[master 00/17] Improve child process starting and management
dshea at redhat.com
Tue Sep 23 19:54:55 UTC 2014
On 09/23/2014 03:48 PM, Colin Walters wrote:
> On Tue, Sep 23, 2014, at 01:28 PM, David Shea wrote:
>> On 09/23/2014 12:17 PM, Colin Walters wrote:
>>> On Sun, Sep 21, 2014, at 03:36 PM, David Shea wrote:
>>>> - None of this ever actually gets called, because gdk calls _exit(1)
>>>> when it loses the X connection.
>>> That behavior is inherited from libX11 - every toolkit does it. The
>>> intention is to tie application lifecycle to that of the display.
>> This is almost true. For example, qt calls exit(1) on an I/O error. The
>> difference there is that 1) it doesn't call _exit, because skipping our
>> shutdown handlers is seriously a jerk thing to do, and 2) there's at
>> least a fixme comment showing they're aware that this behavior isn't
> Looks like this is:
> Which is really i'd say a NetworkManager/nspr bug, but that's the
> current reason.
Fixing a bug in NetworkManager by bypassing NetworkManager's code in a
library is the craziest thing I have heard all day.
And calling exit() would still only half-fix the problem. Being able to
install custom callbacks or something would be ideal.
More information about the anaconda-patches