty() && s != "-" && !(s[0] >= '0' && s[0] <= '9')))
+        if (JSON_HEDLEY_UNLIKELY(!s.empty() && !is_dash(s) && !(s[0] >= '0' && s[0] <= '9')))
         {
             return array_index_status::not_a_number;
         }
@@ -475,9 +505,40 @@ class json_pointer
         return pos == size;
     }
 
+    /// reserve room for @a n elements of an array type that has reserve()
+    template<typename ArrayType>
+    static auto reserve_elements(ArrayType& array, const std::size_t n, detail::priority_tag<1> /*unused*/)
+    -> decltype(array.reserve(n), void())
+    {
+        array.reserve(n);
+    }
+
+    template<typename ArrayType>
+    static void reserve_elements(ArrayType& /*array*/, const std::size_t /*n*/, detail::priority_tag<0> /*unused*/)
+    {}
+
     /// @brief object member access with a key that is only copied if it is inserted
+    ///
+    /// Objects that can find the position of a key and insert there
+    /// (std::map) search once, others look the key up and add it if it is missing.
     template<typename BasicJsonType>
-    static BasicJsonType& object_member(BasicJsonType& object, const string_t& key)
+    static auto object_member_impl(BasicJsonType& object, const string_t& key, detail::priority_tag<1> /*unused*/)
+    -> decltype(object.m_data.m_value.object->emplace_hint(object.m_data.m_value.object->lower_bound(key), key, nullptr), std::declval<BasicJsonType&>())
+    {
+        auto& members = *object.m_data.m_value.object;
+        auto it = members.lower_bound(key);
+        if (it != members.end() && !members.key_comp()(key, it->first))
+        {
+            return it->second;
+        }
+
+        const auto old_capacity = object.container_capacity();
+        it = members.emplace_hint(it, key, nullptr);
+        return object.set_parent_after_insert(it->second, old_capacity);
+    }
+
+    template<typename BasicJsonType>
+    static BasicJsonType& object_member_impl(BasicJsonType& object, const string_t& key, detail::priority_tag<0> /*unused*/)
     {
         auto& members = *object.m_data.m_value.object;
         const auto it = members.find(key);
@@ -488,6 +549,12 @@ class json_pointer
         return object.operator[](key);
     }
 
+    template<typename BasicJsonType>
+    static BasicJsonType& object_member(BasicJsonType& object, const string_t& key)
+    {
+        return object_member_impl(object, key, detail::priority_tag<1> {});
+    }
+
   JSON_PRIVATE_UNLESS_TESTED:
     json_pointer top() const
     {
@@ -720,48 +787,103 @@ class json_pointer
         }
     }
 
+    /*!
+    @brief the values along the pointer last resolved by @ref resolve_plain
+
+    A patch from @ref basic_json::diff lists its operations in document order,
+    so consecutive pointers share most of their reference tokens; the values
+    along the previous pointer are kept to resume from the common prefix.
+    `steps[k]` is the value after the first `k` reference tokens together with
+    the length of the prefix of `*path` that holds them. A value stays valid as
+    long as no operation changes its container: callers keep only values up to
+    the parent of the pointer they modify, or call `reset()`.
+    */
+    template<typename BasicJsonType>
+    struct plain_steps
+    {
+        const string_t* path = nullptr;
+        std::vector<std::pair<std::size_t, BasicJsonType*>> steps{};
+
+        void reset() noexcept
+        {
+            path = nullptr;
+            steps.clear();
+        }
+    };
+
     /*!
     @brief resolve a plain JSON pointer without building its reference tokens
 
-    @param[in] ptr     the value the pointer starts at
+    @param[in] root    the value the pointer starts at
     @param[in] path    a JSON pointer string
     @param[in] length  the number of leading bytes of @a path to resolve; 0 or
                        the position of a slash
     @param[in,out] token  a string to hold the reference tokens, to reuse its storage
+    @param[in,out] cache  the values along the previously resolved pointer;
+                          @a path must outlive the next call
 
     @return the pointed to value, or `nullptr` if the pointer does not denote
-            an existing value, or does so in a way that is not plain: a
-            reference token with an escape, an array index that is malformed or
-            out of range, or a "-". No exception is thrown for these; callers
-            use @ref json_pointer instead, which handles them.
+            an existing value, or does so in a way that is not plain: an array
+            index that is malformed or out of range, or a "-", or a reference
+            token with an escape. No exception is
+            thrown for these; callers use @ref json_pointer instead, which
+            handles them.
     */
     template<typename BasicJsonType>
-    static BasicJsonType* resolve_plain(BasicJsonType* ptr, const string_t& path, const std::size_t length, string_t& token)
+    static BasicJsonType* resolve_plain(BasicJsonType* root, const string_t& path, const std::size_t length, string_t& token, plain_steps<BasicJsonType>& cache)
     {
-        for (std::size_t pos = 0; pos < length;)
+        auto& steps = cache.steps;
+
+        // resume from the deepest step that is also a prefix of this path
+        std::size_t depth = 1;
+        if (cache.path != nullptr && !steps.empty() && steps[0].second == root)
+        {
+            const string_t& previous = *cache.path;
+            for (depth = steps.size(); depth > 1; --depth)
+            {
+                // a step ends at a token boundary of the previous path
+                const std::size_t prefix = steps[depth - 1].first;
+                if (prefix <= length && (prefix == length || path[prefix] == '/') && std::memcmp(path.data(), previous.data(), prefix) == 0)
+                {
+                    break;
+                }
+            }
+            steps.resize(depth);
+        }
+        else
+        {
+            steps.assign(1, std::make_pair(std::size_t{0}, root));
+        }
+        cache.path = &path;
+
+        BasicJsonType* ptr = steps[depth - 1].second;
+        for (std::size_t pos = steps[depth - 1].first; pos < length;)
         {
             if (JSON_HEDLEY_UNLIKELY(path[pos] != '/'))
             {
                 return nullptr;
             }
 
-            // the reference token between the slash at pos and the next one
-            std::size_t end = path.find_first_of('/', pos + 1);
-            if (end == string_t::npos || end > length)
+            // the reference token between the slash at pos and the next one,
+            // which must not have an escape
+            const char* const begin = path.data() + pos + 1;
+            const char* const limit = path.data() + length;
+            const char* end = begin;
+            while (end < limit && *end != '/')
             {
-                end = length;
-            }
-            token.clear();
-            token.append(path.data() + pos + 1, end - pos - 1);
-            if (JSON_HEDLEY_UNLIKELY(token.find_first_of('~') != string_t::npos))
+                if (JSON_HEDLEY_UNLIKELY(*end == '~'))
                 {
                     return nullptr;
                 }
+                ++end;
+            }
 
             switch (ptr->type())
             {
                 case detail::value_t::object:
                 {
+                    token.clear();
+                    token.append(begin, static_cast<std::size_t>(end - begin));
                     const auto it = ptr->m_data.m_value.object->find(token);
                     if (JSON_HEDLEY_UNLIKELY(it == ptr->m_data.m_value.object->end()))
                     {
@@ -773,8 +895,24 @@ class json_pointer
 
                 case detail::value_t::array:
                 {
-                    typename BasicJsonType::size_type idx{};
-                    if (JSON_HEDLEY_UNLIKELY(parse_array_index<BasicJsonType>(token, idx) != array_index_status::ok || idx >= ptr->m_data.m_value.array->size()))
+                    // an index of up to 18 digits without a leading zero fits into size_type
+                    // (9 digits for a 32-bit size_type); everything else is left to json_pointer
+                    constexpr std::size_t max_digits = sizeof(typename BasicJsonType::size_type) >= 8 ? 18 : 9;
+                    const auto digits = static_cast<std::size_t>(end - begin);
+                    if (JSON_HEDLEY_UNLIKELY(digits == 0 || digits > max_digits || (digits > 1 && *begin == '0')))
+                    {
+                        return nullptr;
+                    }
+                    typename BasicJsonType::size_type idx = 0;
+                    for (const char* c = begin; c != end; ++c)
+                    {
+                        if (JSON_HEDLEY_UNLIKELY(*c < '0' || *c > '9'))
+                        {
+                            return nullptr;
+                        }
+                        idx = static_cast<typename BasicJsonType::size_type>((idx * 10) + static_cast<typename BasicJsonType::size_type>(*c - '0'));
+                    }
+                    if (JSON_HEDLEY_UNLIKELY(idx >= ptr->m_data.m_value.array->size()))
                     {
                         return nullptr;
                     }
@@ -793,7 +931,8 @@ class json_pointer
                 default:
                     return nullptr;
             }
-            pos = end;
+            pos = static_cast<std::size_t>(end - path.data());
+            steps.emplace_back(pos, ptr);
         }
 
         return ptr;
@@ -907,7 +1046,7 @@ class json_pointer
 
                 case detail::value_t::array:
                 {
-                    if (JSON_HEDLEY_UNLIKELY(reference_token == "-"))
+                    if (JSON_HEDLEY_UNLIKELY(is_dash(reference_token)))
                     {
                         // "-" always fails the range check
                         JSON_THROW(detail::out_of_range::create(detail::exception_id::pointer_past_the_end_index, detail::concat(
@@ -978,7 +1117,7 @@ class json_pointer
 
                 case detail::value_t::array:
                 {
-                    if (JSON_HEDLEY_UNLIKELY(reference_token == "-"))
+                    if (JSON_HEDLEY_UNLIKELY(is_dash(reference_token)))
                     {
                         // "-" cannot be used for const access
                         JSON_THROW(detail::out_of_range::create(detail::exception_id::pointer_past_the_end_index, detail::concat("array index '-' (", std::to_string(ptr->m_data.m_value.array->size()), ") is out of range"), ptr));
@@ -1040,7 +1179,7 @@ class json_pointer
 
                 case detail::value_t::array:
                 {
-                    if (JSON_HEDLEY_UNLIKELY(reference_token == "-"))
+                    if (JSON_HEDLEY_UNLIKELY(is_dash(reference_token)))
                     {
                         // "-" always fails the range check
                         return nullptr;
@@ -1117,7 +1256,7 @@ class json_pointer
 
                 case detail::value_t::array:
                 {
-                    if (JSON_HEDLEY_UNLIKELY(reference_token == "-"))
+                    if (JSON_HEDLEY_UNLIKELY(is_dash(reference_token)))
                     {
                         // "-" always fails the range check
                         return false;
@@ -1198,33 +1337,32 @@ class json_pointer
         }
         result.reserve(token_count);
 
-        // extract the reference tokens:
-        // - slash: position of the last read slash (or end of string)
-        // - start: position after the previous slash
-        for (
-            // search for the first slash after the first character
-            std::size_t slash = reference_string.find_first_of('/', 1),
-            // set the beginning of the first reference token
-            start = 1;
-            // we can stop if start == 0 (if slash == string_t::npos)
-            start != 0;
-            // set the beginning of the next reference token
-            // (will eventually be 0 if slash == string_t::npos)
-            start = (slash == string_t::npos) ? 0 : slash + 1,
-            // find next slash
-            slash = reference_string.find_first_of('/', start))
+        // extract the reference tokens: every one runs from after a slash to
+        // the next slash or the end of the string; a token without a '~' (the
+        // common case) is stored as it is, others are checked and unescaped
+        const char* const data = reference_string.data();
+        const char* const end = data + size;
+        for (const char* start = data + 1;; ++start)
         {
-            // use the text between the beginning of the reference token
-            // (start) and the last slash (slash).
-            const auto count = (slash == string_t::npos ? reference_string.size() : slash) - start;
-            auto reference_token = string_t(reference_string.data() + start, count);
+            const char* slash = start;
+            bool escaped = false;
+            while (slash != end && *slash != '/')
+            {
+                escaped = escaped || *slash == '~';
+                ++slash;
+            }
 
-            // check reference tokens are properly escaped; a token without
-            // a '~' (the common case) needs neither check nor unescaping
-            std::size_t pos = reference_token.find_first_of('~');
-            if (pos != string_t::npos)
+            const auto count = static_cast<std::size_t>(slash - start);
+            if (JSON_HEDLEY_LIKELY(!escaped))
             {
-                for (; pos != string_t::npos; pos = reference_token.find_first_of('~', pos + 1))
+                result.emplace_back(start, count);
+            }
+            else
+            {
+                auto reference_token = string_t(start, count);
+
+                // check reference tokens are properly escaped
+                for (std::size_t pos = reference_token.find_first_of('~'); pos != string_t::npos; pos = reference_token.find_first_of('~', pos + 1))
                 {
                     JSON_ASSERT(reference_token[pos] == '~');
 
@@ -1238,10 +1376,14 @@ class json_pointer
                 }
 
                 detail::unescape(reference_token);
+                result.push_back(std::move(reference_token));
             }
 
-            // finally, store the reference token
-            result.push_back(std::move(reference_token));
+            if (slash == end)
+            {
+                break;
+            }
+            start = slash;
         }
 
         return result;
@@ -1276,6 +1418,52 @@ class json_pointer
         obj[key] = std::forward<ValueType>(v);
     }
 
+    /// the number of entries that flattening @a value adds: its primitive
+    /// values and empty containers
+    template<typename BasicJsonType>
+    static std::size_t count_flattened(const BasicJsonType& value)
+    {
+        std::size_t count = 0;
+        std::vector<const BasicJsonType*> pending(1, &value);
+        while (!pending.empty())
+        {
+            const BasicJsonType* const v = pending.back();
+            pending.pop_back();
+            if (v->is_array() && !v->m_data.m_value.array->empty())
+            {
+                for (const auto& element : *v->m_data.m_value.array)
+                {
+                    pending.push_back(&element);
+                }
+            }
+            else if (v->is_object() && !v->m_data.m_value.object->empty())
+            {
+                for (const auto& member : *v->m_data.m_value.object)
+                {
+                    pending.push_back(&member.second);
+                }
+            }
+            else
+            {
+                ++count;
+            }
+        }
+        return count;
+    }
+
+    // Reserve the storage for @a value of objects that have a reserve(), so
+    // that it does not have to be moved while it grows
+    template<typename ObjectType, typename BasicJsonType>
+    static auto reserve_flattened(ObjectType& obj, const BasicJsonType& value, detail::priority_tag<1> /*unused*/)
+    -> decltype(obj.reserve(std::size_t {}), void())
+    {
+        obj.reserve(count_flattened(value));
+    }
+
+    template<typename ObjectType, typename BasicJsonType>
+    static void reserve_flattened(ObjectType& /*obj*/, const BasicJsonType& /*value*/, detail::priority_tag<0> /*unused*/)
+    {}
+
     /// append the decimal representation of @a i to @a out
     static void append_index(string_t& out, std::size_t i)
     {
@@ -1385,6 +1573,7 @@ class json_pointer
             }
         };
 
+        reserve_flattened(object, value, detail::priority_tag<1> {});
         enter(value);
         while (!stack.empty())
         {
@@ -1392,8 +1581,9 @@ class json_pointer
             // frame, which would invalidate a reference to it
             const BasicJsonType* const container = stack.back().container;
 
-            // drop the path of the previous child
-            path.resize(stack.back().path_length);
+            // the path of the previous child is replaced by the path of the
+            // next one; for an array element in one resize
+            const std::size_t path_length = stack.back().path_length;
 
             if (container->is_array())
             {
@@ -1407,8 +1597,20 @@ class json_pointer
 
                 // iterate array and use index as a reference string
                 ++stack.back().index;
-                path.push_back('/');
-                append_index(path, i);
+                std::array<char, 24> digits{};
+                std::size_t pos = digits.size();
+                std::size_t rest = i;
+                do
+                {
+                    digits[--pos] = static_cast<char>('0' + (rest % 10));
+                    rest /= 10;
+                }
+                while (rest != 0);
+                const std::size_t count = digits.size() - pos;
+                path.resize(path_length + 1 + count);
+                char* const out = &path[path_length];
+                out[0] = '/';
+                std::memcpy(out + 1, digits.data() + pos, count);
                 enter(array[i]);
             }
             else
@@ -1422,6 +1624,7 @@ class json_pointer
 
