Fix symbol export#2052
Open
vyasr wants to merge 12 commits intorapidsai:mainfrom
Open
Conversation
vyasr
added
improvement
Contributor
Author
|
/ok to test |
|
Auto-sync is disabled for draft pull requests in this repository. Workflows must be run manually. Contributors can view more details about this message here. |
Contributor
Author
|
/ok to test |
3 similar comments
Contributor
Author
|
/ok to test |
Contributor
Author
|
/ok to test |
Contributor
Author
|
/ok to test |
Add CUVS_EXPORT/CUVS_HIDDEN macros and set CXX_VISIBILITY_PRESET hidden on cuvs_objs and cuvs_c targets. Apply namespace CUVS_EXPORT cuvs to all installed C++ headers and CUVS_EXPORT to C API function declarations. Also fix pairwise_distance.cpp to include its own C API header so that the CUVS_EXPORT annotation on the function declaration is visible to the compiler when building with hidden default visibility.
The check-c-abi workflow only uses c/include as its header path. export.hpp was only present in cpp/include, causing CUVS_EXPORT to be undefined when the ABI checker parsed C headers, breaking all function signature comparisons. Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent) Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>
Under -fvisibility=hidden, explicit template instantiations in source files require CUVS_EXPORT to be exported from libcuvs.so. Without this, test binaries linking against libcuvs.so get undefined symbols (e.g. cuvs::neighbors::ivf_flat::index<__half, long>::n_lists()). Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent) Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>
NVCC does not support __attribute__((visibility("default"))) on
explicit class template instantiations (error rapidsai#1091-D). These symbols
are already exported via the namespace-level CUVS_EXPORT on the header
declarations. Only function template instantiations (template void)
need the per-instantiation CUVS_EXPORT in .cu files.
This function was the only C API declaration in pq.h missing the CUVS_EXPORT annotation, causing an undefined symbol error when the Python wheel tried to load pq.abi3.so against libcuvs.so built with -fvisibility=hidden.
… Doxygen - Rename c/include/cuvs/core/export.hpp -> export.h (pure C macros should use .h extension; fixes Java jextract build which only copies *.h files to its staging directory) - Update all 23 C API headers to include export.h instead of export.hpp - In Doxyfile: set MACRO_EXPANSION=YES, EXPAND_ONLY_PREDEF=YES, add CUVS_EXPORT= and CUVS_HIDDEN= to PREDEFINED so Doxygen strips the visibility attribute before generating XML, preventing breathe from choking on CUVS_EXPORT in C function signatures
Temporarily point RAFT_FORK and RAFT_PINNED_TAG at the raft symbol visibility PR branch to verify cuVS works with hidden-default raft.
|
No actionable comments were generated in the recent review. 🎉 ℹ️ Recent review info⚙️ Run configurationConfiguration used: Path: .coderabbit.yaml Review profile: CHILL Plan: Enterprise Run ID: 📒 Files selected for processing (1)
✅ Files skipped from review due to trivial changes (1)
📝 WalkthroughSummary by CodeRabbitRelease Notes
WalkthroughThis PR introduces a symbol visibility control system for cuVS by creating export macros ( ChangesSymbol Export Visibility System
Estimated code review effort🎯 4 (Complex) | ⏱️ ~75 minutes 🚥 Pre-merge checks | ✅ 4 | ❌ 1❌ Failed checks (1 warning)
✅ Passed checks (4 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches🧪 Generate unit tests (beta)
Tip 💬 Introducing Slack Agent: The best way for teams to turn conversations into code.Slack Agent is built on CodeRabbit's deep understanding of your code, so your team can collaborate across the entire SDLC without losing context.
Built for teams:
One agent for your entire SDLC. Right inside Slack. Review rate limit: 9/10 reviews remaining, refill in 6 minutes. Comment |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 2
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (8)
cpp/CMakeLists.txt (1)
931-940:⚠️ Potential issue | 🟠 Major | ⚡ Quick winPropagate the visibility defaults to
cuvs-cagra-searchtoo.
