Refactor host_span to use cuda::std::span internally and remove span_base

[PointKernel]

Description

Contributes to #20539.

This PR refactors cudf::host_span to be backed by cuda::std::span internally instead of the custom cudf::detail::span_base, and removes span_base entirely now that it has no other users. The custom logic needed for copy-engine optimizations is retained: host_span still tracks is_device_accessible and still constructs from cudf-supported host containers.

Not in this PR: removing the custom device_span alias and dropping cudf::dynamic_extent.

Checklist

• I am familiar with the Contributing Guidelines.
• New or existing tests cover these changes.
• The documentation is up to date with these changes.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: b97c90c0-1862-43d8-a54c-5900a9de17ce

📥 Commits

Reviewing files that changed from the base of the PR and between cde923f and 138feb1.

📒 Files selected for processing (1)

• cpp/include/cudf/utilities/span.hpp

🚧 Files skipped from review as they are similar to previous changes (1)

• cpp/include/cudf/utilities/span.hpp

---

📝 Walkthrough

Summary by CodeRabbit

• Refactor
  • Updated span utilities to use newer CUDA span primitives while keeping the existing host_span API and behavior intact.
  • Switched dynamic_extent to use cuda::std::dynamic_extent.
  • Reworked host_span to wrap an internal cuda::std::span and updated its accessors, constructors, subspan, and conversions accordingly.
  • Added missing forward declarations for null-mask helper operations to improve compilation compatibility.
• Chores
  • Refreshed copyright header text.

Walkthrough

This PR refactors cudf::host_span to wrap cuda::std::span, updates dynamic_extent to use cuda::std::dynamic_extent, and adds forward declarations for two cudf::detail bitmask helper templates in null_mask.cuh, along with SPDX header updates.

Changes

host_span Internal Storage Refactor

Layer / File(s)	Summary
dynamic_extent redefinition and includes cpp/include/cudf/utilities/span.hpp	dynamic_extent switches to cuda::std::dynamic_extent, the unused <limits> include is removed, and the SPDX header text is updated.
host_span class definition and constructors cpp/include/cudf/utilities/span.hpp	host_span stops inheriting span_base, stores _span as a cuda::std::span, and initializes _span and _is_device_accessible across the pointer, container, const-container, and copy constructors.
host_span accessor and iterator methods cpp/include/cudf/utilities/span.hpp	operator[], front, back, begin, end, data, size, size_bytes, empty, first, last, subspan, and the conversion operator to std::span<T> now delegate to _span.

null_mask.cuh Forward Declarations

Layer / File(s)	Summary
Forward declarations for bitmask binop helpers cpp/include/cudf/detail/null_mask.cuh	Forward declarations for inplace_bitmask_binop and inplace_segmented_bitmask_binop are added before later template code uses them, and the SPDX header text is updated.

Estimated code review effort: 3 (Moderate) | ~25 minutes

Possibly related PRs

• rapidsai/cudf#22588: Related host_span API and implementation migration involving the same header and span-type changes.

Suggested reviewers: shrshi, mroeschke, wence-

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly summarizes the main change: refactoring host_span to use cuda::std::span and removing span_base.
Description check	✅ Passed	The description matches the changeset and accurately describes the host_span refactor and span_base removal.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

_{Comment @coderabbitai help to get the list of available commands.}

[coderabbitai]

Actionable comments posted: 1

Caution

Some comments are outside the diff and can’t be posted inline due to platform limitations.

⚠️ Outside diff range comments (1)

cpp/include/cudf/utilities/span.hpp (1)

114-127: 🎯 Functional Correctness | 🟠 Major | ⚡ Quick win

HIGH: Preserve allocator device-accessibility in container constructors.

_is_device_accessible now stays at the default false for container-backed spans, so pinned host vectors lose the flag that copy-engine paths rely on.

Proposed fix

 private:
   using span_type = cuda::std::span<T, Extent>;  ///< The underlying span type
+
+  template <typename C>
+  static constexpr bool container_is_device_accessible(C const& in)
+  {
+    if constexpr (requires { in.get_allocator().is_device_accessible(); }) {
+      return in.get_allocator().is_device_accessible();
+    } else {
+      return false;
+    }
+  }
 
 public:
@@
-  constexpr host_span(C& in) : _span{thrust::raw_pointer_cast(in.data()), in.size()}
+  constexpr host_span(C& in)
+    : _span{thrust::raw_pointer_cast(in.data()), in.size()},
+      _is_device_accessible{container_is_device_accessible(in)}
   {
   }
@@
-  constexpr host_span(C const& in) : _span{thrust::raw_pointer_cast(in.data()), in.size()}
+  constexpr host_span(C const& in)
+    : _span{thrust::raw_pointer_cast(in.data()), in.size()},
+      _is_device_accessible{container_is_device_accessible(in)}
   {
   }

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@cpp/include/cudf/utilities/span.hpp` around lines 114 - 127, The
container-backed host_span constructors are not preserving allocator
device-accessibility, so spans built from device-accessible host containers lose
the flag that copy-engine paths depend on. Update the host_span(C&) and
host_span(C const&) constructors to derive and store _is_device_accessible from
the source container’s allocator/device-accessibility trait, consistent with the
existing span state. Use the host_span template constructors and the
container-support checks already present to locate the fix.

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@cpp/include/cudf/utilities/span.hpp`:
- Around line 233-246: The host_span subview helpers first() and last() are
dropping the device-accessibility state, so pinned-memory slices lose the
_is_device_accessible flag while subspan() preserves it. Update host_span::first
and host_span::last to construct the returned host_span with the same
accessibility metadata as the source span, using the existing host_span
constructors/fields around _span and _is_device_accessible so these views stay
device-accessible.

---

Outside diff comments:
In `@cpp/include/cudf/utilities/span.hpp`:
- Around line 114-127: The container-backed host_span constructors are not
preserving allocator device-accessibility, so spans built from device-accessible
host containers lose the flag that copy-engine paths depend on. Update the
host_span(C&) and host_span(C const&) constructors to derive and store
_is_device_accessible from the source container’s allocator/device-accessibility
trait, consistent with the existing span state. Use the host_span template
constructors and the container-support checks already present to locate the fix.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: 114a1e0b-01ec-4f62-b605-abc2e21827e6

📥 Commits

Reviewing files that changed from the base of the PR and between e7cae5e and 93ac913.

📒 Files selected for processing (2)

• cpp/include/cudf/detail/null_mask.cuh
• cpp/include/cudf/utilities/span.hpp

[PointKernel]

Re: the outside-diff comment on the container constructors (host_span(C&) / host_span(C const&)) not preserving allocator device-accessibility — not applying this one.

The container constructors defaulted _is_device_accessible to false before this PR as well, so it isn't a regression. Device-accessibility for cudf's pinned host containers is carried by host_vector's own conversion operator, which already passes get_allocator().is_device_accessible() into the 3-arg host_span ctor:

cudf/cpp/include/cudf/detail/utilities/host_vector.hpp

Lines 217 to 227 in 93ac913

	[[nodiscard]] operator host_span<T const>() const
	{
	return host_span<T const>{
	base::data(), base::size(), base::get_allocator().is_device_accessible()};
	}
	[[nodiscard]] operator host_span<T>()
	{
	return host_span<T>{base::data(), base::size(), base::get_allocator().is_device_accessible()};
	}

The generic host_span(C&) ctor only handles std::vector / thrust::host_vector / std::basic_string, whose allocators have no is_device_accessible(), so the proposed requires-guarded helper would return false for all of them and change nothing.

[PointKernel]

/merge