dask.array.core.map_blocks mis-handles kwargs in task name #8632
Description
Minimal Complete Verifiable Example:
import dask
import dask.array as da
import distributed
import numpy as np
def mul(x, y=None, *, ham=None):
if y is None:
return x * 3
return x * 2
def main():
with distributed.Client() as c:
x = da.arange(6, chunks=3)
d1 = x.map_blocks(mul, ham=1, dtype=np.int_)
d2 = x.map_blocks(mul, {"ham": 1}, dtype=np.int_)
print(dask.compute(d1, d2)) # (array([ 0, 3, 6, 9, 12, 15]), array([ 0, 3, 6, 9, 12, 15]))
print(dask.compute(d2, d1)) # (array([ 0, 2, 4, 6, 8, 10]), array([ 0, 2, 4, 6, 8, 10]))
if __name__ == "__main__":
main()this prints:
(array([ 0, 3, 6, 9, 12, 15]), array([ 0, 3, 6, 9, 12, 15]))
(array([ 0, 2, 4, 6, 8, 10]), array([ 0, 2, 4, 6, 8, 10]))
but I expected
(array([ 0, 3, 6, 9, 12, 15]), array([ 0, 2, 4, 6, 8, 10]))
(array([ 0, 2, 4, 6, 8, 10]), array([ 0, 3, 6, 9, 12, 15]))
Anything else we need to know?:
name is calculated with
Line 678 in 7446308
however kwargs are treated the same as a dict in the last arg
Lines 857 to 858 in 7446308
eg:
>>> import dask.base
>>> dask.base.tokenize("hello", ham="eggs")
'bdafbbae310e8f3f583f8002342c8ebb'
>>> dask.base.tokenize("hello", {"ham":"eggs"})
'bdafbbae310e8f3f583f8002342c8ebb'