monkeylater

Run selected functions later, in background threads.

monkeylater is an experimental Python library for making explicit function calls asynchronous by monkey-patching them. It is aimed at experiment scripts and logging/output paths where writes such as numpy.save or cv2.imwrite should not block the main loop.

Use global patching for scripts and experiment code where you want selected functions to run in the background across a wider section of your program.

import numpy as np
import monkeylater

monkeylater.patch_global((np, "save"))

larger_array = np.random.rand(5000, 5000)
np.save("large_array.npy", larger_array)

# ...
# The save runs in the background, and the main thread can continue doing other work.

Patch multiple functions by passing more explicit (owner, "attribute") tuples:

monkeylater.patch_global((np, "save"), (cv2, "imwrite"))

Use scoped patching when you want a safer, temporary patch around a specific block. The original function is restored when the with block exits, and pending background work is flushed before exit completes.

import numpy as np
import monkeylater

arr = np.arange(9)

with monkeylater.patch((np, "save")):
    np.save("array.npy", arr)

Global mode automatically flushes pending work at process exit. You only need flush() or restore_all() when you want control earlier.

Call flush() when you need outputs to exist now, want exceptions now, or are writing tests:

monkeylater.flush()

Call restore_all() in long-running processes when you want to stop global patching before process exit:

monkeylater.restore_all()

Patched functions return concurrent.futures.Future objects instead of their original return values. Keep the future when you want to wait for one call or inspect its result directly.

with monkeylater.patch((np, "save")):
    save_job = np.save("array.npy", np.arange(9))

save_job.result()

Exceptions raised by the original function are deferred until Future.result(), flush(), or scoped patch exit.

monkeylater does not copy arguments. If you pass a mutable object and modify it immediately after calling the patched function, the background function may see the modified object.

Copy explicitly when needed:

with monkeylater.patch((np, "save")):
    np.save("array.npy", arr.copy())