feat!: merge v6.0.0 (#220)

build!: migrate python bindings from pybind11 to nanobind (#211)

build: migrate python bindings from pybind11 to nanobind


nanobind is a from-scratch rewrite of pybind11 by the same author, targeting
C++17 and trading pybind11’s breadth for a much leaner call path. Measured on
this repository’s own bindings, it is ~2.7x faster on a geometric mean of call
shapes (up to ~6x on method calls, ~9x on overload dispatch), compiles ~3x
faster and produces a ~33% smaller extension module.
The bindings themselves translate almost mechanically. The parts that do not:

pybind11/operators.h has no nanobind counterpart, so each self OP self
expression becomes an explicit lambda. In-place operators return a reference
under rv_policy::none so that a += b keeps returning a itself.
init(factory) becomes a placement-new __init__.
make_iterator now takes a scope and a name for the iterator type.
arg_v(name, value, repr) becomes arg(name).sig(repr) = value.
handle::cast<T>() becomes the free function nanobind::cast<T>(handle).
Path.get_absolute_path is split into two overloads. A Path default
argument makes the type reference-cycle through its own function records,
which nanobind reports as a leak at interpreter shutdown; the overloads also
defer Path::Current() to call time, matching the C++ default.

nanobind is fetched with FetchContent, as magic_enum already is, so no
development image change is needed. It is pinned to v2.9.2: v3.x requires
Python 3.10 and this project still ships 3.9 wheels. Stubs now come from
nanobind’s own stubgen rather than pybind11-stubgen.
One user-visible behaviour change: passing a string that is not exactly one
character where a C++ char is expected (String.char("")) now raises
TypeError rather than ValueError.
Co-Authored-By: Claude Opus 5 (1M context) noreply@anthropic.com

fix: apply compact assertions to the nanobind runtime, and package stubs

Three review findings:

NB_COMPACT_ASSERTIONS was set on the extension target, where it does
nothing: the macro is only read by nanobind’s own src/ files. It now also
reaches the target that compiles nb_combined.cpp, under the same non-Debug
condition. nanobind’s documented manual build applies it to both.

Stub generation wrote nothing into the package. pybind11-stubgen -o .
created the ostk/core/ tree, whereas nanobind’s stubgen -O . writes the
named module’s stubs flat into that directory, so .pyi files landed beside
pyproject.toml and the ostk.core package-data glob never matched them.
The wheel shipped py.typed and no stubs at all, a regression against
pybind11. Output now goes to ${PROJECT_PATH}.

The second String.__add__ overload took String, making it an exact
duplicate of the first. main had self + String() alongside
self += std::string() and std::string() + self; std::string is the
operand that completes that set. Behaviour is unchanged either way, since
implicitly_convertible<std::string, String> already accepted a str, but
the stub now shows String + str instead of the same overload twice.


Not addressed: the BUILD_WHEELS_IN_CMAKE=OFF branch still generates no stubs.
That block is byte-identical to main and never generated them under pybind11
either; nothing in CI or the Makefile sets that option, which defaults to ON.
Adding stubs there is a separate change.
Co-Authored-By: Claude Opus 5 (1M context) noreply@anthropic.com

Update bindings/python/src/OpenSpaceToolkitCorePy/Container/Dictionary.cpp

Update bindings/python/test/types_/test_string.py

feat: upgrade to nanobind 3

Update bindings/python/src/OpenSpaceToolkitCorePy/Container/Dictionary.cpp

chore: style



Co-authored-by: Claude Opus 5 (1M context) noreply@anthropic.com

feat!: deprecate Real and Integer, replace uses with double and int (#214)

feat: deprecate Real and Integer, replace uses with double and int


Mark type::Real and type::Integer as [[deprecated]] and replace all uses
within the library, tests and Python bindings with double and int.