                 // iterate object and use keys as reference string
                 ++stack.back().member;
+                path.resize(path_length);
                 path.push_back('/');
                 detail::escape_into(path, it->first);
                 enter(it->second);
@@ -1517,8 +1720,28 @@ class json_pointer
                 {
                     end = key.size();
                 }
+                const std::size_t token_size = end - pos - 1;
+
+                // a token that is a plain index of up to 9 digits, which is
+                // the case for almost all array elements, is not copied
+                std::size_t index = 0;
+                bool plain_index = current->type() == detail::value_t::array && token_size != 0 && token_size <= 9 && (token_size == 1 || data[pos + 1] != '0');
+                for (std::size_t k = pos + 1; plain_index && k < end; ++k)
+                {
+                    plain_index = data[k] >= '0' && data[k] <= '9';
+                    index = (index * 10) + static_cast<std::size_t>(data[k] - '0');
+                }
+                if (plain_index)
+                {
+                    current = &current->operator[](static_cast<typename BasicJsonType::size_type>(index));
+                    values.push_back(current);
+                    ends.push_back(end);
+                    pos = end;
+                    continue;
+                }
+
                 token.clear();
-                token.append(data + pos + 1, end - pos - 1);
+                token.append(data + pos + 1, token_size);
                 if (token.find_first_of('~') != string_t::npos)
                 {
                     detail::unescape(token);
@@ -1528,10 +1751,13 @@ class json_pointer
                 {
                     case detail::value_t::null:
                     {
-                        if (arrays.contains(data, pos))
+                        // a first child 0 makes the value an array without a look-up
+                        if ((token_size == 1 && data[pos + 1] == '0') || arrays.contains(data, pos))
                         {
                             // some reference token below this position is 0, so the
                             // value is an array
+                            *current = BasicJsonType(detail::value_t::array);
+                            reserve_elements(*current->m_data.m_value.array, 2, detail::priority_tag<1> {});
                             current = &current->operator[](array_index<BasicJsonType>(token));
                         }
                         else

include/nlohmann/detail/output/output_adapters.hpp

@@ -55,6 +55,66 @@ using output_adapter_t = std::shared_ptr<output_adapter_protocol<CharType>>;
 /// output_vector_sink also appends writes of at least this size with insert()
 constexpr std::size_t max_reserve_size = 256;
 
+/// @brief copy @a length characters, at most 32, without a call of memcpy()
+///
+/// Short strings are the typical payload of a binary value, and a call of
+/// memcpy() costs more than copying them. A copy of 16 to 32 characters is two
+/// possibly overlapping 16-byte copies, one of 8 to 15 two 8-byte copies, and
+/// so on down to a single character; each is a fixed-size load and store.
+/// Nothing outside of [src, src + length) is read or written.
+template<typename CharType>
+inline void copy_short(CharType* dst, const CharType* src, const std::size_t length, std::true_type /*is_byte*/) noexcept
+{
+    JSON_ASSERT(length <= 32);
+    if (length >= 16)
+    {
+        std::array<char, 16> first; // NOLINT(cppcoreguidelines-pro-type-member-init,hicpp-member-init): written before read
+        std::array<char, 16> last; // NOLINT(cppcoreguidelines-pro-type-member-init,hicpp-member-init): written before read
+        std::memcpy(first.data(), src, 16);
+        std::memcpy(last.data(), src + (length - 16), 16);
+        std::memcpy(dst, first.data(), 16);
+        std::memcpy(dst + (length - 16), last.data(), 16);
+    }
+    else if (length >= 8)
+    {
+        std::array<char, 8> first; // NOLINT(cppcoreguidelines-pro-type-member-init,hicpp-member-init): written before read
+        std::array<char, 8> last; // NOLINT(cppcoreguidelines-pro-type-member-init,hicpp-member-init): written before read
+        std::memcpy(first.data(), src, 8);
+        std::memcpy(last.data(), src + (length - 8), 8);
+        std::memcpy(dst, first.data(), 8);
+        std::memcpy(dst + (length - 8), last.data(), 8);
+    }
+    else if (length >= 4)
+    {
+        std::array<char, 4> first; // NOLINT(cppcoreguidelines-pro-type-member-init,hicpp-member-init): written before read
+        std::array<char, 4> last; // NOLINT(cppcoreguidelines-pro-type-member-init,hicpp-member-init): written before read
+        std::memcpy(first.data(), src, 4);
+        std::memcpy(last.data(), src + (length - 4), 4);
+        std::memcpy(dst, first.data(), 4);
+        std::memcpy(dst + (length - 4), last.data(), 4);
+    }
+    else if (length != 0)
+    {
+        // 1 to 3: the first, the middle, and the last character
+        const CharType a = src[0];
+        const CharType b = src[length >> 1];
+        const CharType c = src[length - 1];
+        dst[0] = a;
+        dst[length >> 1] = b;
+        dst[length - 1] = c;
+    }
+}
+
+/// @brief copy @a length characters, of a character type wider than a byte
+template<typename CharType>
+inline void copy_short(CharType* dst, const CharType* src, const std::size_t length, std::false_type /*is_byte*/) noexcept
+{
+    std::memcpy(dst, src, length * sizeof(CharType));
+}
+
+/// the number of characters an empty vector is grown to by output_vector_sink
+constexpr std::size_t min_vector_sink_capacity = 64;
+
 /// @brief non-virtual output sink writing into a std::vector
 ///
 /// This sink is not part of the virtual output_adapter_protocol hierarchy: it is
@@ -78,7 +138,7 @@ class output_vector_sink
 {
   public:
     explicit output_vector_sink(std::vector<CharType, AllocatorType>& vec) noexcept
-        : v(&vec), position(vec.size())
+        : v(&vec), cursor(vec.data() + vec.size()), limit(cursor)
     {}
 
     output_vector_sink(const output_vector_sink&) = delete;
@@ -86,7 +146,7 @@ class output_vector_sink
     output_vector_sink& operator=(output_vector_sink&&) = delete;
 
     output_vector_sink(output_vector_sink&& other) noexcept
-        : v(other.v), position(other.position)
+        : v(other.v), cursor(other.cursor), limit(other.limit)
     {
         other.v = nullptr;
     }
@@ -95,7 +155,7 @@ class output_vector_sink
     {
         if (v != nullptr)
         {
-            v->resize(position);
+            v->resize(position());
         }
     }
 
@@ -104,11 +164,11 @@ class output_vector_sink
     JSON_HEDLEY_ALWAYS_INLINE
     void write_character(CharType c)
     {
-        if (JSON_HEDLEY_UNLIKELY(position == v->size()))
+        if (JSON_HEDLEY_UNLIKELY(cursor == limit))
         {
             grow(1);
         }
-        (*v)[position++] = c;
+        *cursor++ = c;
     }
 
     // no JSON_HEDLEY_NON_NULL here: binary_writer legitimately passes a null
@@ -117,7 +177,7 @@ class output_vector_sink
     JSON_HEDLEY_ALWAYS_INLINE
     void write_characters(const CharType* s, std::size_t length)
     {
-        if (JSON_HEDLEY_UNLIKELY(v->size() - position < length))
+        if (JSON_HEDLEY_UNLIKELY(space() < length))
         {
             if (length >= max_reserve_size)
             {
@@ -126,10 +186,15 @@ class output_vector_sink
             }
             grow(length);
         }
-        if (length != 0)
+        if (length <= 32)
         {
-            std::memcpy(v->data() + position, s, length * sizeof(CharType));
-            position += length;
+            copy_short(cursor, s, length, std::integral_constant<bool, sizeof(CharType) == 1>());
+            cursor += length;
+        }
+        else
+        {
+            std::memcpy(cursor, s, length * sizeof(CharType));
+            cursor += length;
         }
     }
 
@@ -144,21 +209,37 @@ class output_vector_sink
     CharType* reserve(const std::size_t length)
     {
         JSON_ASSERT(length <= max_reserve_size);
-        if (JSON_HEDLEY_UNLIKELY(v->size() - position < length))
+        if (JSON_HEDLEY_UNLIKELY(space() < length))
         {
             grow(length);
         }
-        return v->data() + position;
+        return cursor;
     }
 
     /// keep the first @a length characters written to the space from reserve()
     JSON_HEDLEY_ALWAYS_INLINE
     void commit(const std::size_t length) noexcept
     {
-        position += length;
+        cursor += length;
     }
 
+    /// nothing to pass on: the characters are in the vector (see output_adapter_sink)
+    void flush() noexcept
+    {}
+
   private:
+    /// number of characters of *v written so far
+    std::size_t position() const noexcept
+    {
+        return static_cast<std::size_t>(cursor - v->data());
+    }
+
+    /// number of characters that fit the vector behind the write position
+    std::size_t space() const noexcept
+    {
+        return static_cast<std::size_t>(limit - cursor);
+    }
+
     /// make room for at least @a length more characters after the position;
     /// kept out of line so that the write functions stay small enough for
     /// GCC to inline them into the writer
@@ -166,12 +247,15 @@ class output_vector_sink
     void grow(const std::size_t length)
     {
         // use the capacity the vector already has, else at least double it;
-        // never less than 64 characters, so that a small value does not
+        // never less than min_vector_sink_capacity, so that a small value does not
         // reallocate for every few bytes
-        const std::size_t needed = position + length;
-        v->resize((std::max)(std::size_t{64}, needed <= v->capacity()
+        const std::size_t written = position();
+        const std::size_t needed = written + length;
+        v->resize((std::max)(min_vector_sink_capacity, needed <= v->capacity()
                              ? v->capacity()
                              : (std::max)(needed, v->capacity() * 2)));
+        cursor = v->data() + written;
+        limit = v->data() + v->size();
     }
 
     /// append @a length (at least max_reserve_size) characters that do not fit
@@ -181,20 +265,24 @@ class output_vector_sink
     JSON_HEDLEY_NEVER_INLINE
     void append(const CharType* s, const std::size_t length)
     {
-        v->resize(position);
-        if (v->capacity() - position < length)
+        const std::size_t written = position();
+        v->resize(written);
+        if (v->capacity() - written < length)
         {
             // grow geometrically, with room for the writes that follow, so
             // that the next grow() need not reallocate right away
-            v->reserve((std::max)(v->capacity() * 2, position + length + max_reserve_size));
+            v->reserve((std::max)(v->capacity() * 2, written + length + max_reserve_size));
         }
         v->insert(v->end(), s, s + length);
-        position = v->size();
+        cursor = v->data() + v->size();
+        limit = cursor;
     }
 
     std::vector<CharType, AllocatorType>* v;
-    /// number of characters of *v written so far
-    std::size_t position;
+    /// the write position: in *v, at the number of characters written so far
+    CharType* cursor;
+    /// the end of *v
+    CharType* limit;
 };
 
 /// output adapter for byte vectors
@@ -277,9 +365,17 @@ class output_string_adapter : public output_adapter_protocol<CharType>
 ///
 /// Wraps the polymorphic output_adapter_t so the same binary_writer template can
 /// also target arbitrary adapters (output streams, strings, user-provided
-/// adapters) via the `output_adapter`-based overloads. Each write still goes
-/// through one virtual call, exactly as before; only the concrete sinks above
-/// avoid it.
+/// adapters) via the `output_adapter`-based overloads.
+///
+/// A write through the adapter is a virtual call, and for a stream a call with
+/// a sentry on top, which costs more than encoding the few bytes of most values.
+/// The sink therefore collects what is written in a buffer and passes it on in
+/// one piece when it is full, a write is too large for it, or @ref flush is
+/// called. Everything written reaches the adapter in the same order and with
+/// the same bytes; the destructor passes on what is left, so that a write that
+/// throws leaves the adapter with exactly the bytes written before it. Only
+/// @ref flush can report an error of the adapter (the destructor must not
+/// throw), so it has to be called once the value is written.
 template<typename CharType>
 class output_adapter_sink
 {
@@ -290,36 +386,121 @@ class output_adapter_sink
         JSON_ASSERT(oa);
     }
 
+    output_adapter_sink(const output_adapter_sink&) = delete;
+    output_adapter_sink& operator=(const output_adapter_sink&) = delete;
+    output_adapter_sink& operator=(output_adapter_sink&&) = delete;
+
+    output_adapter_sink(output_adapter_sink&& other) noexcept
+        : oa(std::move(other.oa)), used(other.used)
+    {
+        std::memcpy(buffer.data(), other.buffer.data(), other.used * sizeof(CharType));
+        other.used = 0;
+    }
+
+    ~output_adapter_sink()
+    {
+        // a normal return has flushed already, so this is for a write that threw
+        if (used != 0)
+        {
+            JSON_TRY
+            {
+                flush();
+            }
+            JSON_CATCH(...)
+            {
+                // nothing can be done about an adapter that throws here
+            }
+        }
+    }
+
+    JSON_HEDLEY_ALWAYS_INLINE
     void write_character(CharType c)
     {
-        oa->write_character(c);
+        if (JSON_HEDLEY_UNLIKELY(used == buffer_size))
+        {
+            flush();
+        }
+        buffer[used++] = c;
     }
 
     // no JSON_HEDLEY_NON_NULL: forwards (null, 0) for empty payloads, exactly as
     // the type-erased path already did before this sink existed
+    JSON_HEDLEY_ALWAYS_INLINE
     void write_characters(const CharType* s, std::size_t length)
     {
-        oa->write_characters(s, length);
+        if (JSON_HEDLEY_UNLIKELY(buffer_size - used < length))
+        {
+            write_large(s, length);
+            return;
+        }
+        if (length <= 32)
+        {
+            copy_short(buffer.data() + used, s, length, std::integral_constant<bool, sizeof(CharType) == 1>());
+            used += length;
+        }
+        else
+        {
+            std::memcpy(buffer.data() + used, s, length * sizeof(CharType));
+            used += length;
+        }
     }
 
     /// space for writing up to @a length characters (see
     /// output_vector_sink::reserve)
-    CharType* reserve(const std::size_t length) noexcept
+    JSON_HEDLEY_ALWAYS_INLINE
+    CharType* reserve(const std::size_t length)
     {
         JSON_ASSERT(length <= max_reserve_size);
-        static_cast<void>(length);
-        return reserved.data();
+        if (JSON_HEDLEY_UNLIKELY(buffer_size - used < length))
+        {
+            flush();
+        }
+        return buffer.data() + used;
     }
 
-    /// write the first @a length characters written to the space from reserve()
-    void commit(const std::size_t length)
+    /// keep the first @a length characters written to the space from reserve()
+    JSON_HEDLEY_ALWAYS_INLINE
+    void commit(const std::size_t length) noexcept
     {
-        oa->write_characters(reserved.data(), length);
+        used += length;
+    }
+
+    /// pass everything written so far on to the adapter
+    void flush()
+    {
+        if (used != 0)
+        {
+            // reset first: the characters count as written even if the adapter throws
+            const std::size_t length = used;
+            used = 0;
+            oa->write_characters(buffer.data(), length);
+        }
     }
 
   private:
+    /// a write that does not fit the space left in the buffer
+    JSON_HEDLEY_NEVER_INLINE
+    void write_large(const CharType* s, const std::size_t length)
+    {
+        flush();
+        if (length >= buffer_size)
+        {
+            oa->write_characters(s, length);
+        }
+        else
+        {
+            std::memcpy(buffer.data(), s, length * sizeof(CharType));
+            used = length;
+        }
+    }
+
+    /// large enough for reserve() at any time
+    static constexpr std::size_t buffer_size = 4 * max_reserve_size;
+
     output_adapter_t<CharType> oa;
-    std::array<CharType, max_reserve_size> reserved{};
+    /// number of characters of @ref buffer not yet passed on
+    std::size_t used = 0;
+    std::array<CharType, buffer_size> buffer{};
 };
 
 /// @brief whether std::basic_string<CharType> has a non-deprecated std::char_traits

include/nlohmann/detail/output/serializer.hpp

@@ -71,7 +71,7 @@ class serializer
 
     None of @a pretty_print_, @a ensure_ascii_ and @a indent_step_ change over
     the life of the serializer, so they are captured once here instead of
-    being threaded through every call to @ref dump, @ref dump_internal and
+    being threaded through every call to @ref dump, @ref dump_container and
     @ref dump_iteratively.
     */
     serializer(output_adapter_protocol<char>& s, const char ichar,
@@ -80,11 +80,14 @@ class serializer
                const std::size_t indent_step_ = 0,
                error_handler_t error_handler_ = error_handler_t::strict)
         : o(&s)
+        , number_buffer() // NOLINT(readability-redundant-member-init)
+        , string_buffer() // NOLINT(readability-redundant-member-init)
         , indent_char(ichar)
         , pretty_print(pretty_print_)
         , ensure_ascii(ensure_ascii_)
         , indent_step(indent_step_)
         , error_handler(error_handler_)
+        , write_buffer() // NOLINT(readability-redundant-member-init)
     {}
 
     // deleted because of pointer members
@@ -115,18 +118,18 @@ class serializer
     void dump(const BasicJsonType& val,
               const std::size_t current_indent = 0)
     {
-        dump_internal(val, current_indent);
+        dump_element(val, current_indent, 0);
         flush();
     }
 
   JSON_PRIVATE_UNLESS_TESTED:
     /*!
-    @brief worker for @ref dump
+    @brief serialize the object or array @a val
 
-    Identical in behavior to the historical @ref dump, but writes into the
-    serializer's internal @ref write_buffer instead of issuing a virtual call
-    per token. The public @ref dump wraps this and flushes the buffer once the
-    top-level value has been serialized.
+    Writes into the serializer's internal @ref write_buffer instead of issuing
+    a virtual call per token; the public @ref dump flushes the buffer once the
+    top-level value has been serialized. Scalars are written by @ref dump_scalar,
+    which @ref dump_element calls without entering this function.
 
