Train 26: #596, #581

CMakeLists.txt

@@ -378,6 +378,7 @@ if (JINJA2CPP_BUILD_TESTS)
     set(JINJA2CPP_TEST_SOURCES
         test/autoescape_test.cpp
         test/basic_tests.cpp
+        test/big_int_test.cpp
         test/containers_api_test.cpp
         test/errors_test.cpp
         test/expressions_test.cpp

docs/api-2.0.md

@@ -206,6 +206,7 @@ the migration notes fix it), **fix** (a defect; behaviour changes, no rename),
 | `Value(T&&)` for unsigned, `long long`, `short`, `char`, `size_t` | *probe:* stores `bool` at C++14 (`Value(5u).data().index() == 1`), `int64_t` at C++17, and `Value(size_t(5))` does not compile at C++17 | constructors for every integral type storing `int64_t`, as `Reflect` already does; `Value(char)` deleted (ambiguous between a number and a one-letter string) | fix (task 0067) |
 | `Value(nullptr)` | *probe:* picks `Value(const char*)` and throws `logic_error` from `std::string(nullptr)` | `Value(std::nullptr_t)` makes `None` | fix (0067) |
 | `Value(ValuesList&&) noexcept`, `Value(ValuesMap&&) noexcept` | allocate, so `bad_alloc` becomes `std::terminate` | drop `noexcept` there | fix |
+| `Value(uint64_t)` above `INT64_MAX` | cannot be held: wrapped to a negative number in 1.x, `std::overflow_error` since 0172 | `jinja2::BigInt` (`big_int.h`): an integer of any size as canonical decimal text, `TryParse(std::string_view, BigInt&) noexcept`, `Sign`, `ToInt64`/`ToUInt64`, comparisons by value; the last `Value` alternative (`sizeof(Value)` unchanged), with `IsBigInt()`, `AsBigInt()`, `bool AsBigInt(const BigInt*&)`; the integral constructor, `Reflect` and the JSON bindings store it for a value outside `int64_t`. `IsInt()` stays `int64_t` only. Until 0270 a `BigInt` that enters a template fails the render (a top-level render parameter fails it even unread), and `ToString()` prints one only at top level, not inside a list or map; arithmetic on it needs a backend (0271, plan docs/design/0178-big-integer-plan.md) | add (0269); an exhaustive `std::visit` over `GetData()` must handle the new alternative |
 | `IsEqual()` next to `operator==` | redundant | keep both; not worth a break | — |
 | `ValuesMap : std::unordered_map` | hash order | ordered map (task 0043) | break, done in 0043 |
 | free `AsString`/`AsWString` overloads for `std::string`, `std::wstring` and the views | encoding conversion; the `std::string` one returns a `const` value, defeating moves | `ToString`/`ToWString` returning `std::string`/`std::wstring` | alias |
@@ -321,13 +322,13 @@ class Value
 public:
     using ValueData = variant<NoneValue, bool, std::string, std::wstring, string_view, wstring_view,
                               int64_t, double, RecWrapper<ValuesList>, RecWrapper<ValuesMap>,
-                              GenericList, GenericMap, RecWrapper<UserCallable>, Markup, WMarkup>;
+                              GenericList, GenericMap, RecWrapper<UserCallable>, Markup, WMarkup, BigInt>;
 
     Value();
     Value(std::nullptr_t) noexcept;                         // None
     template<class I, std::enable_if_t<detail::IsStoredAsInt<I>::value>* = nullptr>
-    Value(I val) noexcept(!detail::CanExceedInt64<I>)       // above INT64_MAX: std::overflow_error (0172)
-        : m_data(detail::ToStoredInt(val)) {}
+    Value(I val) noexcept(!detail::CanExceedInt64<I>)       // outside int64_t: a BigInt (0269)
+        : m_data(detail::ToStoredInt<ValueData>(val)) {}
     Value(char) = delete;                                   // number or one-letter string? say which
     // ... string, double, list, map, callable constructors as today
 
@@ -341,6 +342,7 @@ public:
     bool IsList() const noexcept;      // ValuesList or GenericList
     bool IsMap() const noexcept;       // ValuesMap or GenericMap
     bool IsCallable() const noexcept;
+    bool IsBigInt() const noexcept;    // an integer outside int64_t
 
     const std::string& AsString() const;                // throws bad_variant_access
     std::string& AsString();                           // a held Markup becomes a plain string first
@@ -493,6 +495,14 @@ inside an `if` statement or an inline if, where it fails when it runs (docs/task
 1.x let a `UserCallable` passed in the render parameters or the globals act as a filter or
 test, found when it ran; `Settings::userCallablesAsFilters = true` keeps that behaviour.
 
