Adapt import loading for Windows compatibility

[stefanhahmann]

This draft PR is a suggestion to fix an issue with appose in Windows. It is very similar to #5 and probably again not the ideal solution.

I am using this env:

name: cellpose3
channels:
  - nvidia
  - pytorch
  - conda-forge
channel_priority: strict
dependencies:
  - python=3.10
  - cellpose==3.1.1.2
  - appose==0.7.0
  - pytorch
  - pytorch-cuda
  - numpy

Within this env I run:

micromamba run -p C:/Users/xxxx/.local/share/appose/cellpose3 C:/Users/xxxx/.local/share/appose/cellpose3/python.exe -c "import appose.python_worker; appose.python_worker.main()"

and then interactively inputing this task to stdin

{"task":"c388f270-d6be-48df-9e02-e42b48cab98b","requestType":"EXECUTE","inputs":{},"script":"import numpy\n"}

This results in a perfectly working process:

[SERVICE-0] {"task":"c388f270-d6be-48df-9e02-e42b48cab98b","requestType":"EXECUTE","inputs":{},"script":"import numpy\n"}
[SERVICE-0] {"task":"c388f270-d6be-48df-9e02-e42b48cab98b","responseType":"LAUNCH"}
[SERVICE-0] {"task":"c388f270-d6be-48df-9e02-e42b48cab98b","responseType":"COMPLETION","outputs":{}}

However, when I try to run this from a Java process similar to the code example (https://github.com/mastodon-sc/mastodon-deep-lineage/blob/appose/src/test/java/org/mastodon/mamut/appose/ApposeCellpose3Example.java), I only get this response and then the whole process is stuck in a dead lock:

[SERVICE-0] {"task":"c388f270-d6be-48df-9e02-e42b48cab98b","requestType":"EXECUTE","inputs":{},"script":"import numpy\n"}
[SERVICE-0] {"task":"c388f270-d6be-48df-9e02-e42b48cab98b","responseType":"LAUNCH"}

The code added in this PR fixes this situation. It seems that loading imports like (e.g.) numpy can lead to deadlocks when being loaded in (multiple) threads.

I am again not sure, if the proposed fix is a good one however. Since the problem occurs only on Windows, one could think about adding if sys.platform == 'win32': or something like this to only perform the extra code on Windows. Also, one could think about making this optional and controlable from outside the python_worker by perhaps only performing the new code, if some global variable is set.

@ctrueden curious to learn your opinion about this.

[stefanhahmann]

I could just verify that Appose 0.8.0 (java) together with Appose 0.7.1 (python) fixes this Windows hanging bug in https://github.com/mastodon-sc/mastodon-deep-lineage/blob/dff692442470e23a224364e3a0f7ffb0862fdb01/src/test/java/org/mastodon/mamut/appose/ApposeCellpose3Example.java

All I needed to do is adding the numpy import to the init() method: Service python = env.python().init( "import numpy" )