Thomas Moreau

I think this feature is indeed a nice one, and would prove very useful in `joblib.Memory`, in particular when interacting with SQL databases or changing environments. A classical hack to...

Ok if you like it then, I did a small PR to try to implement my solution. Let me know what you think of it. In particular, I am not...

From what I understood, the problem is that you have 2 tasks, the first one is long (10s) while the second error immediatly, you need 10s to get the error:...

Yes, because in this case, we call terminate on the `ThreadPool` but this is non-blocking as we do not terminate the threads nor join them. The computations continues in background...

Do you have a minimal reproducible working example, so I can better understand what you are trying to do? Also, when you say it is not working, does it throws...

This is weird. On my computer, I get: 

Actually, when you instantiate `multiprocessing.Queue`, under the hood this sets the default context. So you should actually move `set_start_method` ```python import multiprocessing import sklearn multiprocessing.set_start_method('loky') _proc_log_queue = multiprocessing.Queue() def main():...

This is to prevent you from creating a `multiprocessing.Queue` with the fork context and then launching processes with `loky` context. This is also what causes the large traceback when you...

Hello, thanks for reporting this issue. The `joblib.hash` function is based on the serialization of the underlying object, ie the use of `pickle.dumps`. The string of bytes produced is then...

Thx for the review @jjerphan ! I did update this a bit. I will try to finish this by adding some examples and a bit of doc.