+### 5.7 C++-only filters render, `toxml` is gone (release note)
+
+In 1.x `camelize`, `underscorize`, `escapecpp`, `toxml` and `toyaml` were registered but
+rendered an empty string. In 2.0 the first four render the output that
+[cpp-only-filters.md](cpp-only-filters.md) describes, and `toxml` is removed: a template that
+uses it fails `Template::Load` with `ErrorCode::UnknownFilter` (docs/tasks/0233,
+docs/release-2.0.md D16). Templates that leaned on the empty output drop the filter.
+
 ## 6. Release options
 
 | Option | What ships | Users get | Cost |

docs/cpp-only-filters.md

@@ -0,0 +1,85 @@
+# C++-only filters
+
+Jinja2C++ adds a few filters that Python Jinja2 does not have. A template that uses them
+does not load in Jinja2 (`No filter named ...`). Since 2.0 each has the output described
+here (docs/tasks/0233); before, `camelize`, `underscorize`, `escapecpp`, `toxml` and
+`toyaml` were registered but rendered an empty string.
+
+| Filter | Output |
+|---|---|
+| `applymacro(macro=name, ...)` | The value passed to the macro `name`, with the other arguments |
+| `camelize(upper_first=true)` | An identifier in CamelCase |
+| `underscorize` | An identifier in snake_case |
+| `escapecpp` | The text escaped for a C++ string literal |
+| `toyaml` | The value as a block-style YAML document |
+
+`toxml` was removed in 2.0: JSON-like data has no single XML mapping, so it would be a format
+of our own to define and support (docs/release-2.0.md, D16). A template that uses it fails
+to load with `ErrorCode::UnknownFilter`; `xmlattr` covers attribute lists, and a macro
+can write any XML shape.
+
+## `camelize` and `underscorize`
+
+Both take the value's `str()` and split it into words: runs of letters and digits (Unicode
+`\w` without `_`), split again before an uppercase letter that follows a lowercase letter
+or a digit (`fooBar`, `Vector3D`) and before the last of a run of uppercase letters that a
+lowercase one follows (`HTTPServer` is `HTTP` and `Server`). Everything between words
+(`_`, `-`, spaces, punctuation) is dropped.
+
+- `camelize` gives each word a titlecase first letter and lowercase others and joins them:
+  `device_type`, `device-type` and `DEVICE_TYPE` are `DeviceType`, `HTTPServer` is
+  `HttpServer`. With `upper_first=false` the first word is all lowercase: `deviceType`,
+  `httpServer`.
+- `underscorize` joins the words in lowercase with `_`: `DeviceType` is `device_type`,
+  `getHTTPResponseCode` is `get_http_response_code`, `__init__` is `init`.
+
+Case mapping is Unicode's, as in `upper`/`lower`: `straße_name|camelize` is `StraßeName`.
+
+## `escapecpp`
+
+The value's `str()` as the contents of a C++ string literal, without the quotes: `\` and
+`"` get a backslash, `\n`, `\r`, `\t`, `\a`, `\b`, `\f` and `\v` their escapes, and other
+control characters (below U+0020, and U+007F) three octal digits (`\001`), which cannot run
+into a digit that follows the way `\x` does. Other characters, `'` and non-ASCII ones
+included, are kept, so a UTF-8 template gives a literal for a UTF-8 source file.
+
+```
+const char* text = "{{ message|escapecpp }}";
+```
+
+## `toyaml`
+
+A YAML document that YAML 1.1 and 1.2 loaders (PyYAML, yaml-cpp, ...) read back as the same
+data, with no trailing newline:
+
+- `none` is `null`, booleans `true`/`false`, integers in decimal,
+  floats in Python's shortest form with a `.` always present (`1.0e+20`), `.nan`, `.inf`
+  and `-.inf`.
+- A mapping is `key: value` lines in the mapping's order; a list is `- item` lines; a
+  `dictsort` or `items` pair is a two-item list. Nested collections are indented by two
+  spaces, a collection inside a list starts on the item's line, and an empty one is `[]`
+  or `{}`.
+- A string (a key too) is written plain only when it cannot be read as anything else: it
+  starts with a letter, `_`, `/` or a printable non-ASCII character, holds only those,
+  digits, spaces and `-.()+`, does not end with a space and is not a null or boolean word
+  (`y`, `n`, `yes`, `no`, `on`, `off`, `true`, `false`, `null`, in any case). Any other
+  string is double-quoted with YAML escapes (`\"`, `\\`, `\n`, `\t`, `\xHH`, `\uHHHH`, ...).
+- A key longer than 255 bytes is written as an explicit `? key` with its value on a `:`
+  line below, since YAML limits an implicit key to 1024 characters.
+- Undefined values, `namespace()`, `cycler()`, `joiner()` and macros raise an error
+  ("Object of type Undefined is not YAML serializable"), as `tojson` does.
+
+```
+{{ {'name': 'demo', 'ports': [80, 443], 'tls': {'enabled': true}}|toyaml }}
+```
+
+```yaml
+name: demo
+ports:
+  - 80
+  - 443
+tls:
+  enabled: true
+```
+
+The result is an ordinary string: under autoescape it is HTML-escaped like any other.

