Pretty printers maintenance

[Jacobfaib]

Description

1. Tighten timeout for pretty printers
2. Compile them all in optimized mode
3. Properly handle printing references (previously this wouldn't match the printers, now it does)
4. Don't need to go up a frame to inspect values, we can just do it inside the function itself but then we need to inline the asm() call
5. Unify some of the repeated boilerplate code

Checklist

• New or existing tests cover these changes.
• The documentation is up to date with these changes.

[coderabbitai]

Note

Reviews paused

It looks like this branch is under active development. To avoid overwhelming you with review comments due to an influx of new commits, CodeRabbit has automatically paused this review. You can configure this behavior by changing the reviews.auto_review.auto_pause_after_reviewed_commits setting.

Use the following commands to manage reviews:

• @coderabbitai resume to resume automatic reviews.
• @coderabbitai review to trigger a single review.

Use the checkboxes below for quick actions:

• ▶️ Resume reviews
• 🔍 Trigger review

📝 Walkthrough

Summary by CodeRabbit

• Bug Fixes

  • Improved CUDA pretty-printer handling of reference-qualified values in GDB and LLDB.
  • Corrected type recognition and displayed values for atomic, buffer, complex, event, stream, vector, memory-resource, array, and tuple types.
  • Standardized debugger output with consistent addresses, reference qualifiers, and object formatting.

• Tests

  • Pretty-printer tests now run with optimized settings and shorter timeouts.
  • Simplified test case configuration and strengthened output normalization for more reliable debugger validation.

Walkthrough

Pretty-printer implementations now handle reference-valued CUDA types in GDB and LLDB. Debugger tests use optimized builds, simplified case arguments, shorter timeouts, and expanded output normalization.

Changes

Debugger case execution and build configuration

Layer / File(s)	Summary
Case parsing and debugger execution libcudacxx/test/debugging/CMakeLists.txt, libcudacxx/test/debugging/run_pretty_printer_test.py	Cases now use breakpoint, section, and expression values. GDB and LLDB no longer select caller frames. Breakpoint grouping and output normalization were updated.
GDB reference normalization libcudacxx/share/libcudacxx/gdb/*	GDB pretty-printers strip references from types and values before type matching, value extraction, and field inspection.
LLDB reference normalization libcudacxx/share/libcudacxx/lldb/*	LLDB pretty-printers dereference reference values and canonicalize types before summaries, synthetic providers, and handle inspection.
Optimized debugger fixtures libcudacxx/test/debugging/*	Test sources use KEEP_FOR_DEBUGGER. Case wiring and LLDB expectations now match direct expressions, optimized execution, addresses, and reference formatting.

Possibly related PRs

• NVIDIA/cccl#10520: Directly modifies stream pretty-printers and their associated tests.

Suggested reviewers: bernhardmgruber, davebayer

---

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

[coderabbitai]

Actionable comments posted: 1

🧹 Nitpick comments (1)

libcudacxx/test/debugging/CMakeLists.txt (1)

165-168: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

suggestion: Update the helper documentation for the new build mode.

The documentation for libcudacxx_add_pretty_printer_test still says that targets have host debug information and disabled optimization. The implementation now requests -O3 and NDEBUG, and removes explicit -g. Update that contract after restoring debug information. (raw.githubusercontent.com)

Source: MCP tools

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: 7b8d6e00-ac46-4d8c-a952-ac04b6fa0a24

📥 Commits

Reviewing files that changed from the base of the PR and between fef5a52 and 5b80f36.

📒 Files selected for processing (1)

• libcudacxx/test/debugging/CMakeLists.txt

[coderabbitai]

Actionable comments posted: 2

🧹 Nitpick comments (1)

libcudacxx/test/debugging/run_pretty_printer_test.py (1)

528-530: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

suggestion: Add regression tests for > > >, quoted text containing 42u, addresses, and suffixed integers. Run the targeted pretty-printer test with the repository CMake/Ninja preset and run pre-commit run --all-files before merge.

As per path instructions, files under libcudacxx/test follow .agent/skills/cccl-test/SKILL.md, and changes require targeted validation and pre-commit checks.

Source: Path instructions

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: 055a8252-208d-4627-b6ff-b77c39e5d8c6

📥 Commits

Reviewing files that changed from the base of the PR and between 5cd42f0 and ca7c287.

