This will be technically difficult. It is possible to access memory from different processes using shared memory, but implementing that in minimap2 will be nontrivial.

Can you use multiple threads? You can create one mappy.Aligner object and call Aligner.map on different threads like (I have not tested this part, though):

aligner = mappy.Aligner(fn_index)
# then in each thread
thr_buf = mappy.ThreadBuffer()
for hit in aligner.map(seq, buf=thr_buf):
    ...

I figured this might be quite an undertaking if it were not already built into minimap2/mappy. I am using python's multiprocess module right now, but slimming the memory footprint down for large genomes might warrant a switch to threading.

I will try some tests to see if this is feasible. Thanks or the pointer and use case for mappy.ThreadBuffer.

I have tested the interface and it works quite well indeed. I will note that mixing multiprocessing with multithreading was quite a headache, but it is indeed possible. For others that might be trying this, I had to open all multiprocess objects (Pipes in my case) and start all processes before opening any threading objects.

While it would be nice to have a version that could share state across processes I think this is a sufficient workaround and so I would consider this issue resolved.