docs/parity.md

@@ -4,7 +4,7 @@ How far Jinja2C++ is from Python [Jinja2](https://jinja.palletsprojects.com/) 3.
 area, and which task in `docs/tasks/` closes each gap.
 
 Every statement marked with a case id is backed by the differential corpus in
-`test/parity/` (1938 templates rendered by both engines, see
+`test/parity/` (1939 templates rendered by both engines, see
 [test/parity/README.md](../test/parity/README.md)); `ctest -R parity` re-checks all of
 them. Statements in the last section (API level) are read from the headers and are not
 corpus-checked yet.
@@ -21,7 +21,7 @@ Snapshot of `python3 test/parity/generate.py --report` (Jinja2 3.1.6, Oct 2026):
 | autoescape | 124 | 122 | 1 | 1 | 0 | 0 | 0 | 0 | 0210, 0216 |
 | custom | 28 | 28 | 0 | 0 | 0 | 0 | 0 | 0 |  |
 | errors | 97 | 93 | 0 | 2 | 2 | 0 | 0 | 0 | 0036, 0213 |
-| filters | 308 | 302 | 4 | 2 | 0 | 0 | 0 | 0 | 0101, 0178, 0203, 0215 |
+| filters | 309 | 303 | 4 | 2 | 0 | 0 | 0 | 0 | 0101, 0178, 0203, 0215 |
 | fuzz | 32 | 32 | 0 | 0 | 0 | 0 | 0 | 0 |  |
 | globals | 74 | 73 | 1 | 0 | 0 | 0 | 0 | 0 | 0212 |
 | i18n | 54 | 54 | 0 | 0 | 0 | 0 | 0 | 0 |  |
@@ -37,7 +37,7 @@ Snapshot of `python3 test/parity/generate.py --report` (Jinja2 3.1.6, Oct 2026):
 | tests | 58 | 57 | 0 | 1 | 0 | 0 | 0 | 0 | 0101 |
 | undefined | 106 | 106 | 0 | 0 | 0 | 0 | 0 | 0 |  |
 | whitespace | 54 | 54 | 0 | 0 | 0 | 0 | 0 | 0 |  |
-| **total** | **1938** | **1894** | **25** | **15** | **4** | **0** | **0** | **0** | |
+| **total** | **1939** | **1895** | **25** | **15** | **4** | **0** | **0** | **0** | |
 
 *output*: both render, text differs. *rejects*: C++ errors on a valid template.
 *accepts*: C++ renders a template Jinja2 rejects. *unsupported*: needs an Environment
@@ -255,8 +255,10 @@ storing a container in itself raises instead of printing `[...]` (deliberate, 00
 | `xmlattr` | ✅ | `xmlattr*` | |
 | Unknown filter is an error; in a branch never taken it is not | ✅ | `unknown_filter*` | |
 
-C++-only filters (`camelize`, `underscorize`, `escapecpp`, `toxml`, `toyaml`,
-`applymacro`) and the `startsWith` test are deliberate extensions; Jinja2 rejects them.
+C++-only filters (`camelize`, `underscorize`, `escapecpp`, `toyaml`, `applymacro`; see
+[cpp-only-filters.md](cpp-only-filters.md)) and the `startsWith` test are deliberate
+extensions; Jinja2 rejects them. `toxml` was removed in 2.0 and is an unknown filter, as in
+Jinja2 (0233).
 
