How to access a wrapped function from a Delayed object?

[andersy005]

I'm trying to retrieve a wrapped function of either a partial or curried function, and so far I can't seem to figure out where dask stores information about the wrapped function when using dask.delayed.

Here's a code snippet showing what I'm trying to achieve. I included an example with toolz.curry to demonstrate how I am able to retrieve the wrapper function:

In [37]: import dask

In [38]: import toolz

In [39]: def mul(a, b):
    ...:     return a * b
    ...: 

In [40]: delayed_mul = dask.delayed(mul)

In [41]: curried_mul = toolz.curry(mul)

toolz.curry nests the wrapped function in a .func attribute

In [44]: curried_mul.func
Out[44]: <function __main__.mul(a, b)>
In [48]: curried_mul.func.__name__
Out[48]: 'mul'

I had hopes that dask was doing the same. Well, nope :(

In [50]: delayed_mul.func
Out[50]: Delayed('getattr-469ea79569437ee253de5413f145f0dd')

In [51]: delayed_mul.func.__name__
---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-51-3fd5f0aa9560> in <module>
----> 1 delayed_mul.func.__name__

~/opt/miniconda3/envs/xpersist-dev/lib/python3.9/site-packages/dask/delayed.py in __getattr__(self, attr)
    661         if attr == "dtype" and self._attr == "dtype":
    662             raise AttributeError("Attribute %s not found" % attr)
--> 663         return super().__getattr__(attr)
    664 
    665     @property

~/opt/miniconda3/envs/xpersist-dev/lib/python3.9/site-packages/dask/delayed.py in __getattr__(self, attr)
    538     def __getattr__(self, attr):
    539         if attr.startswith("_"):
--> 540             raise AttributeError("Attribute {0} not found".format(attr))
    541 
    542         if attr == "visualise":

AttributeError: Attribute __name__ not found

[jrbourbeau]

You should be able to inspect the task graph for the delayed object to get the wrapped function. Something along the lines of

In [1]: import dask

In [2]: def mul(a, b):
   ...:     return a * b
   ...:

In [3]: delayed_mul = dask.delayed(mul)

In [4]: delayed_mul.dask[delayed_mul.key]
Out[4]: <function __main__.mul(a, b)>

[andersy005]

Aha!! that's it! Thank you, @jrbourbeau!