     Serializing a container descends into its elements, so a value nested deeply
     enough used to exhaust the call stack and terminate the process with no
@@ -137,20 +140,20 @@ class serializer
 
     @sa https://github.com/nlohmann/json/issues/5387
     */
-    void dump_internal(const BasicJsonType& val,
-                       const std::size_t current_indent = 0,
-                       const std::size_t depth = 0)
-    {
-        switch (val.m_data.m_type)
-        {
-            case value_t::object:
+    void dump_container(const BasicJsonType& val,
+                        const std::size_t current_indent,
+                        const std::size_t depth)
     {
+        JSON_ASSERT(val.m_data.m_type == value_t::object || val.m_data.m_type == value_t::array);
+
         if (JSON_HEDLEY_UNLIKELY(depth >= recursion_depth_limit()))
         {
             dump_iteratively(val, current_indent);
             return;
         }
 
+        if (val.m_data.m_type == value_t::object)
+        {
             if (val.m_data.m_value.object->empty())
             {
                 put_literal("{}");
@@ -159,34 +162,27 @@ class serializer
 
             if (pretty_print)
             {
-                    put_literal("{\n");
-
                 // variable to hold indentation for recursive calls
                 const auto new_indent = next_indent(current_indent, indent_step);
 
+                put_literal_indent("{\n", new_indent);
+
                 // first n-1 elements
                 auto i = val.m_data.m_value.object->cbegin();
                 for (std::size_t cnt = 0; cnt < val.m_data.m_value.object->size() - 1; ++cnt, ++i)
                 {
-                        put_indent(new_indent);
-                        put_char('"');
-                        dump_escaped(i->first);
-                        put_literal("\": ");
-                        dump_internal(i->second, new_indent, depth + 1);
-                        put_literal(",\n");
+                    put_object_key(i->first, "\": ");
+                    dump_element(i->second, new_indent, depth + 1);
+                    put_literal_indent(",\n", new_indent);
                 }
 
                 // last element
                 JSON_ASSERT(i != val.m_data.m_value.object->cend());
                 JSON_ASSERT(std::next(i) == val.m_data.m_value.object->cend());
-                    put_indent(new_indent);
-                    put_char('"');
-                    dump_escaped(i->first);
-                    put_literal("\": ");
-                    dump_internal(i->second, new_indent, depth + 1);
+                put_object_key(i->first, "\": ");
+                dump_element(i->second, new_indent, depth + 1);
 
-                    put_char('\n');
-                    put_indent(current_indent);
+                put_literal_indent("\n", current_indent);
                 put_char('}');
             }
             else
@@ -197,32 +193,19 @@ class serializer
                 auto i = val.m_data.m_value.object->cbegin();
                 for (std::size_t cnt = 0; cnt < val.m_data.m_value.object->size() - 1; ++cnt, ++i)
                 {
-                        put_char('"');
-                        dump_escaped(i->first);
-                        put_literal("\":");
-                        dump_internal(i->second, current_indent, depth + 1);
+                    put_object_key(i->first, "\":");
+                    dump_element(i->second, 0, depth + 1);
                     put_char(',');
                 }
 
                 // last element
                 JSON_ASSERT(i != val.m_data.m_value.object->cend());
                 JSON_ASSERT(std::next(i) == val.m_data.m_value.object->cend());
-                    put_char('"');
-                    dump_escaped(i->first);
-                    put_literal("\":");
-                    dump_internal(i->second, current_indent, depth + 1);
+                put_object_key(i->first, "\":");
+                dump_element(i->second, 0, depth + 1);
 
                 put_char('}');
             }
-
-                return;
-            }
-
-            case value_t::array:
-            {
-                if (JSON_HEDLEY_UNLIKELY(depth >= recursion_depth_limit()))
-                {
-                    dump_iteratively(val, current_indent);
             return;
         }
 
@@ -234,27 +217,24 @@ class serializer
 
         if (pretty_print)
         {
-                    put_literal("[\n");
-
             // variable to hold indentation for recursive calls
             const auto new_indent = next_indent(current_indent, indent_step);
 
+            put_literal_indent("[\n", new_indent);
+
             // first n-1 elements
             for (auto i = val.m_data.m_value.array->cbegin();
                     i != val.m_data.m_value.array->cend() - 1; ++i)
             {
-                        put_indent(new_indent);
-                        dump_internal(*i, new_indent, depth + 1);
-                        put_literal(",\n");
+                dump_element(*i, new_indent, depth + 1);
+                put_literal_indent(",\n", new_indent);
             }
 
             // last element
             JSON_ASSERT(!val.m_data.m_value.array->empty());
-                    put_indent(new_indent);
-                    dump_internal(val.m_data.m_value.array->back(), new_indent, depth + 1);
+            dump_element(val.m_data.m_value.array->back(), new_indent, depth + 1);
 
-                    put_char('\n');
-                    put_indent(current_indent);
+            put_literal_indent("\n", current_indent);
             put_char(']');
         }
         else
@@ -265,39 +245,44 @@ class serializer
             for (auto i = val.m_data.m_value.array->cbegin();
                     i != val.m_data.m_value.array->cend() - 1; ++i)
             {
-                        dump_internal(*i, current_indent, depth + 1);
+                dump_element(*i, 0, depth + 1);
                 put_char(',');
             }
 
             // last element
             JSON_ASSERT(!val.m_data.m_value.array->empty());
-                    dump_internal(val.m_data.m_value.array->back(), current_indent, depth + 1);
+            dump_element(val.m_data.m_value.array->back(), 0, depth + 1);
 
             put_char(']');
         }
-
-                return;
-            }
-
-            case value_t::string:
-            case value_t::binary:
-            case value_t::boolean:
-            case value_t::number_integer:
-            case value_t::number_unsigned:
-            case value_t::number_float:
-            case value_t::discarded:
-            case value_t::null:
-            default:
-                dump_scalar(val, current_indent);
-                return;
-        }
     }
 
   private:
+    /*!
+    @brief serialize an element of a container
+
+    Most elements are scalars, which need none of the state dump_container() keeps
+    for a container; writing them from here spares each the entry and exit of
+    that function.
+    */
+    JSON_HEDLEY_ALWAYS_INLINE void dump_element(const BasicJsonType& val,
+            const std::size_t current_indent,
+            const std::size_t depth)
+    {
+        if (val.m_data.m_type == value_t::object || val.m_data.m_type == value_t::array)
+        {
+            dump_container(val, current_indent, depth);
+        }
+        else
+        {
+            dump_scalar(val, current_indent);
+        }
+    }
+
     /*!
     @brief write out @a val and everything below it without the call stack
 
-    Emits the same bytes as @ref dump_internal, keeping the containers it has
+    Emits the same bytes as @ref dump_container, keeping the containers it has
     entered on an explicit stack instead of descending into them. Only reached
     for values nested deeper than @ref recursion_depth_limit, which is why it is not
     written for speed: walking every value this way measured up to 20% slower on
@@ -325,8 +310,7 @@ class serializer
                 {
                     if (pretty_print)
                     {
-                        put_char('\n');
-                        put_indent(frame.current_indent);
+                        put_literal_indent("\n", frame.current_indent);
                     }
 
                     put_char('}');
@@ -353,16 +337,13 @@ class serializer
                     put_indent(frame.child_indent);
                 }
 
-                put_char('"');
-                dump_escaped(frame.object_it->first);
-
                 if (pretty_print)
                 {
-                    put_literal("\": ");
+                    put_object_key(frame.object_it->first, "\": ");
                 }
                 else
                 {
-                    put_literal("\":");
+                    put_object_key(frame.object_it->first, "\":");
                 }
 
                 const BasicJsonType& element = frame.object_it->second;
@@ -381,8 +362,7 @@ class serializer
                 {
                     if (pretty_print)
                     {
-                        put_char('\n');
-                        put_indent(frame.current_indent);
+                        put_literal_indent("\n", frame.current_indent);
                     }
 
                     put_char(']');
@@ -523,7 +503,7 @@ class serializer
     /*!
     @brief serialize the value @a val, which is neither an object nor an array
 
-    Shared by @ref dump_internal and @ref dump_value, so that a value is written
+    Shared by @ref dump_element and @ref dump_value, so that a value is written
     the same way however deeply it is nested. A binary value is written out here
     in full: it looks like an object, but has no elements to descend into.
 
@@ -537,78 +517,13 @@ class serializer
         {
             case value_t::string:
             {
-                put_char('"');
-                dump_escaped(*val.m_data.m_value.string);
-                put_char('"');
+                dump_string_value(*val.m_data.m_value.string);
                 return;
             }
 
             case value_t::binary:
             {
-                if (pretty_print)
-                {
-                    put_literal("{\n");
-
-                    // variable to hold indentation for the bytes
-                    const auto new_indent = next_indent(current_indent, indent_step);
-
-                    put_indent(new_indent);
-
-                    put_literal("\"bytes\": [");
-
-                    if (!val.m_data.m_value.binary->empty())
-                    {
-                        for (auto i = val.m_data.m_value.binary->cbegin();
-                                i != val.m_data.m_value.binary->cend() - 1; ++i)
-                        {
-                            dump_byte(*i);
-                            put_literal(", ");
-                        }
-                        dump_byte(val.m_data.m_value.binary->back());
-                    }
-
-                    put_literal("],\n");
-                    put_indent(new_indent);
-
-                    put_literal("\"subtype\": ");
-                    if (val.m_data.m_value.binary->has_subtype())
-                    {
-                        dump_integer(val.m_data.m_value.binary->subtype());
-                    }
-                    else
-                    {
-                        put_literal("null");
-                    }
-                    put_char('\n');
-                    put_indent(current_indent);
-                    put_char('}');
-                }
-                else
-                {
-                    put_literal("{\"bytes\":[");
-
-                    if (!val.m_data.m_value.binary->empty())
-                    {
-                        for (auto i = val.m_data.m_value.binary->cbegin();
-                                i != val.m_data.m_value.binary->cend() - 1; ++i)
-                        {
-                            dump_byte(*i);
-                            put_char(',');
-                        }
-                        dump_byte(val.m_data.m_value.binary->back());
-                    }
-
-                    put_literal("],\"subtype\":");
-                    if (val.m_data.m_value.binary->has_subtype())
-                    {
-                        dump_integer(val.m_data.m_value.binary->subtype());
-                        put_char('}');
-                    }
-                    else
-                    {
-                        put_literal("null}");
-                    }
-                }
+                dump_binary_value(current_indent, *val.m_data.m_value.binary);
                 return;
             }
 
@@ -627,19 +542,19 @@ class serializer
 
             case value_t::number_integer:
             {
-                dump_integer(val.m_data.m_value.number_integer);
+                dump_signed_value(val.m_data.m_value.number_integer);
                 return;
             }
 
             case value_t::number_unsigned:
             {
-                dump_integer(val.m_data.m_value.number_unsigned);
+                dump_unsigned_value(val.m_data.m_value.number_unsigned);
                 return;
             }
 
             case value_t::number_float:
             {
-                dump_float(val.m_data.m_value.number_float);
+                dump_float_value(val.m_data.m_value.number_float);
                 return;
             }
 
@@ -662,6 +577,100 @@ class serializer
         }
     }
 
+    /// @brief write the string @a s in quotes; kept out of line, so that the
+    /// other scalars do not pay for its register needs (and the other way round)
+    JSON_HEDLEY_NEVER_INLINE void dump_string_value(const string_t& s)
+    {
+#if JSON_LITTLE_ENDIAN_TARGET
+        if (s.size() <= short_string_max && put_short_quoted(s, "\""))
+        {
+            return;
+        }
+#endif
+        put_char('"');
+        dump_escaped(s);
+        put_char('"');
+    }
+
+    // The numbers, and the rest of what dump_scalar() writes, are left to
+    // functions of their own as well: a single function holding all of them
+    // saves and restores every register any of them needs, on every call. Now
+    // dump_scalar() only dispatches, and each writer pays for what it uses.
+    JSON_HEDLEY_NEVER_INLINE void dump_signed_value(const number_integer_t x)
+    {
+        dump_integer(x);
+    }
+
+    JSON_HEDLEY_NEVER_INLINE void dump_unsigned_value(const number_unsigned_t x)
+    {
+        dump_integer(x);
+    }
+
+    JSON_HEDLEY_NEVER_INLINE void dump_float_value(const number_float_t x)
+    {
+        dump_float(x);
+    }
+
+    JSON_HEDLEY_NEVER_INLINE void dump_binary_value(const std::size_t current_indent, const typename BasicJsonType::binary_t& binary)
+    {
+        if (pretty_print)
+        {
+            // variable to hold indentation for the bytes
+            const auto new_indent = next_indent(current_indent, indent_step);
+
+            put_literal_indent("{\n", new_indent);
+            put_literal("\"bytes\": [");
+
+            if (!binary.empty())
+            {
+                for (auto i = binary.cbegin(); i != binary.cend() - 1; ++i)
+                {
+                    dump_byte(*i);
+                    put_literal(", ");
+                }
+                dump_byte(binary.back());
+            }
+
+            put_literal_indent("],\n", new_indent);
+
+            put_literal("\"subtype\": ");
+            if (binary.has_subtype())
+            {
+                dump_integer(binary.subtype());
+            }
+            else
+            {
+                put_literal("null");
+            }
+            put_literal_indent("\n", current_indent);
+            put_char('}');
+        }
+        else
+        {
+            put_literal("{\"bytes\":[");
+
+            if (!binary.empty())
+            {
+                for (auto i = binary.cbegin(); i != binary.cend() - 1; ++i)
+                {
+                    dump_byte(*i);
+                    put_char(',');
+                }
+                dump_byte(binary.back());
+            }
+
+            put_literal("],\"subtype\":");
+            if (binary.has_subtype())
+            {
+                dump_integer(binary.subtype());
+                put_char('}');
+            }
+            else
+            {
+                put_literal("null}");
+            }
+        }
+    }
 
     /*!
     @brief the indentation level to use for the children of the current value
@@ -678,6 +687,220 @@ class serializer
         return new_indent;
     }
 
