Update to Cython 3.0.0 (#258)

This PR bumps the pinning and fixes the new warnings that appeared

Authors:
  - Vyas Ramasubramani (https://github.com/vyasr)

Approvers:
  - Ray Douglass (https://github.com/raydouglass)
  - Alexey Kamenev (https://github.com/Alexey-Kamenev)
  - Vukasin Milovanovic (https://github.com/vuule)
  - https://github.com/jakirkham

URL: #258

conda/environments/all_cuda-118_arch-x86_64.yaml

@@ -14,7 +14,7 @@ dependencies:
 - cudf==23.10.*
 - cupy>=12.0.0
 - cxx-compiler
-- cython>=0.29,<0.30
+- cython>=3.0.0
 - dask>=2022.05.2
 - distributed>=2022.05.2
 - doxygen=1.8.20

conda/environments/all_cuda-120_arch-x86_64.yaml

@@ -14,7 +14,7 @@ dependencies:
 - cudf==23.10.*
 - cupy>=12.0.0
 - cxx-compiler
-- cython>=0.29,<0.30
+- cython>=3.0.0
 - dask>=2022.05.2
 - distributed>=2022.05.2
 - doxygen=1.8.20

conda/recipes/kvikio/meta.yaml

@@ -57,7 +57,7 @@ requirements:
     - python
     - setuptools
     - pip
-    - cython >=0.29,<0.30
+    - cython >=3.0.0
     {% if cuda_major == "11" %}
     - cudatoolkit
     {% endif %}

dependencies.yaml

@@ -92,7 +92,7 @@ dependencies:
       - output_types: [conda, requirements, pyproject]
         packages:
           - cmake>=3.26.4
-          - cython>=0.29,<0.30
+          - cython>=3.0.0
           - ninja
           - scikit-build>=0.13.1
       - output_types: conda

legate/pyproject.toml

@@ -5,7 +5,7 @@
 build-backend = "setuptools.build_meta"
 requires = [
     "cmake>=3.26.4",
-    "cython>=0.29,<0.30",
+    "cython>=3.0.0",
     "ninja",
     "scikit-build>=0.13.1",
     "setuptools",

python/kvikio/_lib/libkvikio.pyx

@@ -5,6 +5,7 @@
 # cython: language_level=3
 
 import pathlib
+from typing import Optional
 
 from libc.stdint cimport uintptr_t
 from libcpp.utility cimport move, pair
@@ -49,7 +50,7 @@ def memory_deregister(buf) -> None:
     kvikio_cxx_api.memory_deregister(<void*>arr.ptr)
 
 
-def compat_mode() -> int:
+def compat_mode() -> bool:
     return kvikio_cxx_api.compat_mode()
 
 
@@ -124,7 +125,7 @@ cdef class CuFile:
     def open_flags(self) -> int:
         return self._handle.fd_open_flags()
 
-    def pread(self, buf, size: int, file_offset: int, task_size) -> IOFuture:
+    def pread(self, buf, size: Optional[int], file_offset: int, task_size) -> IOFuture:
         cdef pair[uintptr_t, size_t] info = _parse_buffer(buf, size, True)
         return _wrap_io_future(
             self._handle.pread(
@@ -135,7 +136,7 @@ cdef class CuFile:
             )
         )
 
-    def pwrite(self, buf, size: int, file_offset: int, task_size) -> IOFuture:
+    def pwrite(self, buf, size: Optional[int], file_offset: int, task_size) -> IOFuture:
         cdef pair[uintptr_t, size_t] info = _parse_buffer(buf, size, True)
         return _wrap_io_future(
             self._handle.pwrite(
@@ -146,7 +147,7 @@ cdef class CuFile:
             )
         )
 
-    def read(self, buf, size: int, file_offset: int, dev_offset: int) -> int:
+    def read(self, buf, size: Optional[int], file_offset: int, dev_offset: int) -> int:
         cdef pair[uintptr_t, size_t] info = _parse_buffer(buf, size, False)
         return self._handle.read(
             <void*>info.first,
@@ -155,7 +156,7 @@ cdef class CuFile:
             dev_offset,
         )
 
-    def write(self, buf, size: int, file_offset: int, dev_offset: int) -> int:
+    def write(self, buf, size: Optional[int], file_offset: int, dev_offset: int) -> int:
         cdef pair[uintptr_t, size_t] info = _parse_buffer(buf, size, False)
         return self._handle.write(
             <void*>info.first,

python/kvikio/_lib/libnvcomp_ll.pyx

@@ -1,6 +1,8 @@
 # Copyright (c) 2023, NVIDIA CORPORATION. All rights reserved.
 # See file LICENSE for terms.
 