cuvs-cagra-searchis also compiled into the shared library, but it still builds with the compiler defaults. That leaves one object target outside the hidden-default setup, so the new export hygiene can still leak symbols.♻️ Suggested fix
set_target_properties( cuvs_objs PROPERTIES CXX_STANDARD 20 CXX_STANDARD_REQUIRED ON CUDA_STANDARD 20 CUDA_STANDARD_REQUIRED ON POSITION_INDEPENDENT_CODE ON CXX_VISIBILITY_PRESET hidden VISIBILITY_INLINES_HIDDEN ON ) + set_target_properties( + cuvs-cagra-search + PROPERTIES CXX_VISIBILITY_PRESET hidden + VISIBILITY_INLINES_HIDDEN ON + )🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@cpp/CMakeLists.txt` around lines 931 - 940, The cuvs-cagra-search target is not receiving the hidden-visibility defaults, allowing symbol leakage; update the CMake configuration so cuvs-cagra-search is configured like cuvs_objs (or included in the same set_target_properties call) to set CXX_STANDARD 20, CXX_STANDARD_REQUIRED ON, CUDA_STANDARD 20, CUDA_STANDARD_REQUIRED ON, POSITION_INDEPENDENT_CODE ON, CXX_VISIBILITY_PRESET hidden, and VISIBILITY_INLINES_HIDDEN ON for the cuvs-cagra-search target (reference the existing set_target_properties call for cuvs_objs to mirror or extend the same property list).c/src/distance/pairwise_distance.cpp (2)
46-53:⚠️ Potential issue | 🔴 Critical | ⚡ Quick winValidate the public C API inputs before dereferencing them.
translate_exceptionswon't save a nullcuvsResources_torDLManagedTensor*; this path will crash before it can return acuvsError_t. Please reject nullres,x_tensor,y_tensor, anddistances_tensorbefore touchingdl_tensor.
Based on coding guidelines, Input validation must check for negative or invalid dimensions, null pointers, and invalid parameter combinations before GPU operations.🛡️ Suggested fix
extern "C" cuvsError_t cuvsPairwiseDistance(cuvsResources_t res, DLManagedTensor* x_tensor, DLManagedTensor* y_tensor, DLManagedTensor* distances_tensor, cuvsDistanceType metric, float metric_arg) { return cuvs::core::translate_exceptions([=] { + if (res == nullptr || x_tensor == nullptr || y_tensor == nullptr || + distances_tensor == nullptr) { + RAFT_FAIL("cuvsPairwiseDistance requires non-null inputs"); + } auto x_dt = x_tensor->dl_tensor.dtype;🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@c/src/distance/pairwise_distance.cpp` around lines 46 - 53, The public C API cuvsPairwiseDistance must validate inputs before any dereference: add checks at the start of cuvsPairwiseDistance (before calling cuvs::core::translate_exceptions or touching dl_tensor) to reject null pointers for res, x_tensor, y_tensor, and distances_tensor and return an appropriate cuvsError_t; also validate tensor dimensions/sizes and metric/metric_arg combinations (negative dims, mismatched shapes, unsupported cuvsDistanceType) and return errors early so GPU code never runs with invalid inputs. Ensure these checks reference the same parameter names (cuvsResources_t res, DLManagedTensor* x_tensor, DLManagedTensor* y_tensor, DLManagedTensor* distances_tensor, cuvsDistanceType metric, float metric_arg) and occur prior to any use of dl_tensor or other members.
54-56:⚠️ Potential issue | 🔴 Critical | ⚡ Quick winRead dtype metadata from the matching tensors.
y_dtanddist_dtare both taken fromx_tensorright now, so mixed-dtype inputs will pass validation and dispatch the wrong specialization.🐛 Suggested fix
- auto x_dt = x_tensor->dl_tensor.dtype; - auto y_dt = x_tensor->dl_tensor.dtype; - auto dist_dt = x_tensor->dl_tensor.dtype; + auto x_dt = x_tensor->dl_tensor.dtype; + auto y_dt = y_tensor->dl_tensor.dtype; + auto dist_dt = distances_tensor->dl_tensor.dtype;🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@c/src/distance/pairwise_distance.cpp` around lines 54 - 56, The dtype variables are incorrectly read from x_tensor for all three; change the assignments so y_dt reads from y_tensor->dl_tensor.dtype and dist_dt reads from dist_tensor->dl_tensor.dtype (leave x_dt as x_tensor->dl_tensor.dtype) so that mixed-dtype inputs validate and dispatch correctly for functions/logic that use x_dt, y_dt, and dist_dt in pairwise_distance.cpp (variables x_tensor, y_tensor, dist_tensor).c/include/cuvs/preprocessing/quantize/scalar.h (1)
88-121:⚠️ Potential issue | 🟡 Minor | ⚡ Quick winAdd return documentation to the exported quantizer APIs.