+  private:
+    /*!
+    @brief append the object key @a s in quotes, followed by @a tail
+
+    @a tail is the closing quote and the colon, followed by a space when
+    pretty-printing. The callers pass it because each of them already sits on
+    the pretty-printing or the compact path; choosing it here would test
+    @ref pretty_print once more for every key. Most keys are short and plain
+    and are written in one go by put_short_quoted(); everything else takes the
+    general path.
+    */
+    template<std::size_t N>
+    void put_object_key(const string_t& s, const char (&tail)[N]) // NOLINT(cppcoreguidelines-avoid-c-arrays,hicpp-avoid-c-arrays,modernize-avoid-c-arrays)
+    {
+#if JSON_LITTLE_ENDIAN_TARGET
+        if (put_short_quoted(s, tail))
+        {
+            return;
+        }
+#endif
+        put_char('"');
+        dump_escaped(s);
+        put_literal(tail);
+    }
+
+    // put_short_quoted() assembles words from little-endian loads
+#if JSON_LITTLE_ENDIAN_TARGET
+    /// the longest string put_short_quoted() takes
+    static constexpr std::size_t short_string_max = JSON_SCAN_VECTOR ? 32 : 16;
+
+    /// the high bit of every byte of @a v that is not printable ASCII or is a
+    /// quote, a backslash, or DEL (zero if the bytes can be copied as they are)
+    static std::uint64_t swar_not_plain(const std::uint64_t v) noexcept
+    {
+        constexpr std::uint64_t ones = 0x0101010101010101ull;
+        const std::uint64_t q = v ^ 0x2222222222222222ull; // '"'
+        const std::uint64_t b = v ^ 0x5C5C5C5C5C5C5C5Cull; // '\\'
+        return (((q - ones) & ~q) | ((b - ones) & ~b)         // == '"', == '\\'
+                | ((v - 0x2020202020202020ull) & ~v)          // < 0x20
+                | v | (v + ones))                             // >= 0x7F
+               & 0x8080808080808080ull;
+    }
+
+    template<typename WordType>
+    static WordType load_word(const char* p) noexcept
+    {
+        WordType w;
+        std::memcpy(&w, p, sizeof(w));
+        return w;
+    }
+
+    template<typename WordType>
+    static void store_word(char* p, const WordType w) noexcept
+    {
+        std::memcpy(p, &w, sizeof(w));
+    }
+
+    /*!
+    @brief write `"`, @a s and @a tail if @a s is short and needs no escaping
+
+    A string of up to 16 (32 with vector instructions) printable ASCII
+    characters (no quote, backslash, or DEL) is read with two overlapping loads
+    and written with two overlapping stores, instead of scanning it and calling
+    memcpy with a length that is only known at run time. Returns false, having
+    changed nothing, if @a s does not qualify or the write buffer lacks the
+    room.
+    */
+    template<std::size_t N>
+    bool put_short_quoted(const string_t& s, const char (&tail)[N]) // NOLINT(cppcoreguidelines-avoid-c-arrays,hicpp-avoid-c-arrays,modernize-avoid-c-arrays)
+    {
+        const std::size_t n = s.size();
+        if (n > short_string_max || write_buffer_pos + n + N > write_buffer.size())
+        {
+            return false;
+        }
+
+        char* d = write_buffer.data() + write_buffer_pos + 1;
+        const char* const p = s.data();
+
+        if (n >= 8)
+        {
+#if JSON_SCAN_VECTOR
+            if (!(n >= 16 ? copy_plain_vectors(d, p, n) : copy_plain_halves(d, p, n)))
+            {
+                return false;
+            }
+#else
+            const auto a = load_word<std::uint64_t>(p);
+            const auto b = load_word<std::uint64_t>(p + n - 8);
+            if ((swar_not_plain(a) | swar_not_plain(b)) != 0)
+            {
+                return false;
+            }
+            store_word(d, a);
+            store_word(d + n - 8, b);
+#endif
+        }
+        else if (n >= 4)
+        {
+            const auto a = load_word<std::uint32_t>(p);
+            const auto b = load_word<std::uint32_t>(p + n - 4);
+            // the bytes above the string are filled with a plain character
+            const std::uint64_t w = a | (static_cast<std::uint64_t>(b) << (8 * (n - 4)))
+                                    | (0x4141414141414141ull << (8 * n));
+            if (swar_not_plain(w) != 0)
+            {
+                return false;
+            }
+            store_word(d, a);
+            store_word(d + n - 4, b);
+        }
+        else if (n >= 2)
+        {
+            const auto a = load_word<std::uint16_t>(p);
+            const auto b = load_word<std::uint16_t>(p + n - 2);
+            const std::uint64_t w = a | (static_cast<std::uint64_t>(b) << (8 * (n - 2)))
+                                    | (0x4141414141414141ull << (8 * n));
+            if (swar_not_plain(w) != 0)
+            {
+                return false;
+            }
+            store_word(d, a);
+            store_word(d + n - 2, b);
+        }
+        else if (n == 1)
+        {
+            const std::uint64_t w = static_cast<std::uint64_t>(static_cast<unsigned char>(p[0])) | (0x4141414141414141ull << 8u);
+            if (swar_not_plain(w) != 0)
+            {
+                return false;
+            }
+            *d = p[0];
+        }
+
+        d[-1] = '"';
+        std::memcpy(d + n, tail, N - 1);
+        write_buffer_pos += n + N;
+        return true;
+    }
+
+#if JSON_SCAN_VECTOR
+    /// copy the 8 to 15 bytes [@a p, @a p + @a n) to @a d with two overlapping
+    /// words, tested together as one vector, see copy_plain_vectors()
+    static bool copy_plain_halves(char* d, const char* p, const std::size_t n) noexcept
+    {
+#if JSON_SCAN_NEON
+        const uint8x8_t lo = vld1_u8(reinterpret_cast<const std::uint8_t*>(p));
+        const uint8x8_t hi = vld1_u8(reinterpret_cast<const std::uint8_t*>(p + n - 8));
+        const uint8x16_t in = vcombine_u8(lo, hi);
+        const uint8x16_t special = vorrq_u8(vorrq_u8(vceqq_u8(in, vdupq_n_u8('"')), vceqq_u8(in, vdupq_n_u8('\\'))),
+                                            vorrq_u8(vcltq_s8(vreinterpretq_s8_u8(in), vdupq_n_s8(0x20)), vceqq_u8(in, vdupq_n_u8(0x7F))));
+        if (vmaxvq_u8(special) != 0)
+        {
+            return false;
+        }
+        vst1_u8(reinterpret_cast<std::uint8_t*>(d), lo);
+        vst1_u8(reinterpret_cast<std::uint8_t*>(d + n - 8), hi);
+#else
+        const __m128i lo = _mm_loadl_epi64(static_cast<const __m128i*>(static_cast<const void*>(p))); // NOLINT(bugprone-casting-through-void)
+        const __m128i hi = _mm_loadl_epi64(static_cast<const __m128i*>(static_cast<const void*>(p + n - 8))); // NOLINT(bugprone-casting-through-void)
+        const __m128i in = _mm_unpacklo_epi64(lo, hi);
+        const __m128i special = _mm_or_si128(_mm_or_si128(_mm_cmpeq_epi8(in, _mm_set1_epi8('"')), _mm_cmpeq_epi8(in, _mm_set1_epi8('\\'))),
+                                             _mm_or_si128(_mm_cmplt_epi8(in, _mm_set1_epi8(0x20)), _mm_cmpeq_epi8(in, _mm_set1_epi8(0x7F))));
+        if (_mm_movemask_epi8(special) != 0)
+        {
+            return false;
+        }
+        _mm_storel_epi64(static_cast<__m128i*>(static_cast<void*>(d)), lo); // NOLINT(bugprone-casting-through-void)
+        _mm_storel_epi64(static_cast<__m128i*>(static_cast<void*>(d + n - 8)), hi); // NOLINT(bugprone-casting-through-void)
+#endif
+        return true;
+    }
+
+    /// copy the 16 to 32 bytes [@a p, @a p + @a n) to @a d with two overlapping
+    /// vectors, unless one of them is a quote, a backslash, a control
+    /// character, DEL or not ASCII
+    static bool copy_plain_vectors(char* d, const char* p, const std::size_t n) noexcept
+    {
+#if JSON_SCAN_NEON
+        const uint8x16_t a = vld1q_u8(reinterpret_cast<const std::uint8_t*>(p));
+        const uint8x16_t b = vld1q_u8(reinterpret_cast<const std::uint8_t*>(p + n - 16));
+        const uint8x16_t special = vorrq_u8(vorrq_u8(vorrq_u8(vceqq_u8(a, vdupq_n_u8('"')), vceqq_u8(a, vdupq_n_u8('\\'))),
+                                            vorrq_u8(vceqq_u8(b, vdupq_n_u8('"')), vceqq_u8(b, vdupq_n_u8('\\')))),
+                                            vorrq_u8(vorrq_u8(vcltq_s8(vreinterpretq_s8_u8(a), vdupq_n_s8(0x20)), vcltq_s8(vreinterpretq_s8_u8(b), vdupq_n_s8(0x20))),
+                                                vorrq_u8(vceqq_u8(a, vdupq_n_u8(0x7F)), vceqq_u8(b, vdupq_n_u8(0x7F)))));
+        if (vmaxvq_u8(special) != 0)
+        {
+            return false;
+        }
+        vst1q_u8(reinterpret_cast<std::uint8_t*>(d), a);
+        vst1q_u8(reinterpret_cast<std::uint8_t*>(d + n - 16), b);
+#else
+        const __m128i a = _mm_loadu_si128(static_cast<const __m128i*>(static_cast<const void*>(p))); // NOLINT(bugprone-casting-through-void)
+        const __m128i b = _mm_loadu_si128(static_cast<const __m128i*>(static_cast<const void*>(p + n - 16))); // NOLINT(bugprone-casting-through-void)
+        const __m128i quote = _mm_set1_epi8('"');
+        const __m128i backslash = _mm_set1_epi8('\\');
+        const __m128i space = _mm_set1_epi8(0x20);
+        const __m128i del = _mm_set1_epi8(0x7F);
+        const __m128i special = _mm_or_si128(_mm_or_si128(_mm_or_si128(_mm_cmpeq_epi8(a, quote), _mm_cmpeq_epi8(a, backslash)),
+                                             _mm_or_si128(_mm_cmpeq_epi8(b, quote), _mm_cmpeq_epi8(b, backslash))),
+                                             _mm_or_si128(_mm_or_si128(_mm_cmplt_epi8(a, space), _mm_cmplt_epi8(b, space)),
+                                                 _mm_or_si128(_mm_cmpeq_epi8(a, del), _mm_cmpeq_epi8(b, del))));
+        if (_mm_movemask_epi8(special) != 0)
+        {
+            return false;
+        }
+        _mm_storeu_si128(static_cast<__m128i*>(static_cast<void*>(d)), a); // NOLINT(bugprone-casting-through-void)
+        _mm_storeu_si128(static_cast<__m128i*>(static_cast<void*>(d + n - 16)), b); // NOLINT(bugprone-casting-through-void)
+#endif
+        return true;
+    }
+#endif
+#endif
+
   JSON_PRIVATE_UNLESS_TESTED:
     /*!
     @brief dump escaped string
@@ -690,14 +913,23 @@ class serializer
     @param[in] s  the string to escape
 
     Complexity: Linear in the length of string @a s.
+
+    @note kept out of line: with the short-string attempts in front of it gone
+          from here, compilers would otherwise inline it into @ref put_object_key
+          and then no longer into the loops that call that
     */
-    void dump_escaped(const string_t& s)
+    JSON_HEDLEY_NEVER_INLINE void dump_escaped(const string_t& s)
     {
         // Short strings of printable ASCII - most object keys and many values -
         // are checked byte by byte against a table and then copied in one go.
         // For them, this is cheaper than starting the vector scanner below,
         // whose setup costs more than such a string takes to scan. The same
         // check serves both values of ensure_ascii.
+        //
+        // Where put_short_quoted() exists and takes strings of up to 32 bytes,
+        // the callers have tried it already, and repeating the attempt only
+        // costs time for strings that need escaping or are not ASCII.
+#if !JSON_LITTLE_ENDIAN_TARGET || !JSON_SCAN_VECTOR
         if (s.size() <= 32)
         {
             const auto& copyable = ascii_copyable();
@@ -719,6 +951,7 @@ class serializer
                 return;
             }
         }
+#endif
 
         // dispatch once here rather than test the flag inside the loop: it does
         // not change while a string is written, and folding it lets each of the
@@ -796,7 +1029,7 @@ class serializer
                         put_buffer(string_buffer, bytes);
                         bytes = 0;
                     }
-                    put_string(s, i, i + run);
+                    put_run(s, i, run);
                     bytes_after_last_accept = 0;
                     undumped_chars = 0;
                     i += run;
@@ -836,6 +1069,27 @@ class serializer
 
             const auto byte = static_cast<std::uint8_t>(s[i]);
 
+            if (EnsureAscii && state == UTF8_ACCEPT && byte >= 0x80)
+            {
+                // a run of well-formed multi-byte characters is escaped
+                // straight into the write buffer; an ill-formed sequence
+                // ends the run and is left to the byte-at-a-time path
+                if (bytes != 0)
+                {
+                    put_buffer(string_buffer, bytes);
+                    bytes = 0;
+                    bytes_after_last_accept = 0;
+                }
+                const std::size_t escaped = write_u_escapes(reinterpret_cast<const unsigned char*>(s.data()) + i, s.size() - i);
+                if (escaped != 0)
+                {
+                    bytes_after_last_accept = 0;
+                    undumped_chars = 0;
+                    i += escaped - 1;
+                    continue;
+                }
+            }
+
             switch (decode(state, codepoint, byte))
             {
                 case UTF8_ACCEPT:  // decode found a new code point
@@ -1113,6 +1367,12 @@ class serializer
             return;
         }
 
+        // the usual indentation, without a call of memset
+        if (JSON_HEDLEY_LIKELY(put_short_literal_indent("", indent)))
+        {
+            return;
+        }
+
         const std::size_t capacity = write_buffer.size();
 
         // fill whatever room is left in the buffer; this is the whole job
@@ -1146,6 +1406,51 @@ class serializer
         write_buffer_pos = indent;
     }
 
+    /*!
+    @brief append a string literal and then @a indent indentation characters
+
+    The usual line break and indentation go into the buffer with one check of
+    its room and four stores of the indentation, see @ref put_short_literal_indent.
+    */
+    template<std::size_t N>
+    JSON_HEDLEY_ALWAYS_INLINE void put_literal_indent(const char (&s)[N], const std::size_t indent) // NOLINT(cppcoreguidelines-avoid-c-arrays,hicpp-avoid-c-arrays,modernize-avoid-c-arrays)
+    {
+        if (JSON_HEDLEY_LIKELY(put_short_literal_indent(s, indent)))
+        {
+            return;
+        }
+        put_literal(s);
+        put_indent(indent);
+    }
+
+    /*!
+    @brief append a string literal and up to 32 indentation characters in one go
+
+    Four words of the indentation character are stored unconditionally and only
+    the part wanted is claimed, which spares the call of memset. Shared by
+    @ref put_indent (with an empty literal, whose copy compiles to nothing) and
+    @ref put_literal_indent. Returns false, having written nothing, if
+    @a indent is wider or the write buffer lacks the room.
+    */
+    template<std::size_t N>
+    JSON_HEDLEY_ALWAYS_INLINE bool put_short_literal_indent(const char (&s)[N], const std::size_t indent) // NOLINT(cppcoreguidelines-avoid-c-arrays,hicpp-avoid-c-arrays,modernize-avoid-c-arrays)
+    {
+        constexpr std::size_t length = N - 1;
+        if (indent > 32 || write_buffer_pos + length + 32 > write_buffer.size())
+        {
+            return false;
+        }
+        const std::uint64_t word = 0x0101010101010101ull * static_cast<unsigned char>(indent_char);
+        char* const d = write_buffer.data() + write_buffer_pos;
+        std::memcpy(d, s, length);
+        std::memcpy(d + length, &word, sizeof(word));
+        std::memcpy(d + length + 8, &word, sizeof(word));
+        std::memcpy(d + length + 16, &word, sizeof(word));
+        std::memcpy(d + length + 24, &word, sizeof(word));
+        write_buffer_pos += length + indent;
+        return true;
+    }
+
     /*!
     @brief append a string literal to the write buffer
 
@@ -1170,6 +1475,28 @@ class serializer
         write_buffer_pos += length;
     }
 
+    /*!
+    @brief append the characters of @a str in [@a start, @a start + @a length)
+
+    A run of up to 16 characters with 16 or more in the string from @a start on
+    and in the write buffer from its end on is copied as a whole block of 16,
+    which spares the call of memcpy with a length known only at run time; the
+    surplus characters in the buffer are overwritten by what follows.
+    */
+    template<typename StringType>
+    void put_run(const StringType& str, const std::size_t start, const std::size_t length)
+    {
+        // a short run is the exception in text without escapes (CJK, accented
+        // Latin), for which this keeps the long path the fast one
+        if (JSON_HEDLEY_UNLIKELY(length <= 16 && start + 16 <= str.size() && write_buffer_pos + 16 <= write_buffer.size()))
+        {
+            std::memcpy(write_buffer.data() + write_buffer_pos, str.data() + start, 16);
+            write_buffer_pos += length;
+            return;
+        }
+        put_string(str, start, start + length);
+    }
+
     /*!
     @brief append the characters of @a str in [@a start, @a end)
 
