Message271573
Here is a new patch with tests on constructor.
The patch on the feature is quite different: instead of testing for None socket, I choose to delay the != SOCK_STREAM check in the later condition, when we know we treat a socket.
Tests include different constructor forms: with a given socket, with a fileno (didn't work either, before this patch) and without socket nor file descriptor (as in my original test).
I don't have sufficient background to judge if tests will work on all platform (eg. fileno and windows, ...).
Here is the interesting diff of the tests on SSL (before/after applying the patch on the feature):
32c32
< test_constructor (__main__.BasicSocketTests) ... ERROR
---
> test_constructor (__main__.BasicSocketTests) ... ok
519,528d518
< ERROR: test_constructor (__main__.BasicSocketTests)
< ----------------------------------------------------------------------
< Traceback (most recent call last):
< File "test_ssl.py", line 149, in test_constructor
< ss = ssl.SSLSocket()
< File "/usr/lib/python3.4/ssl.py", line 545, in __init__
< if sock.getsockopt(SOL_SOCKET, SO_TYPE) != SOCK_STREAM:
< AttributeError: 'NoneType' object has no attribute 'getsockopt'
<
< ====================================================================== |
|
Date |
User |
Action |
Args |
2016-07-28 17:34:23 | nemunaire | set | recipients:
+ nemunaire, janssen, pitrou, vstinner, giampaolo.rodola, christian.heimes, alex, dstufft, abarry |
2016-07-28 17:34:23 | nemunaire | set | messageid: <1469727263.3.0.760638749708.issue27629@psf.upfronthosting.co.za> |
2016-07-28 17:34:23 | nemunaire | link | issue27629 messages |
2016-07-28 17:34:23 | nemunaire | create | |
|