Object, Logger, Pump, Table, Dictionary and String now use int/double
Array::isNear(tolerance) and Array::toString() support arithmetic types
Object::Integer(…) and Object::Real(…) no longer default to Undefined
Python Integer/Real classes carry deprecation notes in their docstrings

Co-Authored-By: Claude Sonnet 5.5 noreply@anthropic.com

fix: make Array arithmetic isNear overflow-safe and toString locale-independent


isNear(tolerance): compute integral distances in the unsigned domain so
INT_MIN/INT_MAX and unsigned types are handled; keep a separate
floating-point comparison
toString: imbue the classic locale before streaming floating-point values
fix the isNear comparator example in Array.hpp

Co-Authored-By: Claude Sonnet 5.5 noreply@anthropic.com

chore: apply batched suggestions from code review

Co-authored-by: Antoine Paletta 98616558+apaletta3@users.noreply.github.com

feat: address feedback

fix: update Array isNear NaN test for reviewed comparison


The reviewed isNear change compares with distance > tolerance, so a NaN
distance is no longer rejected. Update the NaN test expectation to match.
Co-Authored-By: Claude Sonnet 5.5 noreply@anthropic.com

fix: keep Array isNear NaN-safe for floating-point elements

Revert the distance > tolerance form from the review suggestion: it
treats a NaN distance as near. Restore !(distance <= tolerance) and the
original NaN test expectation.
Co-Authored-By: Claude Sonnet 5.5 noreply@anthropic.com

Co-authored-by: Claude Sonnet 5.5 noreply@anthropic.com
Co-authored-by: Antoine Paletta 98616558+apaletta3@users.noreply.github.com

refactor!: replace Undefined pattern with std::optional (#217)

Path, File and Directory are now always defined, and “no value” is
expressed with std::optional instead of an Undefined() sentinel.

Remove Undefined() and isDefined() from Path, File and Directory, along
with the guards that threw on undefined instances
Remove the unimplemented Undefined()/isDefined() declarations from
System::User, System::Group and Logger::Source
Remove Object::Format::Undefined and Table::Format::Undefined; Object::Parse,
Object::Load, Object::toString, Dictionary::Parse and Table::Load take
std::optional = std::nullopt, and nullopt keeps the previous
auto-detect behaviour (Object::toString still throws without a format)
Python: remove Path/File/Directory undefined() and is_defined() and
Object.Format.Undefined; Object.parse/load and Dictionary.parse default
format to None

Object::Undefined, Real, Integer and Sign are unchanged.
BREAKING CHANGE: removed Path::Undefined, Path::isDefined, File::Undefined,
File::isDefined, Directory::Undefined, Directory::isDefined,
User/Group/Source::Undefined and ::isDefined, Object::Format::Undefined and
Table::Format::Undefined. Callers that need an empty value should use
std::optional, and callers passing Format::Undefined should pass
std::nullopt. Python: Path/File/Directory.undefined() and .is_defined() and
Object.Format.Undefined are removed.
Co-authored-by: Claude Sonnet 5.5 noreply@anthropic.com

refactor!: normalize getters to return const references for stored members (#218)

Only getters are kept. Stored non-trivial members return const T& so the
caller decides whether to copy. Computed and trivial values return by value.

Object: getString/getDictionary/getArray return const T& (with &&
overloads returning a copy so temporaries do not dangle); the const
access{Boolean,Integer,Real,String,Dictionary,Array} overloads are removed
Array: accessFirst/accessLast (const) become getFirst/getLast; accessWhere
is removed in favour of getWhere
Dictionary iterators: const accessKey/accessValue/accessMapIterator become
getKey/getValue/getMapIterator
Exception::getScope, RuntimeError::getMessage/getStackTrace and
File/Directory::getPath return const T&
Python: ConstIterator.access_key/access_value become get_key/get_value

BREAKING CHANGE: the const access* overloads on Object, Array and the
Dictionary iterators are removed or renamed to get*, Array::accessWhere is
removed, Python access_key/access_value are renamed to get_key/get_value, and
Object::getString/getDictionary/getArray are ref-qualified so member pointers
to them are ambiguous.
Co-authored-by: Claude Sonnet 5.5 noreply@anthropic.com

Co-authored-by: Claude Opus 5 (1M context) noreply@anthropic.com
Co-authored-by: Antoine Paletta 98616558+apaletta3@users.noreply.github.com