cuvsScalarQuantizerTrain,cuvsScalarQuantizerTransform, andcuvsScalarQuantizerInverseTransformstill have parameter docs only. As per coding guidelines: public C API functions must include complete Doxygen documentation describing parameters, return values, and side effects.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@c/include/cuvs/preprocessing/quantize/scalar.h` around lines 88 - 121, The three exported functions cuvsScalarQuantizerTrain, cuvsScalarQuantizerTransform, and cuvsScalarQuantizerInverseTransform are missing `@return` (and any side-effect) Doxygen documentation; update each function comment block to include an `@return` describing the cuvsError_t values returned (e.g., CUVS_SUCCESS, error codes) and any side effects (e.g., device/host memory requirements, in-place modification, ownership of DLManagedTensor pointers), so the public C API docs are complete and follow the coding guideline.c/include/cuvs/neighbors/tiered_index.h (1)
214-220:⚠️ Potential issue | 🟡 Minor | ⚡ Quick winAdd the missing
@returnsection tocuvsTieredIndexSearch.The newly exported search API still lacks return-value documentation, which leaves the public C API docs incomplete. As per coding guidelines: public C API functions must include complete Doxygen documentation describing parameters, return values, and side effects.
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@c/include/cuvs/neighbors/tiered_index.h` around lines 214 - 220, Update the Doxygen comment for the exported function cuvsTieredIndexSearch to include a `@return` section that documents the cuvsError_t return values (e.g., CUVS_SUCCESS on success, specific CUVS_* error codes on failure such as invalid arguments, allocation or search errors) and any side-effects; locate the comment block above the cuvsTieredIndexSearch declaration and add a concise `@return` description listing possible return codes and their meaning so the public C API docs are complete.c/include/cuvs/preprocessing/quantize/binary.h (1)
104-139:⚠️ Potential issue | 🟡 Minor | ⚡ Quick winAdd return documentation to the exported binary quantizer APIs.
cuvsBinaryQuantizerTrain,cuvsBinaryQuantizerTransform, andcuvsBinaryQuantizerTransformWithParamsstill have parameter-only docs. As per coding guidelines: public C API functions must include complete Doxygen documentation describing parameters, return values, and side effects.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@c/include/cuvs/preprocessing/quantize/binary.h` around lines 104 - 139, The three exported APIs cuvsBinaryQuantizerTrain, cuvsBinaryQuantizerTransform, and cuvsBinaryQuantizerTransformWithParams are missing `@return` documentation; update each Doxygen comment to include an `@return` that describes the returned cuvsError_t values (e.g., CUVS_SUCCESS on success and possible error codes on failure such as invalid arguments, allocation failures, or transform/train-specific errors) and note side effects like modifications to output tensors (for cuvsBinaryQuantizerTransform and cuvsBinaryQuantizerTransformWithParams the out DLManagedTensor is written, and for cuvsBinaryQuantizerTrain the quantizer is created/initialized). Ensure the `@return` text is clear and consistent across the three functions and references the same error codes used elsewhere in the API.c/include/cuvs/neighbors/hnsw.h (1)
580-585:⚠️ Potential issue | 🟡 Minor | ⚡ Quick winFix broken deserialize example identifiers.
The sample uses misspelled/invalid symbols (
cuvsHnsWIndexParams_t, malformedcuvsHnswDeserializecall), so copy-paste usage from docs will fail.Suggested doc snippet correction
- * cuvsHnsWIndexParams_t hnsw_params; + * cuvsHnswIndexParams_t hnsw_params; * cuvsHnswIndexParamsCreate(&hnsw_params); * hnsw_params->hierarchy = NONE; * hnsw_index->dtype = index->dtype; - * cuvsHnswDeserialize(res, hnsw_params, "/path/to/index", dim, metric hnsw_index); + * cuvsHnswDeserialize(res, hnsw_params, "/path/to/index", dim, metric, hnsw_index);As per coding guidelines
c/include/cuvs/**/*: “For public C API headers, additionally check: Doxygen documentation for all public functions/classes”.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@c/include/cuvs/neighbors/hnsw.h` around lines 580 - 585, The example doc uses misspelled/invalid symbols; replace the incorrect type/name and fix the deserialize call so it matches the public API. Use cuvsHnswIndexParams_t and cuvsHnswIndexParamsCreate for the params variable (hnsw_params) and call cuvsHnswDeserialize with the correct argument list and commas (e.g., cuvsHnswDeserialize(res, hnsw_params, "/path/to/index", dim, metric, hnsw_index)) so the example compiles and reflects the real function signature.c/include/cuvs/neighbors/cagra.h (1)
768-778:⚠️ Potential issue | 🟡 Minor | ⚡ Quick winUpdate the example to match the exported function name.