📒 Files selected for processing (1)

• libcudacxx/test/debugging/run_pretty_printer_test.py

[coderabbitai]

Actionable comments posted: 2

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: 3542af87-4774-4126-884a-e537f0257eb2

📥 Commits

Reviewing files that changed from the base of the PR and between ca7c287 and 88fc073.

📒 Files selected for processing (42)

• libcudacxx/share/libcudacxx/gdb/atomic.py
• libcudacxx/share/libcudacxx/gdb/buffer.py
• libcudacxx/share/libcudacxx/gdb/complex.py
• libcudacxx/share/libcudacxx/gdb/event.py
• libcudacxx/share/libcudacxx/gdb/inplace_vector.py
• libcudacxx/share/libcudacxx/gdb/memory_resource.py
• libcudacxx/share/libcudacxx/gdb/std_array.py
• libcudacxx/share/libcudacxx/gdb/tuple.py
• libcudacxx/share/libcudacxx/lldb/atomic.py
• libcudacxx/share/libcudacxx/lldb/buffer.py
• libcudacxx/share/libcudacxx/lldb/complex.py
• libcudacxx/share/libcudacxx/lldb/event.py
• libcudacxx/share/libcudacxx/lldb/inplace_vector.py
• libcudacxx/share/libcudacxx/lldb/memory_resource.py
• libcudacxx/share/libcudacxx/lldb/std_array.py
• libcudacxx/share/libcudacxx/lldb/tuple.py
• libcudacxx/test/debugging/CMakeLists.txt
• libcudacxx/test/debugging/array/CMakeLists.txt
• libcudacxx/test/debugging/array/lldb.expected
• libcudacxx/test/debugging/array/source.cu
• libcudacxx/test/debugging/atomic/CMakeLists.txt
• libcudacxx/test/debugging/atomic/lldb.expected
• libcudacxx/test/debugging/atomic/source.cu
• libcudacxx/test/debugging/buffer/CMakeLists.txt
• libcudacxx/test/debugging/buffer/lldb.expected
• libcudacxx/test/debugging/buffer/source.cu
• libcudacxx/test/debugging/complex/CMakeLists.txt
• libcudacxx/test/debugging/complex/lldb.expected
• libcudacxx/test/debugging/complex/source.cu
• libcudacxx/test/debugging/event/CMakeLists.txt
• libcudacxx/test/debugging/event/lldb.expected
• libcudacxx/test/debugging/event/source.cu
• libcudacxx/test/debugging/inplace_vector/CMakeLists.txt
• libcudacxx/test/debugging/inplace_vector/lldb.expected
• libcudacxx/test/debugging/inplace_vector/source.cu
• libcudacxx/test/debugging/memory_resource/CMakeLists.txt
• libcudacxx/test/debugging/memory_resource/lldb.expected
• libcudacxx/test/debugging/memory_resource/source.cu
• libcudacxx/test/debugging/run_pretty_printer_test.py
• libcudacxx/test/debugging/tuple/CMakeLists.txt
• libcudacxx/test/debugging/tuple/lldb.expected
• libcudacxx/test/debugging/tuple/source.cu

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

• libcudacxx/test/debugging/CMakeLists.txt

[coderabbitai]

Note

GitHub couldn't provide a complete incremental comparison for this pull request, so CodeRabbit is performing a full review instead. This review may take a little longer.

[miscco]

This looks o much better

[github-actions]

⏱️ CCCL compile-time benchmark comparison: Public headers compile-time bench

Result: 6 regression row(s), 2 improvement row(s) above threshold.

Run	Value
Config	public-headers-gcc13
Baseline	origin/main
Preset	all-dev
Targets	cub.headers.base, thrust.cpp.cuda.headers.base, libcudacxx.test.public_headers
GPU / launch args	rtx2080 / --cuda 13.3 --host gcc13