@@ -1412,6 +1739,78 @@ class serializer
         return escapes;
     }
 
+    /*!
+    @brief escape the well-formed multi-byte characters at the start of @a data
+
+    Writes the "\\uXXXX" escape of each character (two of them for a character
+    above U+FFFF) straight into the write buffer, which is why the caller must
+    have written out what is pending in @ref string_buffer. Stops at the first
+    byte that is below 0x80 or starts an ill-formed or truncated sequence, which
+    the caller handles.
+
+    @return the number of bytes of @a data that have been escaped
+    */
+    JSON_HEDLEY_NEVER_INLINE std::size_t write_u_escapes(const unsigned char* data, std::size_t n)
+    {
+        constexpr const char* nibble_to_hex = "0123456789abcdef";
+        std::size_t i = 0;
+        while (i < n && data[i] >= 0x80)
+        {
+            const std::size_t length = validate_one_utf8(data + i, n - i);
+            if (length == 0)
+            {
+                break;
+            }
+
+            // a lead byte of 0x80 or above starts a sequence of 2 to 4 bytes
+            JSON_ASSERT(length >= 2 && length <= 4);
+
+            const auto b0 = static_cast<std::uint32_t>(data[i]);
+            const auto b1 = static_cast<std::uint32_t>(data[i + 1]);
+            std::uint32_t codepoint = 0;
+            switch (length)
+            {
+                case 2:
+                    codepoint = ((b0 & 0x1Fu) << 6u) | (b1 & 0x3Fu);
+                    break;
+                case 3:
+                    codepoint = ((b0 & 0x0Fu) << 12u) | ((b1 & 0x3Fu) << 6u) | (static_cast<std::uint32_t>(data[i + 2]) & 0x3Fu);
+                    break;
+                default: // 4
+                    codepoint = ((b0 & 0x07u) << 18u) | ((b1 & 0x3Fu) << 12u) | ((static_cast<std::uint32_t>(data[i + 2]) & 0x3Fu) << 6u) | (static_cast<std::uint32_t>(data[i + 3]) & 0x3Fu);
+                    break;
+            }
+
+            // at most two escapes of six bytes each
+            if (JSON_HEDLEY_UNLIKELY(write_buffer.size() - write_buffer_pos < 12))
+            {
+                flush();
+            }
+            char* d = write_buffer.data() + write_buffer_pos;
+            if (codepoint > 0xFFFFu)
+            {
+                d = put_u_escape(d, static_cast<std::uint16_t>(0xD7C0u + (codepoint >> 10u)), nibble_to_hex);
+                codepoint = 0xDC00u + (codepoint & 0x3FFu);
+            }
+            d = put_u_escape(d, static_cast<std::uint16_t>(codepoint), nibble_to_hex);
+            write_buffer_pos = static_cast<std::size_t>(d - write_buffer.data());
+            i += length;
+        }
+        return i;
+    }
+
+    static char* put_u_escape(char* d, const std::uint16_t codeunit, const char* nibble_to_hex) noexcept
+    {
+        const auto unit = static_cast<std::uint32_t>(codeunit);
+        d[0] = '\\';
+        d[1] = 'u';
+        d[2] = nibble_to_hex[(unit >> 12u) & 0x0Fu];
+        d[3] = nibble_to_hex[(unit >> 8u) & 0x0Fu];
+        d[4] = nibble_to_hex[(unit >> 4u) & 0x0Fu];
+        d[5] = nibble_to_hex[unit & 0x0Fu];
+        return d + 6;
+    }
+
     /*!
      * @brief write a lowercase "\\uXXXX" escape sequence into @a string_buffer
      *
@@ -1769,11 +2168,22 @@ class serializer
     /// the output of the serializer (non-owning; the adapter lives at the call site)
     output_adapter_protocol<char>* o = nullptr;
 
+    /// a character buffer left uninitialized: the buffers below are written
+    /// before they are read (only the part written is ever read), and zeroing
+    /// 1.6 KiB took more than half of what dumping a small value costs (the
+    /// user-provided constructor keeps the member initialization, which
+    /// -Weffc++ asks for, from zeroing them)
+    template<std::size_t N>
+    struct uninitialized_buffer : std::array<char, N>
+    {
+        uninitialized_buffer() noexcept {} // NOLINT(modernize-use-equals-default,hicpp-use-equals-default,cppcoreguidelines-pro-type-member-init,hicpp-member-init): see above
+    };
+
     /// a (hopefully) large enough character buffer
-    std::array<char, 64> number_buffer{{}};
+    uninitialized_buffer<64> number_buffer;
 
     /// string buffer
-    std::array<char, 512> string_buffer{{}};
+    uninitialized_buffer<512> string_buffer;
 
     /// the indentation character
     const char indent_char;
@@ -1793,7 +2203,7 @@ class serializer
     /// buffer collecting output before it is flushed to the output adapter, so
     /// that the many small structural writes become few bulk writes
     static constexpr std::size_t write_buffer_size = 1024;
-    std::array<char, write_buffer_size> write_buffer{{}};
+    uninitialized_buffer<write_buffer_size> write_buffer;
     /// number of valid bytes currently held in @ref write_buffer
     std::size_t write_buffer_pos = 0;
 };

include/nlohmann/detail/view/image.hpp

@@ -255,81 +255,6 @@ struct number_token
     bool is_float; ///< a fraction or an exponent
 };
 
-/// advance over ASCII digits, eight bytes at a time
-inline const unsigned char* digit_run(const unsigned char* p, const unsigned char* e) noexcept
-{
-    constexpr std::uint64_t ones = 0x0101010101010101ull;
-    while (e - p >= 8)
-    {
-        // (the digits become 0..9; a byte above 9, or with bit 7 set, is no digit. A carry
-        // out of a byte with bit 7 set only affects the bytes above, after the first one
-        // that count_trailing_zeros finds.)
-        const std::uint64_t x = read_eight_bytes(p) ^ (ones * 0x30u);
-        const std::uint64_t not_digit = ((x + (ones * 0x76u)) | x) & (ones * 0x80u);
-        if (not_digit != 0)
-        {
-            return p + (static_cast<std::size_t>(count_trailing_zeros(not_digit)) / 8);
-        }
-        p += 8;
-    }
-    while (p != e && is_digit(*p))
-    {
-        ++p;
-    }
-    return p;
-}
-
-/// whether the n bytes at p are ASCII digits (n > 0)
-inline bool all_digits(const unsigned char* p, std::size_t n) noexcept
-{
-    const auto digits = [](const unsigned char* q)
-    {
-        constexpr std::uint64_t ones = 0x0101010101010101ull;
-        const std::uint64_t x = read_eight_bytes(q) ^ (ones * 0x30u);
-        return (((x + (ones * 0x76u)) | x) & (ones * 0x80u)) == 0;
-    };
-    if (n >= 8)
-    {
-        for (std::size_t i = 0; i + 8 < n; i += 8)
-        {
-            if (!digits(p + i))
-            {
-                return false;
-            }
-        }
-        return digits(p + n - 8); // (overlapping the bytes before: they are digits)
-    }
-    for (std::size_t i = 0; i < n; ++i)
-    {
-        if (!is_digit(p[i]))
-        {
-            return false;
-        }
-    }
-    return true;
-}
-
-/// Whether a float token has the shape that the parser records for most of
-/// them: [-]digits.digits, with the numbers of digits of the node's layout
-/// (extra). Then the token is a number of that layout, and the general scan
-/// that follows is not needed.
-inline bool plain_float_token(const unsigned char* s, std::size_t len, std::uint16_t extra) noexcept
-{
-    const std::size_t int_digits = extra & 0xFFu;
-    const std::size_t frac_digits = extra >> 8u;
-    if (int_digits == 0 || int_digits >= 255 || frac_digits == 0 || frac_digits >= 255 || len == 0)
-    {
-        return false;
-    }
-    const std::size_t negative = *s == '-' ? 1 : 0;
-    if (len != negative + int_digits + 1 + frac_digits)
-    {
-        return false;
-    }
-    const unsigned char* const p = s + negative;
-    return !(int_digits > 1 && *p == '0') && p[int_digits] == '.' && all_digits(p, int_digits) && all_digits(p + int_digits + 1, frac_digits);
-}
-
 /// whether [s, s + len) is a JSON number (the grammar the parser accepts)
 inline bool scan_number_token(const unsigned char* s, std::size_t len, number_token& t)
 {
@@ -348,7 +273,11 @@ inline bool scan_number_token(const unsigned char* s, std::size_t len, number_to
     }
     else if (*p >= '1' && *p <= '9')
     {
-        p = digit_run(p + 1, e);
+        do
+        {
+            ++p;
+        }
+        while (p != e && is_digit(*p));
     }
     else
     {
@@ -360,7 +289,10 @@ inline bool scan_number_token(const unsigned char* s, std::size_t len, number_to
     if (p != e && *p == '.')
     {
         const unsigned char* const f0 = ++p;
-        p = digit_run(p, e);
+        while (p != e && is_digit(*p))
+        {
+            ++p;
+        }
         if (p == f0)
         {
             return false;
@@ -471,15 +403,10 @@ struct float_range
 template<typename Check>
 bool check_distinct_ranges(std::vector<byte_range>& ranges, Check check)
 {
-    // (save() writes the ranges in the order of the nodes, which is sorted already)
-    const auto before = [](const byte_range & a, const byte_range & b)
+    std::sort(ranges.begin(), ranges.end(), [](const byte_range & a, const byte_range & b)
     {
         return a.off != b.off ? a.off < b.off : a.len < b.len;
-    };
-    if (!std::is_sorted(ranges.begin(), ranges.end(), before))
-    {
-        std::sort(ranges.begin(), ranges.end(), before);
-    }
+    });
     std::size_t end = 0;
     for (std::size_t i = 0; i < ranges.size(); ++i)
     {
@@ -501,8 +428,7 @@ bool check_distinct_ranges(std::vector<byte_range>& ranges, Check check)
 /// record layouts that match it), tokens must not overlap.
 inline bool check_float_ranges(std::vector<float_range>& ranges, const unsigned char* text)
 {
-    // (save() writes the tokens in the order of the nodes, which is sorted already)
-    const auto before = [](const float_range & a, const float_range & b)
+    std::sort(ranges.begin(), ranges.end(), [](const float_range & a, const float_range & b)
     {
         if (a.off != b.off)
         {
@@ -513,11 +439,7 @@ inline bool check_float_ranges(std::vector<float_range>& ranges, const unsigned
             return a.len < b.len;
         }
         return a.extra < b.extra;
-    };
-    if (!std::is_sorted(ranges.begin(), ranges.end(), before))
-    {
-        std::sort(ranges.begin(), ranges.end(), before);
-    }
+    });
     std::size_t end = 0;
     std::size_t i = 0;
     while (i < ranges.size())
@@ -527,18 +449,14 @@ inline bool check_float_ranges(std::vector<float_range>& ranges, const unsigned
         {
             return false;
         }
-        const unsigned char* const s = text + r.off;
-        std::uint16_t layout = r.extra;
-        if (!plain_float_token(s, r.len, r.extra))
-        {
         number_token t{};
+        const unsigned char* const s = text + r.off;
         if (!scan_number_token(s, r.len, t) || !float_token_finite(s, r.len, t))
         {
             return false;
         }
         // the digit layout the parser records (or "many", as compaction writes it)
-            layout = float_layout(t);
-        }
+        const std::uint16_t layout = float_layout(t);
         for (; i < ranges.size() && ranges[i].off == r.off && ranges[i].len == r.len; ++i)
         {
             if (ranges[i].extra != layout && ranges[i].extra != 0xFFFFu)

include/nlohmann/json.hpp

@@ -19,6 +19,7 @@
 #define INCLUDE_NLOHMANN_JSON_HPP_
 
 #include <algorithm> // all_of, find, for_each, none_of
+#include <array> // array
 #include <cmath> // isnan
 #include <cstddef> // nullptr_t, ptrdiff_t, size_t
 #include <cstdint> // uint8_t
@@ -6321,7 +6322,7 @@ public:
             const error_handler_t error_handler = detail::binary_writer_default_error_handler())
     {
         std::vector<std::uint8_t> result;
-        result.reserve(detail::binary_reserve_hint(j));
+        result.reserve(detail::binary_initial_capacity(j));
         vector_writer(result, error_handler).write_cbor(j);
         return result;
     }
@@ -6331,7 +6332,9 @@ public:
     static void to_cbor(const basic_json& j, detail::output_adapter<std::uint8_t> o,
                         const error_handler_t error_handler = detail::binary_writer_default_error_handler())
     {
-        binary_writer<std::uint8_t>(o, error_handler).write_cbor(j);
+        binary_writer<std::uint8_t> writer(o, error_handler);
+        writer.write_cbor(j);
+        writer.flush();
     }
 
     /// @brief create a CBOR serialization of a given JSON value
@@ -6339,7 +6342,9 @@ public:
     static void to_cbor(const basic_json& j, detail::output_adapter<char> o,
                         const error_handler_t error_handler = detail::binary_writer_default_error_handler())
     {
-        binary_writer<char>(o, error_handler).write_cbor(j);
+        binary_writer<char> writer(o, error_handler);
+        writer.write_cbor(j);
+        writer.flush();
     }
 
     /// @brief create a MessagePack serialization of a given JSON value
@@ -6348,7 +6353,7 @@ public:
             const error_handler_t error_handler = error_handler_t::keep)
     {
         std::vector<std::uint8_t> result;
-        result.reserve(detail::binary_reserve_hint(j));
+        result.reserve(detail::binary_initial_capacity(j));
         vector_writer(result, error_handler).write_msgpack(j);
         return result;
     }
@@ -6358,7 +6363,9 @@ public:
     static void to_msgpack(const basic_json& j, detail::output_adapter<std::uint8_t> o,
                            const error_handler_t error_handler = error_handler_t::keep)
     {
-        binary_writer<std::uint8_t>(o, error_handler).write_msgpack(j);
+        binary_writer<std::uint8_t> writer(o, error_handler);
+        writer.write_msgpack(j);
+        writer.flush();
     }
 
     /// @brief create a MessagePack serialization of a given JSON value
@@ -6366,7 +6373,9 @@ public:
     static void to_msgpack(const basic_json& j, detail::output_adapter<char> o,
                            const error_handler_t error_handler = error_handler_t::keep)
     {
-        binary_writer<char>(o, error_handler).write_msgpack(j);
+        binary_writer<char> writer(o, error_handler);
+        writer.write_msgpack(j);
+        writer.flush();
     }
 
     /// @brief create a UBJSON serialization of a given JSON value
@@ -6377,7 +6386,7 @@ public:
             const error_handler_t error_handler = detail::binary_writer_default_error_handler())
     {
         std::vector<std::uint8_t> result;
-        result.reserve(detail::binary_reserve_hint(j));
+        result.reserve(detail::binary_initial_capacity(j));
         vector_writer(result, error_handler).write_ubjson(j, use_size, use_type);
         return result;
     }
@@ -6388,7 +6397,9 @@ public:
                           const bool use_size = false, const bool use_type = false,
                           const error_handler_t error_handler = detail::binary_writer_default_error_handler())
     {
-        binary_writer<std::uint8_t>(o, error_handler).write_ubjson(j, use_size, use_type);
+        binary_writer<std::uint8_t> writer(o, error_handler);
+        writer.write_ubjson(j, use_size, use_type);
+        writer.flush();
     }
 
     /// @brief create a UBJSON serialization of a given JSON value
@@ -6397,7 +6408,9 @@ public:
                           const bool use_size = false, const bool use_type = false,
                           const error_handler_t error_handler = detail::binary_writer_default_error_handler())
     {
-        binary_writer<char>(o, error_handler).write_ubjson(j, use_size, use_type);
+        binary_writer<char> writer(o, error_handler);
+        writer.write_ubjson(j, use_size, use_type);
+        writer.flush();
     }
 
     /// @brief create a BJData serialization of a given JSON value
@@ -6409,7 +6422,7 @@ public:
             const error_handler_t error_handler = detail::binary_writer_default_error_handler())
     {
         std::vector<std::uint8_t> result;
-        result.reserve(detail::binary_reserve_hint(j));
+        result.reserve(detail::binary_initial_capacity(j));
         vector_writer(result, error_handler).write_ubjson(j, use_size, use_type, true, true, version);
         return result;
     }
@@ -6421,7 +6434,9 @@ public:
                           const bjdata_version_t version = bjdata_version_t::draft2,
                           const error_handler_t error_handler = detail::binary_writer_default_error_handler())
     {
-        binary_writer<std::uint8_t>(o, error_handler).write_ubjson(j, use_size, use_type, true, true, version);
+        binary_writer<std::uint8_t> writer(o, error_handler);
+        writer.write_ubjson(j, use_size, use_type, true, true, version);
+        writer.flush();
     }
 
     /// @brief create a BJData serialization of a given JSON value
@@ -6431,7 +6446,9 @@ public:
                           const bjdata_version_t version = bjdata_version_t::draft2,
                           const error_handler_t error_handler = detail::binary_writer_default_error_handler())
     {
-        binary_writer<char>(o, error_handler).write_ubjson(j, use_size, use_type, true, true, version);
+        binary_writer<char> writer(o, error_handler);
+        writer.write_ubjson(j, use_size, use_type, true, true, version);
+        writer.flush();
     }
 
     /// @brief create a BSON serialization of a given JSON value
@@ -6440,7 +6457,7 @@ public:
             const error_handler_t error_handler = detail::binary_writer_default_error_handler())
     {
         std::vector<std::uint8_t> result;
-        result.reserve(detail::binary_reserve_hint(j));
+        result.reserve(detail::binary_initial_capacity(j));
         vector_writer(result, error_handler).write_bson(j);
         return result;
     }
@@ -6450,7 +6467,9 @@ public:
     static void to_bson(const basic_json& j, detail::output_adapter<std::uint8_t> o,
                         const error_handler_t error_handler = detail::binary_writer_default_error_handler())
     {
-        binary_writer<std::uint8_t>(o, error_handler).write_bson(j);
+        binary_writer<std::uint8_t> writer(o, error_handler);
+        writer.write_bson(j);
+        writer.flush();
     }
 
     /// @brief create a BSON serialization of a given JSON value
@@ -6458,7 +6477,9 @@ public:
     static void to_bson(const basic_json& j, detail::output_adapter<char> o,
                         const error_handler_t error_handler = detail::binary_writer_default_error_handler())
     {
-        binary_writer<char>(o, error_handler).write_bson(j);
+        binary_writer<char> writer(o, error_handler);
+        writer.write_bson(j);
+        writer.flush();
     }
 