The docs example calls
cuvsCagraSerializeHnswlib, but the public API declaration iscuvsCagraSerializeToHnswlib. This is easy to copy incorrectly and fail at compile time.Suggested doc fix
- * cuvsCagraSerializeHnswlib(res, "/path/to/index", index); + * cuvsCagraSerializeToHnswlib(res, "/path/to/index", index);As per coding guidelines
c/include/cuvs/**/*: “For public C API headers, additionally check: Doxygen documentation for all public functions/classes”.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@c/include/cuvs/neighbors/cagra.h` around lines 768 - 778, The example in the Doxygen block uses the wrong function name; update the sample call to use the exported API name cuvsCagraSerializeToHnswlib (e.g., replace cuvsCagraSerializeHnswlib(res, "/path/to/index", index); with cuvsCagraSerializeToHnswlib(res, "/path/to/index", index);) so the docs match the declared function cuvsCagraSerializeToHnswlib.
🧹 Nitpick comments (2)
c/include/cuvs/neighbors/hnsw.h (1)
277-282: ⚡ Quick winAdd dedicated Doxygen for
cuvsHnswFromCagraWithDataset.
cuvsHnswFromCagraWithDatasetis now explicitly exported but has no standalone parameter/return docs, unlike adjacent public APIs.Proposed doc block pattern
+/** + * `@brief` Convert a CAGRA index to an HNSW index using an explicit dataset tensor. + * + * `@param`[in] res cuvsResources_t opaque C handle + * `@param`[in] params cuvsHnswIndexParams_t used to configure conversion + * `@param`[in] cagra_index cuvsCagraIndex_t to convert + * `@param`[out] hnsw_index cuvsHnswIndex_t output HNSW index + * `@param`[in] dataset_tensor DLManagedTensor* dataset used for conversion + * `@return` cuvsError_t + */ CUVS_EXPORT cuvsError_t cuvsHnswFromCagraWithDataset(cuvsResources_t res, cuvsHnswIndexParams_t params, cuvsCagraIndex_t cagra_index, cuvsHnswIndex_t hnsw_index, DLManagedTensor* dataset_tensor);As per coding guidelines
c/include/cuvs/**/*: “For public C API headers, additionally check: Doxygen documentation for all public functions/classes”.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@c/include/cuvs/neighbors/hnsw.h` around lines 277 - 282, Add a dedicated Doxygen comment block for the public function cuvsHnswFromCagraWithDataset documenting the purpose and behaviour, listing each parameter (res, params, cagra_index, hnsw_index, dataset_tensor) with types/semantics (ownership, whether pointers may be NULL, expected tensor layout/shape/dtype for dataset_tensor and whether it is consumed or referenced), describe the return value cuvsError_t and possible error conditions, and include any thread-safety or precondition notes similar to the adjacent public APIs' doc style; place this block immediately above the cuvsHnswFromCagraWithDataset declaration so it appears as standalone Doxygen for the exported symbol.c/include/cuvs/neighbors/ivf_pq.h (1)
272-290: ⚡ Quick winExpand getter docs to full Doxygen format (
@param/@return).The newly exported getter APIs in this block currently have brief comments only; please align them with the fully documented style used elsewhere in this header.
As per coding guidelines
c/include/cuvs/**/*: “For public C API headers, additionally check: Doxygen documentation for all public functions/classes”.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@c/include/cuvs/neighbors/ivf_pq.h` around lines 272 - 290, The comments for the public getter APIs (cuvsIvfPqIndexGetNLists, cuvsIvfPqIndexGetDim, cuvsIvfPqIndexGetSize, cuvsIvfPqIndexGetPqDim, cuvsIvfPqIndexGetPqBits, cuvsIvfPqIndexGetPqLen) must be expanded into full Doxygen style: add an `@brief` line, an `@param` for the input index (describe expected handle) and for the output pointer (describe what value is returned and that it must be non-NULL), and an `@return` describing the cuvsError_t return values (e.g., CUVS_SUCCESS on success, error codes on failure). Update each function’s comment block to follow the same pattern used in other header docs so the public C API is fully documented.
