Files
ocp/src/common/occt_handle.h
stroblme 091a2ad0cf Housekeeping ahead of the Inc 1-4 coverage work
- inventory.py --methods: report the instance methods the app calls per class,
  plus chained calls that constrain a return type. --check says which classes
  to bind; this says what to bind on them, which is what writing ~38 module
  TUs needs.
- Drop StlAPI from the inventory: StlAPI_Writer has no app call site (the only
  use was a test fixture, now on the app's own STL writer). 138 symbols / 47
  modules.
- adding-symbols.md: scope the executing-constructor ban to the BRepAlgoAPI
  booleans, which are the only classes with a deferred Set*/Build form —
  BRepMesh_IncrementalMesh, GeomAPI_*, BRepCheck_Analyzer and friends compute
  in their constructor by design and bind as stock. Replace the per-increment
  app-test guidance: backend/tests/conftest.py imports n3xd.main, so no app
  test can collect until the last module is bound. Increments gate on
  stock-recorded fixtures here; the app suite is the Inc 4 gate.
- parity_venv.sh: run the ocp suite in the swapped venv (it imports only
  OCP/n3xd_ocp, so it works throughout).
- Fix a stale macro name in occt_handle.h (ocp_new, not OCP_TRANSIENT_NEW) and
  drop the unused ocp_transient_class helper.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DfriM8XUkn7uYf5Dwe2xo6
2026-08-10 19:23:27 +02:00

125 lines
4.9 KiB
C++

/*
Type caster for opencascade::handle<T> — the ownership core of this binding.
OCCT transients carry their own intrusive, atomic reference count
(Standard_Transient), so a handle keeps an object alive entirely on the C++
side. That is what lets us release the GIL around kernel calls: OCCT may
copy handles on its own worker threads without ever touching Python.
The design mirrors nanobind's own stl/shared_ptr.h caster:
C++ -> Python the wrapper is a *non-owning* nanobind instance pointing
at the C++ object, plus one handle stored in the object's
keep-alive list. Python holds exactly one OCCT reference
per wrapper, released at deallocation. An existing
wrapper is reused (is_new == false), so `a is b` holds for
as long as a wrapper stays alive.
Python -> C++ a plain handle copy (one incref), balanced when the caster
dies after the call. Unlike shared_ptr we do *not* need
to keep the PyObject alive: the OCCT refcount, not the
Python instance, owns the object's memory.
That last sentence is only true if every wrapped transient was heap
allocated and is handle-owned. See occt_transient.h — transient
constructors are bound through ocp_new<T, Args...>(), never nb::init<>, so
Python never owns transient storage. from_python re-checks the invariant
rather than trusting it, because the failure mode is a double free.
*/
#pragma once
#include <nanobind/nanobind.h>
#include <Standard_Handle.hxx>
#include <Standard_Transient.hxx>
#include <type_traits>
NAMESPACE_BEGIN(NB_NAMESPACE)
NAMESPACE_BEGIN(detail)
// Marked NB_NOINLINE so the (identical) body is not duplicated into every
// instantiation of the caster below — there is one per bound transient class.
inline NB_NOINLINE void
occt_handle_keep_alive(opencascade::handle<Standard_Transient> &&h,
PyObject *o) noexcept {
keep_alive(o, new opencascade::handle<Standard_Transient>(std::move(h)),
[](void *p) noexcept {
delete (opencascade::handle<Standard_Transient> *) p;
});
}
template <typename T>
struct type_caster<
opencascade::handle<T>,
enable_if_t<std::is_base_of_v<Standard_Transient, std::decay_t<T>>>> {
static constexpr bool IsClass = true;
using Caster = make_caster<T>;
using Td = std::decay_t<T>;
NB_TYPE_CASTER(opencascade::handle<T>, Caster::Name)
static_assert(is_base_caster_v<Caster>,
"Conversion of opencascade::handle<T> requires that T is "
"bound through nanobind's regular class mechanism.");
bool from_python(handle src, uint8_t flags,
cleanup_list *cleanup) noexcept {
// A null handle is OCCT's absent value and maps to None in both
// directions; several APIs (BRep_Tool::Surface on a degenerate edge,
// for one) legitimately return one.
if (src.is_none()) {
value = Value();
return true;
}
flags &= ~((uint8_t) cast_flags::convert);
Caster caster;
if (!caster.from_python(src, flags, cleanup))
return false;
Td *ptr = caster.operator Td *();
// Invariant check, not defensive programming: a refcount of zero means
// this wrapper owns its storage (nb::init<>, or a by-value return),
// and taking a handle to it would hand OCCT the right to `delete` a
// nanobind instance's memory. Refuse instead — a TypeError beats a
// heap corruption, and it can only be reached by a binding bug.
if (ptr && ptr->GetRefCount() == 0) {
assert(!"OCCT transient is not handle-owned (see occt_transient.h)");
return false;
}
value = Value(ptr);
return true;
}
static handle from_cpp(const Value &v, rv_policy,
cleanup_list *cleanup) noexcept {
Td *ptr = v.get();
if (!ptr)
return none().release();
// Transients are polymorphic, so nb_type_put_p downcasts on the
// dynamic type: BRep_Tool::Surface returning a Geom_Surface handle to
// a plane arrives in Python as Geom_Plane when that class is bound.
bool is_new = false;
handle result = nb_type_put_p(&typeid(Td), &typeid(*ptr), (void *) ptr,
rv_policy::reference, cleanup, &is_new);
// Only a freshly created wrapper takes a reference. Attaching one per
// conversion would pile up redundant handles on a long-lived object
// that crosses the boundary many times.
if (result.is_valid() && is_new)
occt_handle_keep_alive(
opencascade::handle<Standard_Transient>(ptr), result.ptr());
return result;
}
};
NAMESPACE_END(detail)
NAMESPACE_END(NB_NAMESPACE)