We had a problem running code that downloads files from github when porting from python2.7 to python3.[3-7]. Not sure if a bug or not.

With the given code, in python3 a file downloaded in chunks will contain the size of chunks when using the undocumented fp from urlopen(...).fp. In python2, only the chunk payload would make it into the file.

We assume that we can just use the urlopen response directly as a fix (without '.fp'), but though it might still be nice to report the difference.

Short code:
resp = urlopen('http://someurl')
fhand = os.fdopen(fdesc, "wb")
shutil.copyfileobj(resp.fp, fhand)   # using .fp here is the dodgy part
fhand.close()

The attached script demonstrates the difference:

$ python --version
Python 2.7.15rc1
$ python urllib_issue.py 
127.0.0.1 - - [12/Sep/2018 01:27:28] "GET /downloads/1.0.tar.gz HTTP/1.1" 200 -

$ python3 --version
Python 3.6.5
$ python3 urllib_issue.py 
127.0.0.1 - - [12/Sep/2018 01:27:37] "GET /downloads/1.0.tar.gz HTTP/1.1" 200 -
Traceback (most recent call last):
  File "urllib_issue.py", line 87, in <module>
    assert data == FILE_CONTENT, '%s, %s'%(len(FILE_CONTENT), len(data))
AssertionError: 100000, 100493
!!! BASH reports ERROR: shell returned 1

Hello @tkruse, I have made some research and found that when using the Chunked transfer encoding [1], each chunk is preceded by its size in bytes, something that really happen if you check the content of one downloaded file from the example you provided [2]. So far, I would say that this is not a bug, it is just how the transfer encoding works.

[1]: https://en.wikipedia.org/wiki/Chunked_transfer_encoding
[2]: https://gist.github.com/jaswdr/95b2adc519d986c00b17f6572d470f2a