     /// @brief create a BON8 serialization of a given JSON value
@@ -6466,7 +6487,7 @@ public:
     static std::vector<std::uint8_t> to_bon8(const basic_json& j)
     {
         std::vector<std::uint8_t> result;
-        result.reserve(detail::binary_reserve_hint(j));
+        result.reserve(detail::binary_initial_capacity(j));
         vector_writer(result).write_bon8(j);
         return result;
     }
@@ -6475,14 +6496,18 @@ public:
     /// @sa https://json.nlohmann.me/api/basic_json/to_bon8/
     static void to_bon8(const basic_json& j, detail::output_adapter<std::uint8_t> o)
     {
-        binary_writer<std::uint8_t>(o).write_bon8(j);
+        binary_writer<std::uint8_t> writer(o);
+        writer.write_bon8(j);
+        writer.flush();
     }
 
     /// @brief create a BON8 serialization of a given JSON value
     /// @sa https://json.nlohmann.me/api/basic_json/to_bon8/
     static void to_bon8(const basic_json& j, detail::output_adapter<char> o)
     {
-        binary_writer<char>(o).write_bon8(j);
+        binary_writer<char> writer(o);
+        writer.write_bon8(j);
+        writer.flush();
     }
 
     /// @brief create a JSON value from an input in CBOR format
@@ -6878,6 +6903,72 @@ public:
     /// @name JSON Patch functions
     /// @{
 
+  private:
+    /// the members of an operation of a JSON patch that are of interest
+    struct patch_members
+    {
+        basic_json* op = nullptr;
+        basic_json* path = nullptr;
+        basic_json* value = nullptr;
+        basic_json* from = nullptr;
+    };
+
+    /// @brief find the members "op", "path", "value", and "from" of the object @a val in one pass
+    static patch_members find_patch_members(const basic_json& val, std::true_type /*key_type is string_t*/)
+    {
+        patch_members members;
+        for (auto& member : *val.m_data.m_value.object)
+        {
+            const string_t& key = member.first;
+            switch (key.size())
+            {
+                case 2:
+                    if (key == "op")
+                    {
+                        members.op = &member.second;
+                    }
+                    break;
+                case 4:
+                    if (key == "path")
+                    {
+                        members.path = &member.second;
+                    }
+                    else if (key == "from")
+                    {
+                        members.from = &member.second;
+                    }
+                    break;
+                case 5:
+                    if (key == "value")
+                    {
+                        members.value = &member.second;
+                    }
+                    break;
+                default:
+                    break;
+            }
+        }
+        return members;
+    }
+
+    /// @brief find the members by lookup for key types other than string_t
+    static patch_members find_patch_members(const basic_json& val, std::false_type /*key_type is not string_t*/)
+    {
+        const auto find_member = [&val](const char* name) -> basic_json *
+        {
+            const auto it = val.m_data.m_value.object->find(string_t(name));
+            return it != val.m_data.m_value.object->end() ? &it->second : nullptr;
+        };
+
+        patch_members members;
+        members.op = find_member("op");
+        members.path = find_member("path");
+        members.value = find_member("value");
+        members.from = find_member("from");
+        return members;
+    }
+
+  public:
     /// @brief applies a JSON patch in-place without copying the object
     /// @sa https://json.nlohmann.me/api/basic_json/patch/
     void patch_inplace(const basic_json& json_patch)
@@ -7051,18 +7142,18 @@ public:
         // a diff creates them: apply the operation and return true; return
         // false, with nothing changed, for all others
         string_t token;
-        const string_t value_key = "value";
-        const auto try_plain = [&result, &token, &value_key](patch_operations operation, const string_t& path, const basic_json & val)
+        typename json_pointer::template plain_steps<basic_json> steps;
+        const auto try_plain = [&result, &token, &steps](patch_operations operation, const string_t& path, const patch_members & members, const basic_json & val)
         {
-            if (path.empty() || path.find_first_of('~') != string_t::npos)
+            if (path.empty())
             {
                 return false;
             }
 
             // a missing "value" is reported below
-            const auto value_it = val.m_data.m_value.object->find(value_key);
-            const basic_json& value = value_it != val.m_data.m_value.object->end() ? value_it->second : val;
-            if (operation != patch_operations::remove && value_it == val.m_data.m_value.object->end())
+            const basic_json* const value_ptr = members.value;
+            const basic_json& value = value_ptr != nullptr ? *value_ptr : val;
+            if (operation != patch_operations::remove && value_ptr == nullptr)
             {
                 return false;
             }
@@ -7071,7 +7162,7 @@ public:
             {
                 case patch_operations::replace:
                 {
-                    basic_json* const target = json_pointer::resolve_plain(&result, path, path.size(), token);
+                    basic_json* const target = json_pointer::resolve_plain(&result, path, path.size(), token, steps);
                     if (target == nullptr)
                     {
                         return false;
@@ -7082,7 +7173,7 @@ public:
 
                 case patch_operations::test:
                 {
-                    const basic_json* const target = json_pointer::resolve_plain(&result, path, path.size(), token);
+                    const basic_json* const target = json_pointer::resolve_plain(&result, path, path.size(), token, steps);
                     // a failing test is reported below
                     return target != nullptr && *target == value;
                 }
@@ -7102,13 +7193,17 @@ public:
                     }
                     --slash;
 
-                    basic_json* const parent = json_pointer::resolve_plain(&result, path, slash, token);
+                    basic_json* const parent = json_pointer::resolve_plain(&result, path, slash, token, steps);
                     if (parent == nullptr)
                     {
                         return false;
                     }
                     token.clear();
                     token.append(path.data() + slash + 1, path.size() - slash - 1);
+                    if (token.find_first_of('~') != string_t::npos)
+                    {
+                        return false;
+                    }
 
                     if (operation == patch_operations::add)
                     {
@@ -7164,31 +7259,32 @@ public:
             }
         };
 
+        // the name of the first member to check
+        const string_t op_member = "op";
+
         // iterate and apply the operations
         for (const auto& val : json_patch)
         {
             // wrapper to get a value for an operation
-            const auto get_value = [&val](const string_t& op,
-                                          const string_t& member,
+            const auto get_value = [&val](basic_json * found,
+                                          const string_t& op,
+                                          const char* member,
                                           bool string_type) -> basic_json &
             {
-                // find value
-                auto it = val.m_data.m_value.object->find(member);
-
                 // check if the desired value is present
-                if (JSON_HEDLEY_UNLIKELY(it == val.m_data.m_value.object->end()))
+                if (JSON_HEDLEY_UNLIKELY(found == nullptr))
                 {
-                    detail::throw_patch_missing_member(op, member, false, val);
+                    detail::throw_patch_missing_member(op, string_t(member), false, val);
                 }
 
                 // check if the result is of type string
-                if (JSON_HEDLEY_UNLIKELY(string_type && !it->second.is_string()))
+                if (JSON_HEDLEY_UNLIKELY(string_type && !found->is_string()))
                 {
-                    detail::throw_patch_missing_member(op, member, true, val);
+                    detail::throw_patch_missing_member(op, string_t(member), true, val);
                 }
 
                 // no error: return value
-                return it->second;
+                return *found;
             };
 
             // type check: every element of the array must be an object
@@ -7197,26 +7293,30 @@ public:
                 detail::throw_patch_not_an_array(val);
             }
 
+            const patch_members members = find_patch_members(val, std::is_same<typename object_t::key_type, string_t> {});
+
             // collect mandatory members
-            const string_t& op = *get_value("op", "op", true).m_data.m_value.string;
-            const string_t& path = *get_value(op, "path", true).m_data.m_value.string;
+            const string_t& op = *get_value(members.op, op_member, "op", true).m_data.m_value.string;
+            const string_t& path = *get_value(members.path, op, "path", true).m_data.m_value.string;
             const patch_operations operation = get_op(op);
 
             // the usual operations on a plain pointer are applied without
             // building the pointer's reference tokens; anything else (and
             // every error) is handled below
-            if (try_plain(operation, path, val))
+            if (try_plain(operation, path, members, val))
             {
                 continue;
             }
 
+            // the general operations may change any container
+            steps.reset();
             json_pointer ptr(path);
 
             switch (operation)
             {
                 case patch_operations::add:
                 {
-                    operation_add(ptr, get_value("add", "value", false));
+                    operation_add(ptr, get_value(members.value, "add", "value", false));
                     break;
                 }
 
@@ -7229,13 +7329,13 @@ public:
                 case patch_operations::replace:
                 {
                     // the "path" location must exist - use at()
-                    result.at(ptr) = get_value("replace", "value", false);
+                    result.at(ptr) = get_value(members.value, "replace", "value", false);
                     break;
                 }
 
                 case patch_operations::move:
                 {
-                    const string_t& from_path = *get_value("move", "from", true).m_data.m_value.string;
+                    const string_t& from_path = *get_value(members.from, "move", "from", true).m_data.m_value.string;
                     json_pointer from_ptr(from_path);
 
                     if (JSON_HEDLEY_UNLIKELY(is_proper_prefix(from_ptr, ptr)))
@@ -7258,7 +7358,7 @@ public:
 
                 case patch_operations::copy:
                 {
-                    const string_t& from_path = *get_value("copy", "from", true).m_data.m_value.string;
+                    const string_t& from_path = *get_value(members.from, "copy", "from", true).m_data.m_value.string;
                     const json_pointer from_ptr(from_path);
 
                     // The copy is functionally identical to an "add"
@@ -7276,7 +7376,7 @@ public:
                     {
                         // check if "value" matches the one at "path"
                         // the "path" location must exist - use at()
-                        success = (result.at(ptr) == get_value("test", "value", false));
+                        success = (result.at(ptr) == get_value(members.value, "test", "value", false));
                     }
                     JSON_INTERNAL_CATCH (out_of_range&)
                     {
@@ -7368,13 +7468,18 @@ public:
     /// @brief append an operation @a op for @a path to @a result, with @a value if it is not null
     static void diff_operation(basic_json& result, const char* op, const string_t& path, const basic_json* value)
     {
+        // the keys are distinct and in the order of std::map; the members are
+        // added without looking them up
         basic_json entry(value_t::object);
-        entry["op"] = op;
-        entry["path"] = path;
+        auto& members = *entry.m_data.m_value.object;
+        auto hint = members.end();
+        json_pointer::add_flattened(members, hint, string_t("op"), string_t(op), detail::priority_tag<2> {});
+        json_pointer::add_flattened(members, hint, string_t("path"), path, detail::priority_tag<2> {});
         if (value != nullptr)
         {
-            entry["value"] = *value;
+            json_pointer::add_flattened(members, hint, string_t("value"), *value, detail::priority_tag<2> {});
         }
+        entry.set_parents();
         result.push_back(std::move(entry));
     }
 
@@ -7396,6 +7501,62 @@ public:
         diff_operation(result, "add", path, &value);
     }
 
+    /// @brief the values in the target of the members of a source object, for the
+    ///        objects of up to 16 members without allocating
+    ///
+    /// reserve() must be called with the number of values that are added.
+    class diff_target_values
+    {
+      public:
+        diff_target_values() noexcept
+            : inline_items() // NOLINT(readability-redundant-member-init)
+        {}
+        diff_target_values(const diff_target_values&) = delete;
+        diff_target_values& operator=(const diff_target_values&) = delete;
+        diff_target_values(diff_target_values&&) = delete;
+        diff_target_values& operator=(diff_target_values&&) = delete;
+        ~diff_target_values() = default;
+
+        void reserve(const std::size_t n)
+        {
+            if (n > inline_size)
+            {
+                large.resize(n);
+                items = large.data();
+            }
+        }
+
+        void push_back(const basic_json* value)
+        {
+            items[count++] = value;
+        }
+
+        std::size_t size() const noexcept
+        {
+            return count;
+        }
+
+        const basic_json* operator[](const std::size_t i) const noexcept
+        {
+            return items[i];
+        }
+
+      private:
+        static constexpr std::size_t inline_size = 16;
+        /// the values of a small object, left uninitialized: they are written
+        /// before they are read (the user-provided constructor keeps the
+        /// member initialization, which -Weffc++ asks for, from zeroing them)
+        struct uninitialized_items : std::array<const basic_json*, inline_size>
+        {
+            uninitialized_items() noexcept {} // NOLINT(modernize-use-equals-default,hicpp-use-equals-default,cppcoreguidelines-pro-type-member-init,hicpp-member-init): see above
+        };
+
+        std::size_t count = 0;
+        std::vector<const basic_json*> large{}; // NOLINT(readability-redundant-member-init)
+        uninitialized_items inline_items;
+        const basic_json** items = inline_items.data();
+    };
+
     /// @brief append "/" and the decimal representation of @a index to @a path
     static void diff_append_index(string_t& path, std::size_t index)
     {
@@ -7476,8 +7637,9 @@ public:
     operations that remove every member of @a source and add every member of
     @a target to @a result, and returns false.
     */
+    template<typename TargetValues>
     static bool diff_object_keys(basic_json& result, const basic_json& source, const basic_json& target,
-                                 const string_t& path, std::vector<const basic_json*>& target_values,
+                                 const string_t& path, TargetValues& target_values,
                                  std::vector<const_iterator>& added)
     {
         // first pass: one lookup per source key
@@ -7691,7 +7853,7 @@ public:
 
             case value_t::object:
             {
-                std::vector<const basic_json*> target_values;
+                diff_target_values target_values;
                 std::vector<const_iterator> added;
                 if (diff_object_keys(result, source, target, path, target_values, added))
                 {

include/nlohmann/json_view.hpp

@@ -723,12 +723,8 @@ class basic_json_view
     template<typename, bool> friend class basic_json_document;
     template<typename, bool> friend class basic_json_view;
     template<typename, typename> friend class detail::view::editor;
-    template<typename, typename> friend class detail::view::view_side;
     friend iterator;
 
-    /// whether the view is one of an editable document
-    static constexpr bool is_editable = Editable;
-
     basic_json_view(const document_data* d, const node* n) noexcept
         : m_doc(d), m_node(n)
     {}