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Inline comments:
In `@c/include/cuvs/neighbors/ivf_flat.h`:
- Around line 159-163: Add full Doxygen comment blocks for the two new public
getters cuvsIvfFlatIndexGetNLists and cuvsIvfFlatIndexGetDim: replace the brief
inline comments with complete blocks that include `@brief` describing purpose,
`@param` for the cuvsIvfFlatIndex_t index and the output pointer (n_lists or dim)
documenting expected ownership/valid pointer requirements, `@return` documenting
the cuvsError_t return values (e.g., CUVS_SUCCESS and possible error codes) and
any side effects (none), and any notes about thread-safety or valid index
states; ensure the function names cuvsIvfFlatIndexGetNLists and
cuvsIvfFlatIndexGetDim appear in the blocks so the API docs pick them up.
In `@cpp/cmake/thirdparty/get_raft.cmake`:
- Around line 9-10: The code currently hardcodes RAFT_FORK and RAFT_PINNED_TAG
in get_raft.cmake (symbols RAFT_FORK and RAFT_PINNED_TAG); remove these fixed
set(...) calls and instead restore upstream defaults or make them configurable:
introduce CMake cache variables (e.g., RAFT_FORK and RAFT_PINNED_TAG via
option/set with CACHE STRING) with defaults pointing to the official upstream
(rapidsai/main or appropriate tag) and allow overrides from the environment or
CI (e.g., honor RAFT_FORK/RAFT_PINNED_TAG CMake cache entries or an env var like
RAFT_FORK_OVERRIDE); alternatively, delete the test pin here and move any
CI-only override into CI config so normal builds resolve upstream. Ensure
references still use the RAFT_FORK and RAFT_PINNED_TAG symbols but no longer
hardcode the test fork/tag in production CMake.
---
Outside diff comments:
In `@c/include/cuvs/neighbors/cagra.h`:
- Around line 768-778: The example in the Doxygen block uses the wrong function
name; update the sample call to use the exported API name
cuvsCagraSerializeToHnswlib (e.g., replace cuvsCagraSerializeHnswlib(res,
"/path/to/index", index); with cuvsCagraSerializeToHnswlib(res,
"/path/to/index", index);) so the docs match the declared function
cuvsCagraSerializeToHnswlib.
In `@c/include/cuvs/neighbors/hnsw.h`:
- Around line 580-585: The example doc uses misspelled/invalid symbols; replace
the incorrect type/name and fix the deserialize call so it matches the public
API. Use cuvsHnswIndexParams_t and cuvsHnswIndexParamsCreate for the params
variable (hnsw_params) and call cuvsHnswDeserialize with the correct argument
list and commas (e.g., cuvsHnswDeserialize(res, hnsw_params, "/path/to/index",
dim, metric, hnsw_index)) so the example compiles and reflects the real function
signature.
In `@c/include/cuvs/neighbors/tiered_index.h`:
- Around line 214-220: Update the Doxygen comment for the exported function
cuvsTieredIndexSearch to include a `@return` section that documents the
cuvsError_t return values (e.g., CUVS_SUCCESS on success, specific CUVS_* error
codes on failure such as invalid arguments, allocation or search errors) and any
side-effects; locate the comment block above the cuvsTieredIndexSearch
declaration and add a concise `@return` description listing possible return codes
and their meaning so the public C API docs are complete.
In `@c/include/cuvs/preprocessing/quantize/binary.h`:
- Around line 104-139: The three exported APIs cuvsBinaryQuantizerTrain,
cuvsBinaryQuantizerTransform, and cuvsBinaryQuantizerTransformWithParams are
missing `@return` documentation; update each Doxygen comment to include an `@return`
that describes the returned cuvsError_t values (e.g., CUVS_SUCCESS on success
and possible error codes on failure such as invalid arguments, allocation
failures, or transform/train-specific errors) and note side effects like
modifications to output tensors (for cuvsBinaryQuantizerTransform and
cuvsBinaryQuantizerTransformWithParams the out DLManagedTensor is written, and
for cuvsBinaryQuantizerTrain the quantizer is created/initialized). Ensure the
`@return` text is clear and consistent across the three functions and references
the same error codes used elsewhere in the API.
