blob: 4dcc5412c7e90b7f36e147235e046593ee320936 (
plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
|
[[!comment format=mdwn
username="http://joeyh.name/"
ip="4.154.0.63"
subject="comment 1"
date="2013-08-26T19:51:48Z"
content="""
So this is a crash in the gnutls library used for XMPP. Someone else using OSX reported a similar crash to me by email, reproducible reliably by setting up xmpp with google. This is great debugging info:
<pre>
Thread 3 Crashed:
0 libsystem_c.dylib 0x00007fff86ae0bf9 pthread_mutex_lock + 20
1 H 0x000000010e9fd29f gnutls_system_mutex_lock + 12
2 H 0x000000010ea7fa29 wrap_nettle_rnd_refresh + 20
3 H 0x000000010e9fee89 gnutls_deinit + 42
4 git-annex 0x000000010caf0a3a 0x10b5e6000 + 22063674
</pre>
Looks like `gnutls_deinit` was called and it crashed there, inside pthread code. So I think git-annex has already managed to connect to the XMPP server (to test it) and the cleanup is where it's crashing.
This is the second time I have seen a gnutls-related crash using XMPP. The other one was tracked down by John Millikin to a resource allocation bug in haskell-gnutls and fixed.
I have written a test case that reproduces the problem -- just forking a dozen threads that each try to connect to the google xmpp server and then close the connection. After a dozen or so succeed, one will reliably cause a segfault. Forwarded this test case to John.
"""]]
|