tests/src/unit-binary_writer_sinks.cpp

@@ -492,3 +492,278 @@ TEST_CASE("binary_reserve_hint never over-reserves")
         CHECK(nlohmann::detail::binary_reserve_hint(json({{"a", 1}, {"b", 2}})) == 5);
     }
 }
+
+namespace
+{
+
+// numbers around every boundary of the integer encodings, and floats of each
+// kind (exact in float, needing double, special values)
+std::vector<json> boundary_numbers()
+{
+    std::vector<json> numbers;
+    for (int k = 0; k < 64; ++k)
+    {
+        const std::uint64_t p = std::uint64_t{1} << k;
+        for (const std::uint64_t d :
+                {
+                    p - 1, p, p + 1
+                })
+        {
+            numbers.emplace_back(d);
+            if (d <= static_cast<std::uint64_t>((std::numeric_limits<std::int64_t>::max)()))
+            {
+                numbers.emplace_back(static_cast<std::int64_t>(d));
+                numbers.emplace_back(-static_cast<std::int64_t>(d));
+                numbers.emplace_back(-static_cast<std::int64_t>(d) - 1);
+            }
+        }
+    }
+    numbers.emplace_back((std::numeric_limits<std::int64_t>::min)());
+    numbers.emplace_back((std::numeric_limits<std::uint64_t>::max)());
+    for (const double f :
+            {
+                0.0, -0.0, 1.0, -1.0, 0.5, 0.1, 3.25, 1e300, -1e-300, 3.4028234663852886e38, 1e39,
+                std::numeric_limits<double>::infinity(), -std::numeric_limits<double>::infinity(),
+                std::numeric_limits<double>::quiet_NaN()
+            })
+    {
+        numbers.emplace_back(f);
+    }
+    return numbers;
+}
+
+// the concatenation of the encodings of the elements of @a numbers[first, first + count)
+template<typename Single>
+std::vector<std::uint8_t> concat_singles(const std::vector<json>& numbers, const std::size_t first, const std::size_t count, Single single)
+{
+    std::vector<std::uint8_t> bytes;
+    for (std::size_t i = first; i < first + count; ++i)
+    {
+        const auto one = single(numbers[i]);
+        bytes.insert(bytes.end(), one.begin(), one.end());
+    }
+    return bytes;
+}
+
+} // namespace
+
+// Runs of numbers in arrays are encoded straight into space reserved from the
+// sink, by code that is separate from the encoding of a single value. Both have
+// to agree on every number: an array is its head and the encodings of its
+// elements, one after the other.
+TEST_CASE("runs of numbers in arrays are encoded as the numbers one by one")
+{
+    const std::vector<json> numbers = boundary_numbers();
+
+    // arrays of every length that a one-byte head covers, so that runs are as
+    // long as, and longer and shorter than, what fits one reservation
+    for (const std::size_t count :
+            {
+                1u, 2u, 3u, 5u, 14u, 15u, 23u
+            })
+    {
+        for (std::size_t first = 0; first + count <= numbers.size(); first += 7)
+        {
+            const json array(std::vector<json>(numbers.begin() + static_cast<std::ptrdiff_t>(first),
+                                               numbers.begin() + static_cast<std::ptrdiff_t>(first + count)));
+            CAPTURE(array.dump(-1, ' ', false, json::error_handler_t::replace))
+
+            std::vector<std::uint8_t> expected{static_cast<std::uint8_t>(0x80 + count)};
+            auto elements = concat_singles(numbers, first, count, [](const json & j)
+            {
+                return json::to_cbor(j);
+            });
+            expected.insert(expected.end(), elements.begin(), elements.end());
+            CHECK(json::to_cbor(array) == expected);
+
+            expected.assign(1, static_cast<std::uint8_t>(count <= 15 ? 0x90 + count : 0xDC));
+            if (count > 15)
+            {
+                expected.push_back(0);
+                expected.push_back(static_cast<std::uint8_t>(count));
+            }
+            elements = concat_singles(numbers, first, count, [](const json & j)
+            {
+                return json::to_msgpack(j);
+            });
+            expected.insert(expected.end(), elements.begin(), elements.end());
+            CHECK(json::to_msgpack(array) == expected);
+
+            for (const bool bjdata :
+                    {
+                        false, true
+                    })
+            {
+                expected.assign(1, static_cast<std::uint8_t>('['));
+                elements = concat_singles(numbers, first, count, [bjdata](const json & j)
+                {
+                    return bjdata ? json::to_bjdata(j) : json::to_ubjson(j);
+                });
+                expected.insert(expected.end(), elements.begin(), elements.end());
+                expected.push_back(static_cast<std::uint8_t>(']'));
+                CHECK((bjdata ? json::to_bjdata(array) : json::to_ubjson(array)) == expected);
+            }
+
+            // BON8 has no integers above the int64 range
+            bool representable = true;
+            for (std::size_t i = first; i < first + count; ++i)
+            {
+                representable = representable && bon8_representable(numbers[i]);
+            }
+            if (representable)
+            {
+                expected.assign(1, static_cast<std::uint8_t>(count <= 4 ? 0x80 + count : 0x85));
+                elements = concat_singles(numbers, first, count, [](const json & j)
+                {
+                    return json::to_bon8(j);
+                });
+                expected.insert(expected.end(), elements.begin(), elements.end());
+                if (count > 4)
+                {
+                    expected.push_back(0xFE);
+                }
+                CHECK(json::to_bon8(array) == expected);
+            }
+            else
+            {
+                CHECK_THROWS_AS(json::to_bon8(array), json::out_of_range&);
+            }
+        }
+    }
+}
+
+// An array that is written in more than one reservation (more numbers than fit
+// one), and arrays of an index with more digits, round trip for every format.
+TEST_CASE("large arrays and many digits of indices round trip")
+{
+    json array = json::array();
+    for (int i = 0; i < 12000; ++i)
+    {
+        switch (i % 4)
+        {
+            case 0:
+                array.push_back(i);
+                break;
+            case 1:
+                array.push_back(-i);
+                break;
+            case 2:
+                array.push_back(i * 0.25);
+                break;
+            default:
+                array.push_back(i % 8 == 3);
+                break;
+        }
+    }
+    json document = json::object();
+    document["numbers"] = array;
+    document["nested"] = json::array({array, "text", array});
+
+    CHECK(json::from_cbor(json::to_cbor(document)) == document);
+    CHECK(json::from_msgpack(json::to_msgpack(document)) == document);
+    CHECK(json::from_ubjson(json::to_ubjson(document)) == document);
+    CHECK(json::from_ubjson(json::to_ubjson(document, true, true)) == document);
+    CHECK(json::from_bjdata(json::to_bjdata(document)) == document);
+    CHECK(json::from_bson(json::to_bson(document)) == document);
+    CHECK(json::from_bon8(json::to_bon8(document)) == document);
+}
+
+// Strings and keys are written with their head in one reservation up to a
+// length; around that length, and with every kind of head, they must read back.
+TEST_CASE("strings of every head length round trip")
+{
+    json document = json::object();
+    json array = json::array();
+    for (const std::size_t length :
+            {
+                0u, 1u, 2u, 3u, 4u, 7u, 8u, 15u, 16u, 17u, 23u, 24u, 25u, 31u, 32u, 33u, 63u, 127u, 128u, 129u, 200u, 245u, 246u, 247u, 248u, 249u, 250u, 251u, 252u, 253u, 254u, 255u, 256u, 257u, 300u, 1000u, 65535u, 65536u, 70000u
+            })
+    {
+        const std::string text(length, 'x');
+        array.push_back(text);
+        document[std::string(length, 'k')] = text;
+    }
+    document["array"] = array;
+
+    CHECK(json::from_cbor(json::to_cbor(document)) == document);
+    CHECK(json::from_msgpack(json::to_msgpack(document)) == document);
+    CHECK(json::from_ubjson(json::to_ubjson(document)) == document);
+    CHECK(json::from_ubjson(json::to_ubjson(document, true, true)) == document);
+    CHECK(json::from_bjdata(json::to_bjdata(document)) == document);
+    CHECK(json::from_bson(json::to_bson(document)) == document);
+    CHECK(json::from_bon8(json::to_bon8(document)) == document);
+}
+
+#if !defined(JSON_NOEXCEPTION)
+// What a write that throws leaves behind in an output adapter is what was written
+// before it, whatever the adapter and however the adapter sink buffers. (Without
+// exceptions, the writes in CHECK_THROWS_AS do not run.)
+TEST_CASE("a write that throws leaves the same bytes in every kind of output")
+{
+    json array = json::array();
+    for (int i = 0; i < 700; ++i)
+    {
+        array.push_back(i * 1000);
+        array.push_back("text " + std::to_string(i));
+    }
+    array.push_back("bad \xFF byte");
+    json document = json::object();
+    document["first"] = array;
+    document["second"] = 1;
+
+    const auto strict = json::error_handler_t::strict;
+
+    SECTION("cbor")
+    {
+        std::vector<std::uint8_t> as_vector;
+        std::string as_string;
+        std::ostringstream as_stream;
+        CHECK_THROWS_AS(json::to_cbor(document, as_vector, strict), json::type_error&);
+        CHECK_THROWS_AS(json::to_cbor(document, as_string, strict), json::type_error&);
+        CHECK_THROWS_AS(json::to_cbor(document, as_stream, strict), json::type_error&);
+        CHECK(!as_vector.empty());
+        CHECK(std::string(reinterpret_cast<const char*>(as_vector.data()), as_vector.size()) == as_string); // NOLINT(cppcoreguidelines-pro-type-reinterpret-cast)
+        CHECK(as_stream.str() == as_string);
+    }
+
+    SECTION("msgpack")
+    {
+        std::vector<std::uint8_t> as_vector;
+        std::string as_string;
+        std::ostringstream as_stream;
+        CHECK_THROWS_AS(json::to_msgpack(document, as_vector, strict), json::type_error&);
+        CHECK_THROWS_AS(json::to_msgpack(document, as_string, strict), json::type_error&);
+        CHECK_THROWS_AS(json::to_msgpack(document, as_stream, strict), json::type_error&);
+        CHECK(!as_vector.empty());
+        CHECK(std::string(reinterpret_cast<const char*>(as_vector.data()), as_vector.size()) == as_string); // NOLINT(cppcoreguidelines-pro-type-reinterpret-cast)
+        CHECK(as_stream.str() == as_string);
+    }
+
+    SECTION("ubjson")
+    {
+        std::vector<std::uint8_t> as_vector;
+        std::string as_string;
+        std::ostringstream as_stream;
+        CHECK_THROWS_AS(json::to_ubjson(document, as_vector, false, false, strict), json::type_error&);
+        CHECK_THROWS_AS(json::to_ubjson(document, as_string, false, false, strict), json::type_error&);
+        CHECK_THROWS_AS(json::to_ubjson(document, as_stream, false, false, strict), json::type_error&);
+        CHECK(!as_vector.empty());
+        CHECK(std::string(reinterpret_cast<const char*>(as_vector.data()), as_vector.size()) == as_string); // NOLINT(cppcoreguidelines-pro-type-reinterpret-cast)
+        CHECK(as_stream.str() == as_string);
+    }
+
+    SECTION("bson")
+    {
+        // the size pass finds the string before anything is written
+        std::vector<std::uint8_t> as_vector;
+        std::string as_string;
+        std::ostringstream as_stream;
+        CHECK_THROWS_AS(json::to_bson(document, as_vector, strict), json::type_error&);
+        CHECK_THROWS_AS(json::to_bson(document, as_string, strict), json::type_error&);
+        CHECK_THROWS_AS(json::to_bson(document, as_stream, strict), json::type_error&);
+        CHECK(as_vector.empty());
+        CHECK(as_string.empty());
+        CHECK(as_stream.str().empty());
+    }
+}
+#endif

tests/src/unit-custom-number-types.cpp

@@ -285,6 +285,58 @@ TEST_CASE("custom number types: integers beyond 64 bits")
         CHECK(wide_json::to_bjdata(wide_json::array({uint_big, uint_big}), true, true) == expected_bjdata_array);
     }
 
+    SECTION("integers in arrays and objects")
+    {
+        // runs of numbers and scalar object members are encoded by loops of
+        // their own for the number types up to 64 bits; a wider number type
+        // takes the generic path, which checks the range of every integer
+        const wide_json run = wide_json::array({make_int(1), make_int(-3), make_uint(7), wide_json(1.5), int_big, make_int(5)});
+        const wide_json run_negative = wide_json::array({make_int(1), make_int(-3), int_big_negative});
+        const wide_json run_unsigned = wide_json::array({make_uint(1), make_uint(7), uint_big});
+        const wide_json members = wide_json::object({{"a", make_int(1)}, {"b", wide_json(1.5)}, {"c", int_big}});
+
+        const std::vector<wide_json> failing = {run, run_negative, run_unsigned, members, wrap(run), wrap(members)};
+        for (const auto& value : failing)
+        {
+            CHECK_THROWS_AS(_ = wide_json::to_cbor(value), out_of_range&);
+            CHECK_THROWS_AS(_ = wide_json::to_msgpack(value), out_of_range&);
+            CHECK_THROWS_AS(_ = wide_json::to_bson(wrap(value)), out_of_range&);
+            CHECK_THROWS_AS(_ = wide_json::to_bon8(value), out_of_range&);
+        }
+
+#if !defined(JSON_NOEXCEPTION)
+        // a failing write leaves no more than the elements before the failing one
+        // (without exceptions, the writes in CHECK_THROWS_AS do not run)
+        std::vector<std::uint8_t> out;
+        CHECK_THROWS_AS(wide_json::to_cbor(run, out), out_of_range&);
+        // (the head of the array, 1, -3, 7, and 1.5 as a single-precision float)
+        CHECK(out == std::vector<std::uint8_t>({0x86, 0x01, 0x22, 0x07, 0xFA, 0x3F, 0xC0, 0x00, 0x00}));
+        out.clear();
+        CHECK_THROWS_AS(wide_json::to_bon8(run, out), out_of_range&);
+        CHECK(!out.empty());
+#endif
+
+        // the same values that fit are written exactly as with 64-bit number types
+        const wide_json fits = wide_json::array({make_int(1), make_int(-3), make_uint(7), wide_json(1.5), make_int(min_int64), make_int(max_int64), make_int(200), make_int(-70000)});
+        const nlohmann::json fits64 = nlohmann::json::array({1, -3, 7, 1.5, (std::numeric_limits<std::int64_t>::min)(), (std::numeric_limits<std::int64_t>::max)(), 200, -70000});
+        CHECK(wide_json::to_cbor(fits) == nlohmann::json::to_cbor(fits64));
+        CHECK(wide_json::to_msgpack(fits) == nlohmann::json::to_msgpack(fits64));
+        CHECK(wide_json::to_ubjson(fits) == nlohmann::json::to_ubjson(fits64));
+        CHECK(wide_json::to_ubjson(fits, true, true) == nlohmann::json::to_ubjson(fits64, true, true));
+        CHECK(wide_json::to_bjdata(fits) == nlohmann::json::to_bjdata(fits64));
+        CHECK(wide_json::to_bjdata(fits, true, true) == nlohmann::json::to_bjdata(fits64, true, true));
+        CHECK(wide_json::to_bon8(fits) == nlohmann::json::to_bon8(fits64));
+        CHECK(wide_json::to_bson(wrap(fits)) == nlohmann::json::to_bson(nlohmann::json::object({{"v", fits64}})));
+
+        const wide_json fit_members = wide_json::object({{"a", make_int(1)}, {"b", make_int(-300)}, {"c", make_uint(max_uint64)}, {"d", wide_json(2.5)}});
+        const nlohmann::json fit_members64 = nlohmann::json::object({{"a", 1}, {"b", -300}, {"c", (std::numeric_limits<std::uint64_t>::max)()}, {"d", 2.5}});
+        CHECK(wide_json::to_cbor(fit_members) == nlohmann::json::to_cbor(fit_members64));
+        CHECK(wide_json::to_msgpack(fit_members) == nlohmann::json::to_msgpack(fit_members64));
+        CHECK(wide_json::to_ubjson(fit_members) == nlohmann::json::to_ubjson(fit_members64));
+        CHECK(wide_json::to_bjdata(fit_members) == nlohmann::json::to_bjdata(fit_members64));
+        CHECK(wide_json::to_bson(fit_members) == nlohmann::json::to_bson(fit_members64));
+    }
+
     SECTION("BJData ND-array")
     {
         // an ND-array element beyond 64 bits does not fit any dtype, so the

tests/src/unit-json_view_dump.cpp

@@ -345,35 +345,6 @@ TEST_CASE("json_view comparison")
         }
     }
 
-    SECTION("objects whose first keys differ or repeat")
-    {
-        const std::array<const char*, 17> texts =
-        {
-            {
-                R"({"a":1,"b":2})", R"({"a":1,"c":2})", R"({"b":1,"a":2})", R"({"a":1,"a":2,"b":3})", R"({"a":2,"b":3})",
-                R"({"a":1,"b":1,"a":2})", R"({"a":2,"b":1})", R"({"a":1,"a":2})", R"({"b":1})", R"({"a":1,"b":2,"c":3,"d":4,"e":5})",
-                R"({"a":1,"b":2,"c":3,"d":4,"f":5})", R"({"a":1,"b":2,"c":3,"a":4,"e":5})", R"({"a":4,"b":2,"c":3,"e":5})",
-                R"({"\u0061":1,"b":2})", R"({})", R"([{"a":1,"b":2}])", R"([{"a":1,"c":2}])"
-            }
-        };
-        for (const char* const x : texts)
-        {
-            for (const char* const y : texts)
-            {
-                CAPTURE(x)
-                CAPTURE(y)
-                const ordered_json_document dx = ordered_json_document::parse(x);
-                const ordered_json_document dy = ordered_json_document::parse(y);
-                const ordered_json_editable_document ex = ordered_json_editable_document::parse(x);
-                const ordered_json jx = ordered_json::parse(x);
-                const ordered_json jy = ordered_json::parse(y);
-                CHECK((dx.root() == dy.root()) == (jx == jy));
-                CHECK((dx.root() == jy) == (jx == jy));
-                CHECK((ex.root() == dy.root()) == (jx == jy));
-            }
-        }
-    }
-
     SECTION("numbers, duplicate keys, member order")
     {
         const auto same = [](const char* x, const char* y)
@@ -414,151 +385,6 @@ TEST_CASE("json_view comparison")
         CHECK(!(null_document.root() == discarded));
     }
 