+from __future__ import annotations
+
 from abc import ABC, abstractmethod
 from enum import IntEnum
 
@@ -84,7 +86,7 @@ class nvCompBatchAlgorithm(ABC):
         self,
         size_t batch_size,
         size_t max_uncompressed_chunk_bytes,
-    ) -> (nvcompStatus_t, size_t):
+    ) -> tuple[nvcompStatus_t, size_t]:
         """Algorithm-specific implementation."""
         ...
 
@@ -409,7 +411,7 @@ class nvCompBatchAlgorithmLZ4(nvCompBatchAlgorithm):
         self,
         size_t batch_size,
         size_t max_uncompressed_chunk_bytes,
-    ) -> (nvcompStatus_t, size_t):
+    ) -> tuple[nvcompStatus_t, size_t]:
         cdef size_t temp_bytes = 0
 
         err = nvcompBatchedLZ4CompressGetTempSize(
@@ -671,7 +673,7 @@ class nvCompBatchAlgorithmGdeflate(nvCompBatchAlgorithm):
         self,
         size_t batch_size,
         size_t max_uncompressed_chunk_bytes,
-    ) -> (nvcompStatus_t, size_t):
+    ) -> tuple[nvcompStatus_t, size_t]:
         cdef size_t temp_bytes = 0
 
         err = nvcompBatchedGdeflateCompressGetTempSize(
@@ -806,7 +808,7 @@ class nvCompBatchAlgorithmZstd(nvCompBatchAlgorithm):
         self,
         size_t batch_size,
         size_t max_uncompressed_chunk_bytes,
-    ) -> (nvcompStatus_t, size_t):
+    ) -> tuple[nvcompStatus_t, size_t]:
         cdef size_t temp_bytes = 0
 
         err = nvcompBatchedZstdCompressGetTempSize(
@@ -941,7 +943,7 @@ class nvCompBatchAlgorithmSnappy(nvCompBatchAlgorithm):
         self,
         size_t batch_size,
         size_t max_uncompressed_chunk_bytes,
-    ) -> (nvcompStatus_t, size_t):
+    ) -> tuple[nvcompStatus_t, size_t]:
         cdef size_t temp_bytes = 0
 
         err = nvcompBatchedSnappyCompressGetTempSize(

python/kvikio/_lib/nvcomp_cxx_api.pxd

@@ -73,17 +73,23 @@ cdef extern from "nvcomp/nvcompManager.hpp" namespace 'nvcomp':
             size_t uncompressed_buffer_size) except +
         nvcompStatus_t* get_status() const
         CompressionConfig(CompressionConfig& other)
-        CompressionConfig& operator=(CompressionConfig&& other) except +
         CompressionConfig& operator=(const CompressionConfig& other) except +
+        # Commented as Cython doesn't support rvalues, but a user can call
+        # `move` with the existing operator and generate correct C++ code
+        # xref: https://github.com/cython/cython/issues/1445
+        # CompressionConfig& operator=(CompressionConfig&& other) except +
 
     cdef cppclass DecompressionConfig "nvcomp::DecompressionConfig":
         size_t decomp_data_size
         uint32_t num_chunks
         DecompressionConfig(PinnedPtrPool[nvcompStatus_t]& pool) except +
         nvcompStatus_t* get_status() const
         DecompressionConfig(DecompressionConfig& other)
-        DecompressionConfig& operator=(DecompressionConfig&& other) except +
         DecompressionConfig& operator=(const DecompressionConfig& other) except +
+        # Commented as Cython doesn't support rvalues, but a user can call
+        # `move` with the existing operator and generate correct C++ code
+        # xref: https://github.com/cython/cython/issues/1445
+        # DecompressionConfig& operator=(DecompressionConfig&& other) except +
 
     cdef cppclass nvcompManagerBase "nvcomp::nvcompManagerBase":
         CompressionConfig configure_compression(

python/pyproject.toml

@@ -5,7 +5,7 @@
 build-backend = "setuptools.build_meta"
 requires = [
     "cmake>=3.26.4",
-    "cython>=0.29,<0.30",
+    "cython>=3.0.0",
     "ninja",
     "scikit-build>=0.13.1",
     "setuptools",