Artifacts: reports and traces

Direct file processing

-f file-processing exclusive --sort total

🔴 Direct file processing — Regressions

Rank	Regression impact	Selected Δ	Baseline	Current	Event	Matched traces
1	0.961435	0.961435	20.036825	20.998260	Processing Header File: libcudacxx/include/cuda/__device/physical_device.h	94
2	0.749965	0.749965	10.916407	11.666372	Processing Header File: libcudacxx/include/cuda/std/__iterator/concepts.h	437
3	0.308120	0.308120	11.491588	11.799708	Processing Header File: libcudacxx/include/cuda/std/__type_traits/type_list.h	477
4	0.263980	0.263980	5.674576	5.938556	Processing Header File: libcudacxx/include/cuda/std/__iterator/iterator_traits.h	464
5	0.252777	0.252777	1.872785	2.125562	Processing Header File: cub/cub/device/dispatch/tuning/tuning_scan.cuh	68
6	0.229325	0.229325	2.913285	3.142610	Processing Header File: libcudacxx/include/cuda/std/__ranges/concepts.h	405

🟢 Direct file processing — Improvements

Rank	Improvement impact	Selected Δ	Baseline	Current	Event	Matched traces
1	1.350740	-1.350740	9.915596	8.564856	Processing Header File: libcudacxx/include/cuda/std/__cccl/prologue.h	550
2	0.387724	-0.387724	3.412547	3.024823	Processing Header File: libcudacxx/include/cuda/std/__cccl/epilogue.h	550

[github-actions]

🥳 CI Workflow Results

🟩 Finished in 2h 24m: Pass: 100%/68 | Total: 1d 02h | Max: 57m 54s | Hits: 98%/306720

See results here.

AI failure analysis

1. Merged RAPIDS environment requests incompatible rapids-logger versions · 1 job

Explanation: The RAPIDS build stops before compilation because its generated Conda environment simultaneously requires rapids-logger 0.2.x and 0.3.x. The PR's debugger changes are not reached or implicated by the failure.

Evidence:

Build RAPIDS (optional) / rmm ucxx kvikio rapidsmpf cudf cudf_kafka, step 6

2026-08-07T18:38:42.1008703Z - rapids-logger==0.2.*,>=0.0.0a0
2026-08-07T18:38:42.1009044Z - rapids-logger==0.3.*,>=0.0.0a0
2026-08-07T18:39:16.4634196Z error    libmamba Could not solve for environment specs

Root cause: Dependency files from the selected main-branch RAPIDS repositories were merged into one environment while different repositories required mutually exclusive rapids-logger minor versions. The log does not retain enough attribution to determine which generated per-repository environment supplied each constraint. Sources: .github/workflows/build-rapids.yml:66, ci/rapids/post-create-command.sh:38, ci/rapids/cuda13.3-conda/devcontainer.json:90.

Suggested next steps: Inspect the generated per-repository Conda YAML files to identify the sources of the 0.2.x and 0.3.x constraints, then align those downstream dependencies or temporarily pin the affected RAPIDS repository to a compatible commit. Reproduce narrowly with `RAPIDS_LIBS='rmm ucxx kvikio rapidsmpf cudf cudf_kafka' .devcontainer/launch.sh -d -c 13.3 -H rapids-conda -- ./ci/rapids/rapids-entrypoint.sh /bin/bash -li -c 'build-all -j0 -v'`.

Copy this prompt into a coding agent

Repository: https://github.com/NVIDIA/cccl
Workflow run: https://github.com/NVIDIA/cccl/actions/runs/31207630862
Failure group: Merged RAPIDS environment requests incompatible rapids-logger versions
Affected jobs:
- Build RAPIDS (optional) / rmm ucxx kvikio rapidsmpf cudf cudf_kafka: https://github.com/NVIDIA/cccl/actions/runs/31207630862/job/92962684267

Reproduce the RAPIDS environment-generation failure for `rmm ucxx kvikio rapidsmpf cudf cudf_kafka` using the CUDA 13.3 RAPIDS Conda devcontainer. Before Conda solves the merged environment, inspect every generated per-repository YAML and identify exactly which repository requests `rapids-logger==0.2.*` and which requests `rapids-logger==0.3.*`. Verify the intended rapids-logger version for the pinned 26.10 RAPIDS stack, then implement the smallest fix: preferably align the outdated downstream dependency; if that cannot be changed here, pin the drifting repository through the appropriate `RAPIDS_<lib>_GIT_REPO` workflow variable to a mutually compatible commit. Confirm that the regenerated merged environment contains only one compatible rapids-logger constraint, run the focused environment creation/build command, and avoid changing the unrelated debugger-printer code.

Jobs:

• Build RAPIDS (optional) / rmm ucxx kvikio rapidsmpf cudf cudf_kafka