[Misc] Migrate to nanobind#759
Open
hughperkins wants to merge 12 commits into
Open
Conversation
Port the build system and all C++ binding units to nanobind: - pyproject.toml + cmake: nanobind deps, PythonNanobind.cmake, nanobind_add_module/stub - export.* / interfaces_registry: NB_MODULE, namespace nb, stl header swaps - export_lang/math/misc/stream: API renames, rv_policy, getstate/setstate pickle, init lambdas, eigen, enum tag struct, set_lib_dir bytes arg, Program weakref - dlpack_funcs: nb capsule/object, manual PyCapsule for DLPack consumed protocol - memory_usage_monitor: drop pybind11 embed, read RSS from /proc on Linux - py_exception_translator: register from NB_MODULE (not static init) to avoid dlopen crash CPU smoke test green (kernel + field ops).
- NdarrayCxx: nb::is_weak_referenceable() (frontend caches weakrefs to ndarray kernel args for launch-context eviction); fixes 22 ndarray test failures. - export.h: include nanobind/stl/variant.h so ASTBuilder::create_print's std::vector<std::variant<Expr,std::string>> arg converts (fixes print tests).
…o_dlpack - expr.py: coerce np.bool_ -> native bool before make_const_expr_bool (nanobind's bool caster is strict; rejected numpy bool). Fixes ~42 const-bool failures (reduction, ifte, ast_refactor, diag, ...). - dlpack_funcs.cpp: include <nanobind/stl/string.h> so nb::cast<std::string> for the torch __version__ check resolves the string caster. Without it nanobind treated std::string as an unregistered bound type and threw std::bad_cast, breaking all field-backend to_dlpack (57 tensor-layout/dlpack failures).
- test_utils.exclude_arch_platform: use 'is not None' instead of truthiness. nanobind enum members are falsy when their int value is 0 (Arch.x64 == 0), unlike pybind11 enums; a bare 'if exclude_arch' silently failed to exclude the x64/cpu arch (test_test exclude_* + test_field sparse_not_supported). - make_global_load_stmt / make_global_store_stmt: add nb::arg().none() so None maps to nullptr for the Stmt* operands as pybind11 did (test_binding).
…ments Rename QD_INTERFACE_DEF_WITH_PYBIND11 -> QD_INTERFACE_DEF_WITH_NANOBIND (the macro already emits nanobind::module_ calls; only the name lagged) and update the few remaining comments that referenced pybind11 (snode copy-ctor, launch_context_builder bulk-arg note, logging singleton rationale) to be binding-accurate. No functional change.
Collaborator
Author
|
Genesis uni tests passing: 
|
Collaborator
Author
|
Apears we no longer need .pyi stub postprocessing: 
|
# Conflicts: # cmake/PythonNumpyPybind11.cmake # pyproject.toml
nanobind_add_stub emits the .pyi directly (richer, more accurate types than the old pybind11-stubgen output), so the YAML signature-rewrite step is gone. Remove the now-unused tools/gen_stubs.py, stub_replacements_funcs.yaml, stub_replacements_global.yaml, and the ruamel.yaml build dep they were the only consumer of.
Collaborator
Author
|
Genesis benchmark results: 
|
The old pybind11-stubgen YAML post-processing injected readable parameter names
(e.g. get_tensor_type's shape/element_type). Restore them the nanobind-native way via
nb::arg(...) on the affected defs (get_tensor_type, make_const_expr_{bool,int,fp},
create_kernel, create_function, set_arg_ndarray, make_get_element_expr) so the generated
.pyi shows named, keyword-callable params instead of arg0/arg1. Verified against a freshly
generated stub. All call sites are positional, so no runtime impact.
Pure formatting (line-wrapping/indentation) from pre-commit clang-format on the nanobind-ported files; no functional change.
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: db14c1aeb7
ℹ️ About Codex in GitHub
Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".
Comment on lines
+42
to
+43
| #endif | ||
| } |
There was a problem hiding this comment.
When start_memory_monitoring() is used on macOS or Windows, this replacement now falls into the non-Linux branch and records 0 for every sample. The previous psutil-based implementation worked anywhere psutil was available, so this silently disables the exported memory monitor for non-Linux users instead of reporting process RSS or failing clearly.
Useful? React with 👍 / 👎.
Collaborator
Author
|
reran genesis benchmarks after changes above. Still looking good: 
|
Collaborator
Author
|
Genesis unit tests still passing: 
|
The nanobind-generated .pyi now types these binding params precisely (vs the loose/Any signatures from the old pybind11-stubgen + YAML postprocessing), which surfaced 3 real type mismatches in the frontend: - expr.py make_const_expr_fp/int: coerce np.floating/np.integer to native float/int (mirrors the existing bool(val) coercion for make_const_expr_bool). - impl.py expr_subscript: add `assert indices_expr_group is not None` before the tensor-index call (same narrowing already used earlier in subscript()). Runtime behavior unchanged (CPU suite already green); these are static-typing fixes only.
…trations clang-tidy's bugprone-unused-raii flags the discarded `nb::exception<T>(...)` temporaries (the old pybind11 code used the free function py::register_exception, which isn't flagged). The construction is the registration; the handle is meant to be discarded. Suppress per line with NOLINTNEXTLINE, matching the existing convention already used for `nb::class_<BitStructType>` in this file.
# Conflicts: # quadrants/python/memory_usage_monitor.cpp # quadrants/python/memory_usage_monitor.h
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Issue: #
Brief Summary
copilot:summary
Walkthrough
copilot:walkthrough