In `@c/include/cuvs/preprocessing/quantize/scalar.h`:
- Around line 88-121: The three exported functions cuvsScalarQuantizerTrain,
cuvsScalarQuantizerTransform, and cuvsScalarQuantizerInverseTransform are
missing `@return` (and any side-effect) Doxygen documentation; update each
function comment block to include an `@return` describing the cuvsError_t values
returned (e.g., CUVS_SUCCESS, error codes) and any side effects (e.g.,
device/host memory requirements, in-place modification, ownership of
DLManagedTensor pointers), so the public C API docs are complete and follow the
coding guideline.
In `@c/src/distance/pairwise_distance.cpp`:
- Around line 46-53: The public C API cuvsPairwiseDistance must validate inputs
before any dereference: add checks at the start of cuvsPairwiseDistance (before
calling cuvs::core::translate_exceptions or touching dl_tensor) to reject null
pointers for res, x_tensor, y_tensor, and distances_tensor and return an
appropriate cuvsError_t; also validate tensor dimensions/sizes and
metric/metric_arg combinations (negative dims, mismatched shapes, unsupported
cuvsDistanceType) and return errors early so GPU code never runs with invalid
inputs. Ensure these checks reference the same parameter names (cuvsResources_t
res, DLManagedTensor* x_tensor, DLManagedTensor* y_tensor, DLManagedTensor*
distances_tensor, cuvsDistanceType metric, float metric_arg) and occur prior to
any use of dl_tensor or other members.
- Around line 54-56: The dtype variables are incorrectly read from x_tensor for
all three; change the assignments so y_dt reads from y_tensor->dl_tensor.dtype
and dist_dt reads from dist_tensor->dl_tensor.dtype (leave x_dt as
x_tensor->dl_tensor.dtype) so that mixed-dtype inputs validate and dispatch
correctly for functions/logic that use x_dt, y_dt, and dist_dt in
pairwise_distance.cpp (variables x_tensor, y_tensor, dist_tensor).
In `@cpp/CMakeLists.txt`:
- Around line 931-940: The cuvs-cagra-search target is not receiving the
hidden-visibility defaults, allowing symbol leakage; update the CMake
configuration so cuvs-cagra-search is configured like cuvs_objs (or included in
the same set_target_properties call) to set CXX_STANDARD 20,
CXX_STANDARD_REQUIRED ON, CUDA_STANDARD 20, CUDA_STANDARD_REQUIRED ON,
POSITION_INDEPENDENT_CODE ON, CXX_VISIBILITY_PRESET hidden, and
VISIBILITY_INLINES_HIDDEN ON for the cuvs-cagra-search target (reference the
existing set_target_properties call for cuvs_objs to mirror or extend the same
property list).
---
Nitpick comments:
In `@c/include/cuvs/neighbors/hnsw.h`:
- Around line 277-282: Add a dedicated Doxygen comment block for the public
function cuvsHnswFromCagraWithDataset documenting the purpose and behaviour,
listing each parameter (res, params, cagra_index, hnsw_index, dataset_tensor)
with types/semantics (ownership, whether pointers may be NULL, expected tensor
layout/shape/dtype for dataset_tensor and whether it is consumed or referenced),
describe the return value cuvsError_t and possible error conditions, and include
any thread-safety or precondition notes similar to the adjacent public APIs' doc
style; place this block immediately above the cuvsHnswFromCagraWithDataset
declaration so it appears as standalone Doxygen for the exported symbol.
In `@c/include/cuvs/neighbors/ivf_pq.h`:
- Around line 272-290: The comments for the public getter APIs
(cuvsIvfPqIndexGetNLists, cuvsIvfPqIndexGetDim, cuvsIvfPqIndexGetSize,
cuvsIvfPqIndexGetPqDim, cuvsIvfPqIndexGetPqBits, cuvsIvfPqIndexGetPqLen) must be
expanded into full Doxygen style: add an `@brief` line, an `@param` for the input
index (describe expected handle) and for the output pointer (describe what value
is returned and that it must be non-NULL), and an `@return` describing the
cuvsError_t return values (e.g., CUVS_SUCCESS on success, error codes on
failure). Update each function’s comment block to follow the same pattern used
in other header docs so the public C API is fully documented.
