ImportError: Import by filename is not supported when unpickleing

[Steven D'Aprano]

On Thursday 28 July 2016 12:39, Larry Martell wrote:

> I have an object of type Target:
> 
> (Pdb) type(target)
> <class 'workitem.Target'>
> 
> And I pickle it like this:
> 
> (Pdb) type(pickle.dumps(target))
> <type 'str'>
> 
> And then it looks like this:
> 
> (Pdb) pickle.dumps(target)
> 
"ccopy_reg\n_reconstructor\np0\n(cworkitem\nTarget\np1\nc__builtin__\nobject\np2\nNtp3\nRp4\n(dp5\nS'histologySections'\np6\n(lp7\nsS'registrationTransforms'\np8\nV\np9\nsS'valueMaps'\np10\n(dp11\nsS'targetID'\np12\nVRight-
CarotidArtery\np13\nsS'targetInitializer'\np14\ng9\nsS'regions'\np15\n(dp16\nsS'bodySite'\np17\nVRightCarotid\np18\nsS'targetLocalFolderName'\np19\nV21597135/wi-54976537/Right-
CarotidArtery\np20\nsS'readingsLocalFileName'\np21\ng9\nsS'probabilityMaps'\np22\n(dp23\nsS'targetPath'\np24\ng9\nsb."

Gag me with a spoon!

What happens if you do 

s = pickle.dumps(target)
pickle.loads(s)

Do you get the same error?


> And I pass it into a subprocess.Popen call. Then in the program that
> is called, it comes in as a string:
> 
> (Pdb) type(args.target)
> <type 'str'>
> 
> With the same content:
> 
> (Pdb) args.target
> 
"ccopy_reg\n_reconstructor\np0\n(cworkitem\nTarget\np1\nc__builtin__\nobject\np2\nNtp3\nRp4\n(dp5\nS'histologySections'\np6\n(lp7\nsS'registrationTransforms'\np8\nV\np9\nsS'valueMaps'\np10\n(dp11\nsS'targetID'\np12\nVRight-
CarotidArtery\np13\nsS'targetInitializer'\np14\ng9\nsS'regions'\np15\n(dp16\nsS'bodySite'\np17\nVRightCarotid\np18\nsS'targetLocalFolderName'\np19\nV21597135/wi-54976537/Right-
CarotidArtery\np20\nsS'readingsLocalFileName'\np21\ng9\nsS'probabilityMaps'\np22\n(dp23\nsS'targetPath'\np24\ng9\nsb."

Are you sure it's the same content, and not just the same to the naked eye?

> And when I unpickle it I get the error:
> 
> (Pdb) pickle.loads(args.target)
> ***ImportError: Import by filename is not supported.

Does the target process already have workitem imported? What happens if you 
import workitem first?




By the way, just in case this is relevant to you... pickle is an insecure 
format. If your target process is running in a context where it can receive 
pickles from untrusted clients, you're vulnerable to them running arbitrary 
code on your machine.

If you are trying to do a remote procedure call, rather than invent your own, 
you should use a reliable, trusted library. I've used both rpyc and Pyro and 
can recommend them both:

https://pypi.python.org/pypi/rpyc/
https://pypi.python.org/pypi/Pyro4/


-- 
Steve