 ## Statements (`statements`)
 

docs/tasks/0233-cpp-only-filter-stubs.md

@@ -1,9 +1,9 @@
 ---
-status: open
+status: done
 priority: medium
 area: api
 issues: ["https://github.com/jinja2cpp/Jinja2Cpp/issues/41"]
-touches: [src/filter_factories.cpp, src/string_converter_filter.cpp, src/serialize_filters.cpp]
+touches: [src/filter_factories.cpp, src/string_converter_filter.cpp, src/serialize_filters.cpp, src/filters.h, test/filters_test.cpp, docs/cpp-only-filters.md]
 shares: [docs/parity.md]
 ---
 # C++-only filters `camelize`, `underscorize`, `escapecpp`, `toxml`, `toyaml` do nothing
@@ -29,3 +29,9 @@ migration notes (0077) say which.
 
 **Release decision (Ruslan 2026-10-18, D16 in docs/release-2.0.md).** Implement `camelize`,
 `underscorize`, `escapecpp` and `toyaml` with documented output; remove `toxml`.
+
+**Outcome.** `camelize(upper_first=true)`, `underscorize`,
+`escapecpp` and `toyaml` render the output documented in `docs/cpp-only-filters.md`, with
+rows in `test/filters_test.cpp`; `toxml` is removed and fails Load with `UnknownFilter`
+(corpus case `filters.unknown_filter_toxml`). Release note: `docs/api-2.0.md` 5.7, for
+0077's `MIGRATION.md`. PR #581.

docs/tasks/0269-public-big-int-value.md

@@ -1,5 +1,5 @@
 ---
-status: open
+status: done
 priority: high
 area: api
 depends: [0172, 0178]
@@ -23,3 +23,5 @@ that enters a template raises OverflowError. Coordinate with 0250's phase 0253,
 rows), abidiff shows `sizeof(Value)` unchanged, and docs/api-2.0.md documents `BigInt`.
 
 Plan: [docs/design/0178-big-integer-plan.md](../design/0178-big-integer-plan.md).
+
+Done in #596.

docs/tasks/README.md

@@ -300,7 +300,7 @@ files go under `touches` (0012 and 0024 each rewrite about a hundred rows).
 | [0230](0230-boost-free-core.md) | A Boost-free core | build | medium | open |
 | [0231](0231-robin-hood-successor.md) | robin_hood's upstream is archived | perf | high | open |
 | [0232](0232-dependency-pin-drift.md) | Dependency pins disagree across modes and some are mutable | build | low | open |
-| [0233](0233-cpp-only-filter-stubs.md) | C++-only filters `camelize`, `underscorize`, `escapecpp`, `toxml`, `toyaml` do nothing | api | medium | open |
+| [0233](0233-cpp-only-filter-stubs.md) | C++-only filters `camelize`, `underscorize`, `escapecpp`, `toxml`, `toyaml` do nothing | api | medium | done |
 | [0234](0234-github-tracker-cleanup.md) | Close GitHub issues that master already fixes, before 2.0 | release | medium | open |
 | [0235](0235-parity-feature-rows-drift.md) | `docs/parity.md` feature rows contradict master | parity | medium | open |
 | [0236](0236-meta-find-undeclared-variables.md) | Template introspection: `meta.find_undeclared_variables`, `Template.module` | api | low | open |
@@ -334,7 +334,7 @@ files go under `touches` (0012 and 0024 each rewrite about a hundred rows).
 | [0266](0266-confined-filesystem.md) | `ConfinedFileSystem`: a loader that never leaves its root | api | medium | open |
 | [0267](0267-untrusted-mode-fuzzing.md) | Fuzz targets for templates written by untrusted users | robustness | high | open |
 | [0268](0268-untrusted-mode-followups.md) | Untrusted mode: additive follow-ups after the freeze | api | low | open |
-| [0269](0269-public-big-int-value.md) | Public `BigInt` value (0178 phase 1, G1) | api | high | open |
+| [0269](0269-public-big-int-value.md) | Public `BigInt` value (0178 phase 1, G1) | api | high | done |
 | [0270](0270-big-int-core-text-form.md) | Big integers in the core, text form (0178 phase 2) | parity | medium | open |
 | [0271](0271-big-int-backend-interface.md) | Big-integer backend interface and overflow promotion (0178 phase 3) | api | medium | open |
 | [0272](0272-big-int-boost-backend.md) | Boost.Multiprecision backend and `JINJA2CPP_WITH_BIGINT` (0178 phase 4) | build | medium | open |