Franck Charras

~~You're trying to parallelize a task on chunks over all available cpu processes, and you want that each task created this way also run some function in parallel over all...

Second try: now my understanding is that the issue you mention is purely because of return types, you can add this instruction just before returning `paths`: ``` paths = (path...

AFAIK `Parallel` does not use `Logger` inheritance, relevant threads https://github.com/joblib/joblib/issues/1483 https://github.com/joblib/joblib/pull/1494 . In your example: ``` from prefect import flow, task from joblib import Parallel, delayed def sqrt(x): print(x**0.5) @task...

Regarding the second and third points, in fact it's not an issue because `_ensure_native_byte_order` is only used in `joblib.Parallel` calls when auto-memmapping is triggered. Byteorder for small arrays has always...

@alexisthual do you want to try this branch on your usecase and confirm that it fixes it ?

Currently there isn't a way to get those kind of insights built in joblib, but I agree it can be very useful to reason about efficiency of multiprocessing. I wonder...

Thanks for the PR @androids-electric-sheep , is it reasonable to say that there might still be differences in the code that is returned ? then shipping this into a new...

Thanks for the PR @androids-electric-sheep, can you give a bit more context about what is the current behavior, does it fail if it's not an int ? (what about n_jobs=3.0,...

Did a bit of digging, the `filename` attribute is lost when going through [this function](https://github.com/joblib/joblib/blame/main/joblib/numpy_pickle_utils.py#L66-L73). The bug was most likely added there https://github.com/joblib/joblib/commit/0569c89fcc3015734af2fb5e58e88037deccb4dd

Not related to https://github.com/joblib/joblib/commit/0569c89fcc3015734af2fb5e58e88037deccb4dd Can be fixed by passing `inplace=True` to `byteswap`, indeed it makes sense that we're loosing the filename attribute if we're making a copy. What is even...