added a TLS using example for task

[dietmarkuehl]

No description provided.

[Copilot]

Pull request overview

Adds a new example that demonstrates using a thread-local allocator with beman::execution::task, and wires it into the examples build.

Changes:

• Add examples/alloc-2.cpp showcasing a TLS-backed allocator_type for tasks.
• Register the new alloc-2 example target in examples/CMakeLists.txt.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 4 comments.

File	Description
examples/alloc-2.cpp	New allocator example using thread_local state to select a std::pmr::memory_resource for task allocations.
examples/CMakeLists.txt	Adds alloc-2 to the list of built examples (non-MSVC).

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

File header comment says examples/alloc-1.cpp but this is alloc-2.cpp, which can confuse readers and tooling that relies on the header banner. Update the comment to match the actual filename.

Suggested change

	// examples/alloc-1.cpp -*-C++-*-
	// examples/alloc-2.cpp -*-C++-*-

[Copilot]

The throwing global operator new(std::size_t) is required to throw std::bad_alloc on allocation failure; returning nullptr is undefined behavior for callers that assume a valid pointer. Consider checking the malloc result and throwing, or implement only the nothrow overload if you intend to return null.

Suggested change

	auto p = std::malloc(n);
	std::cout << " global new(" << n << ")->" << p << "\n";
	void* p = std::malloc(n);
	std::cout << " global new(" << n << ")->" << p << "\n";
	if (!p) {
	throw std::bad_alloc();
	}

[Copilot]

std::pmr::memory_resource::do_allocate(bytes, alignment) must honor the requested alignment. Ignoring the alignment parameter and using malloc can return misaligned storage for over-aligned allocations. Use an aligned allocation strategy (e.g., ::operator new(bytes, std::align_val_t(alignment)) + matching aligned delete, or std::aligned_alloc where available).

Suggested change

	void* do_allocate(std::size_t n, std::size_t) override {
	auto p{std::malloc(n)};
	std::cout << " resource::allocate(" << n << ")->" << p << "\n";
	return p;
	}
	void do_deallocate(void* p, std::size_t n, std::size_t) override {
	std::cout << " resource::deallocate(" << p << ", " << n << ")\n";
	std::free(p);
	void* do_allocate(std::size_t n, std::size_t alignment) override {
	void* p = nullptr;
	#if defined(__cpp_aligned_new)
	if (alignment > __STDCPP_DEFAULT_NEW_ALIGNMENT__) {
	p = ::operator new(n, std::align_val_t(alignment));
	} else
	#endif
	{
	p = ::operator new(n);
	}
	std::cout << " resource::allocate(" << n << ")->" << p << "\n";
	return p;
	}
	void do_deallocate(void* p, std::size_t n, std::size_t alignment) override {
	std::cout << " resource::deallocate(" << p << ", " << n << ")\n";
	#if defined(__cpp_aligned_new)
	if (alignment > __STDCPP_DEFAULT_NEW_ALIGNMENT__) {
	::operator delete(p, n, std::align_val_t(alignment));
	} else
	#endif
	{
	::operator delete(p, n);
	}

[Copilot]

tls_allocator::set(&res) stores a pointer to a stack resource in a thread_local that outlives res. Even though this example exits soon after, leaving a thread_local pointing at a soon-to-be-destroyed object is easy to copy/paste into real code and becomes a dangling pointer hazard. Consider resetting back to std::pmr::new_delete_resource() after the sync_wait, or wrap the change in an RAII guard object.

[coveralls]

coverage: 94.212%. remained the same
when pulling ebc86d4 on tls-task
into 768ee35 on main.