AppleClang's libc++ does not support std::from_chars for floating-point
types. Add a compile-time detection macro (CHAISCRIPT_HAS_FLOAT_FROM_CHARS)
and fall back to strtod/strtof/strtold on libc++ platforms. Centralize the
fallback in parse_num and route bootstrap.hpp and json.hpp through it to
keep the workaround in one place.
Requested by @lefticus in PR #669 review.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The custom parse_num floating-point implementation produced slightly
different results from the C++ compiler's own literal parsing (e.g.
1.1e-4 was off by a few ULPs). Replace it with std::from_chars which
is locale-independent and matches compiler precision exactly. Also
migrate integer parsing in buildInt from std::stoll/stoull to
std::from_chars, fix the JSON number parser to pass the full numeric
string to from_chars instead of splitting mantissa/exponent and using
std::pow, and update parse_string to use from_chars for arithmetic types.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* Fix#146: Add configuration options to selectively disable built-in functions
Add new Options enum values (No_Stdlib, No_IO, No_Prelude, No_JSON) that
allow users to control which parts of the standard library are registered.
Std_Lib::library() now accepts an options vector, and the ChaiScript
convenience class forwards its options to the library builder. This enables
use cases where only custom functions should be exposed to script users.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* Address review: split Options into Options and Library_Options enums
Separate system-level options (No_Load_Modules, Load_Modules, No_External_Scripts,
External_Scripts) from library-level options (No_Stdlib, No_IO, No_Prelude, No_JSON)
into two distinct enum types. Add Library_Options as a parameter to the ChaiScript
constructor. Update tests to demonstrate both ChaiScript_Basic (explicit Std_Lib::library
call) and ChaiScript (library options via constructor parameter) usage.
Requested by @lefticus in PR #642 review.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* Add cheatsheet documentation for Options and Library_Options
Documents the two-enum configuration system: Options (engine-level:
load_module, use, eval_file) and Library_Options (stdlib-level:
No_Stdlib, No_IO, No_Prelude, No_JSON), with usage examples for
both ChaiScript and ChaiScript_Basic constructors.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: leftibot <leftibot@users.noreply.github.com>
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Several STL bootstrap functions unconditionally instantiated copy-dependent
operations (copy constructor, assignment, push_back/push_front by const ref,
insert_at, and resize with fill value), causing compilation failures when
registering containers of move-only types like std::unique_ptr. Guard these
operations with if constexpr(std::is_copy_constructible_v<value_type>) so they
are only compiled when the element type supports copying.
Co-authored-by: leftibot <leftibot@users.noreply.github.com>
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Co-authored-by: Jason Turner <jason@emptycrate.com>
The function_less_than comparator used by std::stable_sort violated the
strict-weak ordering requirement in two ways: (1) functions with different
arities but matching overlapping parameters were treated as equivalent,
breaking transitivity, and (2) the dynamic_object_type_name comparison
silently fell through when one side had an empty type name. Fixed by
ordering by arity when overlapping parameters match, and imposing a total
order on dynamic object type names.
Co-authored-by: leftibot <leftibot@users.noreply.github.com>
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Co-authored-by: Jason Turner <jason@emptycrate.com>
* Fix#655: Join async threads before engine destruction to prevent heap-use-after-free
Issues #632 and #636 (PRs #651 and #653) both stem from the same root cause: async
threads spawned via async() can outlive the Dispatch_Engine, accessing shared state
(global objects map, type maps) after it has been destroyed. The fix moves async()
registration from the stdlib module into ChaiScript_Basic, where spawned threads are
tracked via Dispatch_Engine. The engine's destructor now joins all outstanding async
threads before destroying shared data structures.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* Address review: follow rule of 5, explicitly default move operations
Requested by @lefticus in PR #656 review.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: leftibot <leftibot@users.noreply.github.com>
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Previously, to_<type> only accepted a string or the same type (identity),
and <type>() constructors only accepted Boxed_Number. This left gaps such as
to_int(char), to_char(int), char(string), int(string), and to_double(char).
