def test_sieve_multiple(self, context: Context) -> None:
"""Test proper implementation of the filtering mechanism."""
context.project.runtime_environment.cuda_version = "10.0"
source = Source("https://pypi.org/simple")
pv_1 = PackageVersion(
name="tensorflow-gpu",
version="==1.12.0",
develop=False,
index=source,
)
pv_2 = PackageVersion(
name="tensorflow",
version="==2.0.0",
develop=False,
index=source,
)
pv_3 = PackageVersion(
name="tensorflow",
version="==1.13.0",
develop=False,
index=source,
)
unit = TensorFlowCUDASieve()
with unit.assigned_context(context):
unit.pre_run()
result = list(unit.run((pv for pv in (pv_1, pv_2, pv_3))))
assert len(result) == 1
assert result[0] == pv_2
def test_unknown_tensorflow(self, context: Context, package_name: str) -> None:
"""Test not discarding if an unknown TensorFlow release is spotted."""
context.project.runtime_environment.cuda_version = "10.0"
package_version = PackageVersion(
name=package_name,
version="==42.30.03",
develop=False,
index=Source("https://pypi.org/simple"),
)
unit = TensorFlowCUDASieve()
with unit.assigned_context(context):
unit.pre_run()
result = list(unit.run((pv for pv in (package_version,))))
assert len(result) == 1
assert result[0] is package_version, "The pipeline unit should keep the unknown TensorFlow release"
def test_run_no_yield(self, context: Context, pv: Tuple[str, str], cuda_version: str) -> None:
"""Test discarding packages that do not conform to the support matrix.
See the official docs for listing:
https://www.tensorflow.org/install/source#gpu
"""
context.project.runtime_environment.cuda_version = cuda_version
package_version = PackageVersion(
name=pv[0],
version=f"=={pv[1]}",
develop=False,
index=Source("https://pypi.org/simple"),
)
unit = TensorFlowCUDASieve()
with unit.assigned_context(context):
unit.pre_run()
result = list(unit.run((pv for pv in (package_version,))))
assert len(result) == 0
def test_run_yield(self, context: Context, package_name: str, package_version: str, cuda_version: str) -> None:
"""Test packages the pipeline unit yields respecting CUDA version used.
See the official docs for listing:
https://www.tensorflow.org/install/source#gpu
"""
context.project.runtime_environment.cuda_version = cuda_version
package_version = PackageVersion(
name=package_name,
version=f"=={package_version}",
develop=False,
index=Source("https://pypi.org/simple"),
)
unit = TensorFlowCUDASieve()
with unit.assigned_context(context):
unit.pre_run()
result = list(unit.run((pv for pv in (package_version,))))
assert len(result) == 1
assert result[0] is package_version