-    SECTION("values that are the same text, or the same value written differently, at any depth")
-    {
-        // nested arrays and objects around depths where the walk over equal text gives up
-        for (const int depth :
-                {
-                    1, 5, 47, 48, 49, 50, 60, 300
-                })
-        {
-            CAPTURE(depth)
-            std::string head;
-            std::string tail;
-            for (int i = 0; i < depth; ++i)
-            {
-                head += i % 2 == 0 ? R"({"k":[1,"s",{"x":true},)" : R"([null,2.5,{"y":)";
-                tail.insert(0, i % 2 == 0 ? "]}" : "}]");
-            }
-            // the same text, one number changed, and one number written differently
-            const auto with_number = [&](const char* number)
-            {
-                std::string result = head;
-                result += number;
-                result += tail;
-                return result;
-            };
-            const std::string text = with_number("7");
-            const std::string changed = with_number("8");
-            const std::string spelled = with_number("7.0");
-            const json_document a = json_document::parse(text);
-            const json_document b = json_document::parse(text);
-            const json_document c = json_document::parse(changed);
-            const json_document d = json_document::parse(spelled);
-            CHECK(a.root() == b.root());
-            CHECK(a.root() != c.root());
-            CHECK(c.root() != a.root());
-            CHECK(a.root() == d.root());
-            CHECK(d.root() == a.root());
-            CHECK(a.root() == json::parse(text));
-            CHECK(a.root() != json::parse(changed));
-            CHECK(json::parse(spelled) == a.root());
-            const nlohmann::json_editable_document e = nlohmann::json_editable_document::parse(text);
-            CHECK(a.root() == e.root());
-            CHECK(e.root() == a.root());
-            CHECK(c.root() != e.root());
-            CHECK(e.root() == e.root());
-            const ordered_json_document oa = ordered_json_document::parse(text);
-            const ordered_json_document oc = ordered_json_document::parse(changed);
-            CHECK(oa.root() != oc.root());
-            CHECK(oa.root() == ordered_json::parse(text));
-            CHECK(oc.root() != ordered_json::parse(text));
-        }
-
-        // not the same text, but the same value
-        const auto same = [](const char* x, const char* y)
-        {
-            const json_document dx = json_document::parse(x);
-            const json_document dy = json_document::parse(y);
-            const bool equal = dx.root() == dy.root();
-            CHECK(equal == (json::parse(x) == json::parse(y)));
-            CHECK(equal == (dx.root() == json::parse(y)));
-            return equal;
-        };
-        CHECK(same(R"({"a":1,"a":2})", R"({"a":3,"a":2})"));
-        CHECK(same(R"({"a":1,"b":[1,{"c":2,"c":3}]})", R"({"b":[1,{"c":9,"c":3}],"a":1})"));
-        CHECK(!same(R"({"a":1,"a":2})", R"({"a":2,"a":3})"));
-        CHECK(same("[1.0,-0,0,1e2]", "[1.00,0,-0,100.0]"));
-        CHECK(same(R"(["\u00e9","\n"])", "[\"\xc3\xa9\",\"\\n\"]"));
-        CHECK(!same(R"({"a":[1,2,{"b":null}]})", R"({"a":[1,2,{"b":false}]})"));
-        CHECK(!same(R"({"a":[1,2,{"b":null}]})", R"({"a":[1,2,{"c":null}]})"));
-        CHECK(!same(R"({"a":[1,2,{"b":null}]})", R"({"a":[1,2,{"b":null,"c":1}]})"));
-
-        // edited documents: moved sequences, links, and values of new nodes
-        nlohmann::json_editable_document edited = nlohmann::json_editable_document::parse(R"({"a":[1,2,3],"b":{"c":"d"},"e":0.5})");
-        const json_document plain = json_document::parse(R"({"a":[1,2,3,4],"b":{"c":"x","y":null},"e":0.5,"z":[]})");
-        edited.push_back(edited.root()["a"], 4);
-        edited.set(edited.root()["b"], "c", "x");
-        edited.set(edited.root()["b"], "y", nullptr);
-        edited.set(edited.root(), "z", json::array());
-        CHECK(edited.root() == plain.root());
-        CHECK(plain.root() == edited.root());
-        edited.erase(edited.root()["a"], 0);
-        CHECK(edited.root() != plain.root());
-        CHECK(edited.root()["a"] == json::parse("[2,3,4]"));
-        const json_document tail = json_document::parse("[2,3,4]");
-        CHECK(edited.root()["a"] == tail.root());
-    }
-
-    SECTION("large documents, with a difference at any place")
-    {
-        const auto make = [](std::size_t count, std::size_t changed, bool swap_members)
-        {
-            std::string text = R"({"b":1,"list":[)";
-            for (std::size_t i = 0; i < count; ++i)
-            {
-                text += (i != 0 ? "," : "");
-                const std::string id = std::to_string(i == changed ? i + 1 : i);
-                if (swap_members && i % 3 == 0)
-                {
-                    text += R"({"v":[1.5,"s",{"k":)";
-                    text += id;
-                    text += R"(}],"id":)";
-                    text += id;
-                    text += "}";
-                }
-                else
-                {
-                    text += R"({"id":)";
-                    text += id;
-                    text += R"(,"v":[1.5,"s",{"k":)";
-                    text += id;
-                    text += "}]}";
-                }
-            }
-            return text + R"(],"a":{"y":2,"x":3}})";
-        };
-        const std::size_t count = 3000;
-        for (const std::size_t changed :
-                {
-                    count, std::size_t{0}, std::size_t{1}, count / 2, count - 1
-                })
-        {
-            for (const bool swap :
-                    {
-                        false, true
-                    })
-            {
-                CAPTURE(changed)
-                CAPTURE(swap)
-                const std::string text_a = make(count, count, false);
-                const std::string text_b = make(count, changed, swap);
-                const json_document a = json_document::parse(text_a);
-                const json_document b = json_document::parse(text_b);
-                const bool equal = changed == count;
-                CHECK((a.root() == b.root()) == equal);
-                CHECK((b.root() == a.root()) == equal);
-                CHECK((a.root() == json::parse(text_b)) == equal);
-                CHECK((json::parse(text_b) == a.root()) == equal);
-                CHECK((a.root()["list"] == b.root()["list"]) == equal);
-                const ordered_json_document oa = ordered_json_document::parse(text_a);
-                const ordered_json_document ob = ordered_json_document::parse(text_b);
-                CHECK((oa.root() == ob.root()) == (equal && !swap));
-                CHECK((oa.root() == ordered_json::parse(text_b)) == (equal && !swap));
-            }
-        }
-    }
-
     SECTION("deep nesting")
     {
         const std::string deep = std::string(100000, '[') + std::string(100000, ']');

tests/src/unit-json_view_image.cpp

@@ -682,66 +682,6 @@ TEST_CASE("json_view images: ownership")
 #if !defined(JSON_NOEXCEPTION)
 TEST_CASE("json_view images: errors")
 {
-#if !defined(JSON_NOEXCEPTION)
-    SECTION("float tokens: a changed byte is accepted if the text is still a number with the digits the node records")
-    {
-        for (const char* const token :
-                {
-                    "-12.375", "0.5", "123456.0", "-0.0", "9999999999999999999.5", "0.000000000000000000000001", "1.25e10", "5E-3"
-                })
-        {
-            CAPTURE(token)
-            const std::string text = std::string("[") + token + "]";
-            const json_document d = json_document::parse(text);
-            const std::vector<std::uint8_t> image = d.save();
-            const std::size_t at = text_at(image) + 1; // the first byte of the token
-            const std::uint16_t recorded = node_at(image, 1).extra;
-            CHECK(loaded_dump(image) == json::parse(text).dump());
-            for (std::size_t pos = 0; token[pos] != '\0'; ++pos)
-            {
-                for (unsigned int byte = 0; byte < 256; ++byte)
-                {
-                    std::vector<std::uint8_t> changed = image;
-                    changed[at + pos] = static_cast<std::uint8_t>(byte);
-                    std::string t = token;
-                    t[pos] = static_cast<char>(byte);
-                    // the digits of the integer part and of the fraction of the changed token
-                    std::size_t integer = 0;
-                    std::size_t fraction = 0;
-                    std::size_t k = (t[0] == '-') ? 1 : 0;
-                    for (; k < t.size() && t[k] >= '0' && t[k] <= '9'; ++k)
-                    {
-                        ++integer;
-                    }
-                    if (k < t.size() && t[k] == '.')
-                    {
-                        for (++k; k < t.size() && t[k] >= '0' && t[k] <= '9'; ++k)
-                        {
-                            ++fraction;
-                        }
-                    }
-                    // (as a JSON text, the token may have white space around it)
-                    const bool float_token = json::accept(t) && t.find_first_of(std::string(" \t\r\n\0", 5)) == std::string::npos && (t.find('.') != std::string::npos || t.find_first_of("eE") != std::string::npos)
-                                             && ((integer | (fraction << 8U)) == recorded);
-                    CAPTURE(pos)
-                    CAPTURE(byte)
-                    std::string result;
-                    try
-                    {
-                        result = loaded_dump(changed);
-                    }
-                    catch (const json::parse_error& e)
-                    {
-                        CHECK(std::string(e.what()) == check_failed);
-                        result = "rejected";
-                    }
-                    CHECK((result != "rejected") == float_token);
-                }
-            }
-        }
-    }
-#endif
-
     SECTION("a literal as the root: dump() after loading")
     {
         for (const char* text :

tests/src/unit-serialization.cpp

@@ -1580,3 +1580,94 @@ TEST_CASE("invalid UTF-8 at every position is handled like the reference")
         }
     }
 }
+
+TEST_CASE("short strings are written like the reference, as values and as keys")
+{
+    // strings of up to 16 characters (32 with vector instructions) that need no
+    // escaping are written as a whole (together with the quotes and, for a key,
+    // the colon); everything else, including a string that is merely short,
+    // takes the general path.
+    // Every length around the edges of the size classes of the fast path is
+    // combined with each kind of byte that must send a string to the general
+    // path, at each position, and compared with the reference for values and,
+    // via the value, for keys.
+    const std::vector<std::string> specials =
+    {
+        "\"", "\\", "\n", "\x1f", "\x7f", "\xc3\xa9", "\xe2\x82\xac", "\xf0\x9f\x98\x80", "\xff", "\x80", "\xc3"
+    };
+
+    for (std::size_t length = 0; length <= 36; ++length)
+    {
+        const std::string plain(length, 'k');
+        check_dump_against_reference(plain, false);
+
+        for (const auto& special : specials)
+        {
+            for (std::size_t pos = 0; pos <= length; ++pos)
+            {
+                std::string s = plain;
+                s.insert(pos, special);
+                check_dump_against_reference(s, false);
+
+                for (const bool ensure_ascii :
+                        {
+                            false, true
+                        })
+                {
+                    // a key and a value are written alike, in compact and
+                    // in pretty-printed form
+                    const json value = s;
+                    const json object = json::object({{s, 1}});
+                    for (const auto error_handler :
+                            {
+                                json::error_handler_t::replace, json::error_handler_t::keep
+                            })
+                    {
+                        const std::string quoted = value.dump(-1, ' ', ensure_ascii, error_handler);
+                        CHECK(object.dump(-1, ' ', ensure_ascii, error_handler) == "{" + quoted + ":1}");
+                        CHECK(object.dump(2, ' ', ensure_ascii, error_handler) == "{\n  " + quoted + ": 1\n}");
+                    }
+                }
+            }
+        }
+    }
+}
+
+TEST_CASE("runs of multi-byte characters are escaped like the reference under ensure_ascii")
+{
+    // a run of well-formed multi-byte characters is escaped straight into the
+    // write buffer; it has to flush that mid-run, stop at an ASCII character or
+    // an ill-formed sequence and leave those to the general path, and pick up
+    // behind what the general path has pending
+    const std::vector<std::string> characters =
+    {
+        "\xc2\x80", "\xdf\xbf", "\xe0\xa0\x80", "\xe2\x82\xac", "\xed\x9f\xbf", "\xef\xbf\xbf",
+        "\xf0\x90\x80\x80", "\xf0\x9f\x98\x80", "\xf4\x8f\xbf\xbf"
+    };
+    const std::vector<std::string> interruptions =
+    {
+        "", "a", "\n", "\"", "\x7f", "\xff", "\x80", "\xc3", "\xe2\x82", "\xf0\x9f\x98", "\xed\xa0\x80"
+    };
+
+    for (std::size_t shift = 0; shift <= 13; ++shift)
+    {
+        for (const auto& interruption : interruptions)
+        {
+            for (const std::size_t count :
+                    {
+                        std::size_t{1}, std::size_t{2}, std::size_t{85}, std::size_t{200}
+                    })
+            {
+                std::string s(shift, 'x');
+                for (std::size_t i = 0; i < count; ++i)
+                {
+                    s += characters[(i + shift) % characters.size()];
+                }
+                s += interruption;
+                s += characters[shift % characters.size()];
+                s += characters[(shift + 3) % characters.size()];
+                check_dump_against_reference(s, shift < 2);
+            }
+        }
+    }
+}

tests/src/unit-stl_helpers.cpp

@@ -78,7 +78,8 @@ TEST_CASE("detail::bit_cast")
 
     // to and from byte arrays
     const auto bytes = bit_cast<std::array<std::uint8_t, 4>>(std::uint32_t{0x01020304u});
-    CHECK(bit_cast<std::uint32_t>(bytes) == 0x01020304u);
+    // qualified: with a std::array argument, ADL would also find std::bit_cast in C++20
+    CHECK(nlohmann::detail::bit_cast<std::uint32_t>(bytes) == 0x01020304u);
 
     static_assert(noexcept(bit_cast<std::uint64_t>(1.0)), "bit_cast must be noexcept");
 }

tests/src/unit-stream-bulk-scan.cpp

@@ -48,14 +48,16 @@ namespace
 class chunked_streambuf : public std::streambuf
 {
   public:
-    chunked_streambuf(std::string s, std::size_t chunk_size)
-        : m_data(std::move(s)), m_chunk_size(chunk_size)
+    // The first @a unbuffered characters are served without a get area, the
+    // rest in get areas of @a chunk_size.
+    chunked_streambuf(std::string s, std::size_t chunk_size, std::size_t unbuffered = 0)
+        : m_data(std::move(s)), m_chunk_size(chunk_size), m_unbuffered(unbuffered)
     {}
 
   protected:
     int_type underflow() override
     {
-        if (m_chunk_size == 0)
+        if (m_chunk_size == 0 || m_pos < m_unbuffered)
         {
             return m_pos < m_data.size() ? traits_type::to_int_type(m_data[m_pos]) : traits_type::eof();
         }
@@ -77,7 +79,7 @@ class chunked_streambuf : public std::streambuf
 
     int_type uflow() override
     {
-        if (m_chunk_size == 0)
+        if (m_chunk_size == 0 || m_pos < m_unbuffered)
         {
             return m_pos < m_data.size() ? traits_type::to_int_type(m_data[m_pos++]) : traits_type::eof();
         }
@@ -88,6 +90,7 @@ class chunked_streambuf : public std::streambuf
     std::string m_data;
     std::size_t m_pos = 0;
     std::size_t m_chunk_size;
+    std::size_t m_unbuffered;
 };
 
 // how a test reads from the stream
@@ -101,9 +104,9 @@ enum class read_mode
 // Read @a input from a chunked_streambuf and describe everything observable:
 // the values or the exception, the stream state after each value, and the
 // characters left in the stream.
-std::string observe(const std::string& input, std::size_t chunk_size, read_mode mode)
+std::string observe(const std::string& input, std::size_t chunk_size, read_mode mode, std::size_t unbuffered = 0)
 {
-    chunked_streambuf buf(input, chunk_size);
+    chunked_streambuf buf(input, chunk_size, unbuffered);
     std::istream is(&buf);
     std::string result;
     for (int i = 0; i < 4; ++i)
@@ -212,6 +215,31 @@ TEST_CASE("parsing from a std::istream scans the get area in bulk")
         }
     }
 
+    SECTION("results do not depend on a get area that appears late")
+    {
+        // the first characters come without a get area, then the streambuf
+        // starts to use one: the token that is read at that point is complete
+        for (const auto& input : corpus())
+        {
+            const std::string expected = observe(input, 0, read_mode::extract);
+            for (const std::size_t unbuffered :
+                    {
+                        1u, 2u, 3u, 5u, 8u, 13u, 21u
+                    })
+            {
+                for (const std::size_t chunk_size :
+                        {
+                            1u, 2u, 3u, 7u, 64u
+                        })
+                {
+                    CAPTURE(input)
+                    CAPTURE(unbuffered)
+                    CAPTURE(chunk_size)
+                    CHECK(observe(input, chunk_size, read_mode::extract, unbuffered) == expected);
+                }
+            }
+        }
+    }
 #endif // !defined(JSON_NOEXCEPTION)
 
     SECTION("values are the same as when parsing from a string")

tools/generate_natvis/generate_natvis.py

@@ -139,9 +139,12 @@ if __name__ == '__main__':
 
     namespaces += [f'{namespaces[0]}::{ns}' for ns in inline_namespaces]
 
+    # The template renders a local XML file, not a web response, and autoescape is enabled.
+    # nosemgrep: python.flask.security.xss.audit.direct-use-of-jinja2.direct-use-of-jinja2
     env = jinja2.Environment(loader=jinja2.FileSystemLoader(searchpath=sys.path[0]), autoescape=True, trim_blocks=True,
                                                             lstrip_blocks=True, keep_trailing_newline=True)
     template = env.get_template('nlohmann_json.natvis.j2')
+    # nosemgrep: python.flask.security.xss.audit.direct-use-of-jinja2.direct-use-of-jinja2
     natvis = template.render(namespaces=namespaces)
 
     output = args.output or os.path.join(args.repo_root, 'nlohmann_json.natvis')