Add two overloads to bootstrap_pod_type: to_<name>(Boxed_Number) enables
cross-type numeric conversions (e.g. to_int('A') → 65, to_char(65) → 'A'),
and <name>(string) via parse_string enables construction from strings
(e.g. char("A") → 'A', int("65") → 65, double("3.5") → 3.5).
Co-authored-by: leftibot <leftibot@users.noreply.github.com>
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* Fix#405: push_back() on script-created vector has no effect with vector_conversion
When both vector_type<T> and vector_conversion<T> are registered, the
dispatch scoring treats all parameter mismatches equally. This causes
the C++ push_back (for the converted type) to be selected over the
built-in one when both have the same number of type differences. The
converted function operates on a temporary copy of the vector, silently
discarding the mutation. The fix deprioritizes overloads that require
type conversion on the first parameter (the object/receiver), ensuring
functions matching the receiver type exactly are always tried first.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* Address review: remove issue references from comments, add round-trip conversion tests
Requested by @lefticus in PR #663 review.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: leftibot <leftibot@users.noreply.github.com>
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* Fix#17: Add const local variable support to ChaiScript
Adds `const var`, `const auto`, and `const` as variable declaration
syntax that creates immutable local variables. A const_override flag
on Boxed_Value enables script-level constness without changing the
C++ type system integration. The parser, optimizer, and evaluator
are extended with Const_Var_Decl and Const_Assign_Decl AST nodes
that mirror their non-const counterparts but mark the value as const
after initialization.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* Address review: remove const_override, set const flag directly on Type_Info
Replace the m_const_override bool on Boxed_Value::Data with a
Type_Info::make_const() method that sets the const bit in m_flags
directly. This ensures constness is visible everywhere consistently,
including places that check get_type_info().is_const() directly.
Requested by @lefticus in PR #643 review.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: leftibot <leftibot@users.noreply.github.com>
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* Fix#201: Add class inheritance support with Derived : Base syntax
Classes can now inherit methods and attributes from a base class using
C++-style syntax: `class Derived : Base { ... }`. Base class methods and
attributes are automatically available on derived objects. Derived classes
can override base methods by defining a method with the same name.
Inheritance relationships are tracked to support proper type matching
in the dispatch system.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* Address review: use implicit derived-to-base matching instead of copying base class functions
Instead of copying all base class methods/attributes into derived classes,
make the type matching system recognize inheritance relationships. Base class
methods now naturally match derived objects through dynamic_object_typename_match,
and dispatch ordering ensures derived overrides are preferred over base methods.
This is simpler (net -25 lines) and avoids duplicating function registrations.
Requested by @lefticus in PR #641 review.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* Add tests for passing derived objects to functions expecting Base
Tests cover: free functions calling base methods on derived objects,
polymorphic dispatch through containers, base attribute access on
derived objects, and multi-level inheritance (GrandChild : Derived : Base).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* Add typed parameter tests for class inheritance
Use typed function signatures (e.g., `def call_do_something(Base obj)`)
instead of untyped parameters to test that derived objects are accepted
by functions expecting a base type, with correct polymorphic dispatch.
Requested by @lefticus in PR #641 review.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: leftibot <leftibot@users.noreply.github.com>
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
I initially tried to use the existing .clang-format file,
but it does not match the code style (at least with clang-format 11)
and the formatting is not consistent across files.
Therefore, I decided to rewrite the .clang-format with some personal
preferences.
Used command
find . -iname "*.hpp" -o -iname "*.cpp" | xargs clang-format -i -style=file
is_pod_v was deprecated in C++20, is_pod_v can be
replaced with is_trivial_v && is_standard_layout_v.
I don't see any benefit from is_standard_layout_v,
but I could have missed something.
The underlying pair that is dereferenced from the iterator has always `const` qualified `first` member (key type). Therefore, an unnecessary temporary was created and bounded to the const ref to the pair. This could be also fixed with `for (const auto &p : from_map)`.
- code (on MSVC) was asserting due to trying to dereference invalid
pointers (dereferencing the end iterator, even if only to get its
address!).
- when a Function_Params is constructed with an empty vector, you
can't return the address of the vec.front() -- instead we use
nullptr for the m_begin and m_end pointers.
