Fall back to serial mode does not work in current version

I tried to use joblib with Chaquopy, but Android does not have a functioning sem_open implementation, so I received the following warning as expected:

_multiprocessing_helpers.py:38: UserWarning: This platform lacks a functioning sem_open implementation, therefore, the required synchronization primitives needed will not function, see issue 3770.. joblib will operate in serial mode

With joblib 0.11, the import still succeeds. But with joblib 0.13.0 it fails, because joblib unconditionally imports loky, which unconditionally imports multiprocessing.synchronize.

Traceback from https://github.com/chaquo/chaquopy/issues/21#issuecomment-448568723 (edited for clarity):

 File "/android_asset/chaquopy/requirements-common.zip/joblib/__init__.py", line 119, in <module>
    from .parallel import Parallel
  File "/android_asset/chaquopy/requirements-common.zip/joblib/parallel.py", line 33, in <module>
    from .externals import loky
  File "/android_asset/chaquopy/requirements-common.zip/joblib/externals/loky/__init__.py", line 11, in <module>
    from .backend.context import cpu_count
  File "/android_asset/chaquopy/requirements-common.zip/joblib/externals/loky/backend/__init__.py", line 13, in <module>
    from multiprocessing import synchronize
  File "stdlib/multiprocessing/synchronize.py", line 34, in <module>
ImportError: This platform lacks a functioning sem_open implementation, therefore, the required synchronization primitives needed will not function, see issue 3770.

[mhsmith]

Running this workaround before the import will allow it to succeed:

import _multiprocessing
_multiprocessing.sem_unlink = None

Of course, process-based parallelism will still not work if the underlying platform doesn’t support it, so you’ll have to use the parallel_backend context manager to use thread-based parallelism instead.

[mhsmith]

Chaquopy 8.0.1 now has a workaround for this problem. Process-based parallelism is still unavailable, but joblib can at least be imported, and then you can use the parallel_backend context manager to use thread-based parallelism instead.

To upgrade, edit your app’s top-level build.gradle file and change the version number of com.chaquo.python:gradle.