🪄 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: 5ae306dc-74cc-4afc-b172-aa2efd8b1289
📒 Files selected for processing (69)
c/CMakeLists.txtc/include/cuvs/cluster/kmeans.hc/include/cuvs/core/c_api.hc/include/cuvs/core/export.hc/include/cuvs/distance/distance.hc/include/cuvs/distance/pairwise_distance.hc/include/cuvs/neighbors/all_neighbors.hc/include/cuvs/neighbors/brute_force.hc/include/cuvs/neighbors/cagra.hc/include/cuvs/neighbors/common.hc/include/cuvs/neighbors/hnsw.hc/include/cuvs/neighbors/ivf_flat.hc/include/cuvs/neighbors/ivf_pq.hc/include/cuvs/neighbors/mg_cagra.hc/include/cuvs/neighbors/mg_common.hc/include/cuvs/neighbors/mg_ivf_flat.hc/include/cuvs/neighbors/mg_ivf_pq.hc/include/cuvs/neighbors/nn_descent.hc/include/cuvs/neighbors/refine.hc/include/cuvs/neighbors/tiered_index.hc/include/cuvs/neighbors/vamana.hc/include/cuvs/preprocessing/pca.hc/include/cuvs/preprocessing/quantize/binary.hc/include/cuvs/preprocessing/quantize/pq.hc/include/cuvs/preprocessing/quantize/scalar.hc/src/distance/pairwise_distance.cppcpp/CMakeLists.txtcpp/cmake/thirdparty/get_raft.cmakecpp/doxygen/Doxyfilecpp/include/cuvs/cluster/agglomerative.hppcpp/include/cuvs/cluster/kmeans.hppcpp/include/cuvs/cluster/spectral.hppcpp/include/cuvs/core/bitmap.hppcpp/include/cuvs/core/bitset.hppcpp/include/cuvs/core/export.hppcpp/include/cuvs/distance/distance.hppcpp/include/cuvs/distance/grammian.hppcpp/include/cuvs/distance/kde.hppcpp/include/cuvs/neighbors/all_neighbors.hppcpp/include/cuvs/neighbors/ball_cover.hppcpp/include/cuvs/neighbors/brute_force.hppcpp/include/cuvs/neighbors/cagra.hppcpp/include/cuvs/neighbors/common.hppcpp/include/cuvs/neighbors/composite/index.hppcpp/include/cuvs/neighbors/dynamic_batching.hppcpp/include/cuvs/neighbors/epsilon_neighborhood.hppcpp/include/cuvs/neighbors/hnsw.hppcpp/include/cuvs/neighbors/ivf_flat.hppcpp/include/cuvs/neighbors/ivf_pq.hppcpp/include/cuvs/neighbors/knn_merge_parts.hppcpp/include/cuvs/neighbors/nn_descent.hppcpp/include/cuvs/neighbors/refine.hppcpp/include/cuvs/neighbors/scann.hppcpp/include/cuvs/neighbors/tiered_index.hppcpp/include/cuvs/neighbors/vamana.hppcpp/include/cuvs/preprocessing/pca.hppcpp/include/cuvs/preprocessing/quantize/binary.hppcpp/include/cuvs/preprocessing/quantize/pq.hppcpp/include/cuvs/preprocessing/quantize/scalar.hppcpp/include/cuvs/preprocessing/spectral_embedding.hppcpp/include/cuvs/selection/select_k.hppcpp/include/cuvs/stats/silhouette_score.hppcpp/include/cuvs/stats/trustworthiness_score.hppcpp/include/cuvs/util/cutlass_utils.hppcpp/include/cuvs/util/file_io.hppcpp/include/cuvs/util/host_memory.hppcpp/src/distance/kde.cucpp/src/neighbors/epsilon_neighborhood.cucpp/src/neighbors/ivf_flat_index.cpp
robertmaynard
approved these changes
May 1, 2026
The ABI checker scans all.h to verify all C headers are listed. export.h was missing after the rename from export.hpp.
PREDEFINED alone is insufficient because Doxygen parses the source
#define in export.h/hpp and overrides the PREDEFINED value, expanding
CUVS_EXPORT to __attribute__((visibility("default"))) in the XML
output. Excluding the files prevents the redefinition, so the
PREDEFINED CUVS_EXPORT= (expand to nothing) takes effect.
Contributor
Author
|
CI on b3d67e9 shows a passing run (up to the usual flaky failures) on a run containing the raft changes. |
This reverts commit 0d94e7d.
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
2 participants
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.
This PR adds symbol visibility controls to raft to avoid exporting weak symbols that it shouldn't.
Contributes to https://github.com/rapidsai/build-infra/issues/53