Re: [Gnustep-cvs] r23144 - in /libs/back/trunk: ChangeLog Source/x11/XGServer.m

[David Ayers]

Richard Frith-Macdonald schrieb:
> 
> On 7 Jul 2006, at 19:28, David Ayers wrote:
> 
>>
>>> URL: http://svn.gna.org/viewcvs/gnustep?rev=23144&view=rev
>>> Log:
>>> Shut down cleanly on receipt of SIGTERM
>>>
>>> Modified:
>>>     libs/back/trunk/ChangeLog
>>>     libs/back/trunk/Source/x11/XGServer.m
>>
>>
>> I'm not sure if we should instead add an NSInvocation of terminate: to
>> the current or even the main run loop.  Calling terminate: in the  signal
>> handler seems almost a bit more dangerous than killing the process  as we
>> don't really know in which state we are when terminate: may start  saving
>> potentially inconsistent data...
> 
> 
> Good idea ... while the terminate: method implementation should  really
> be responsible for ensuring that it shuts down cleanly, it's  certainly
> true that a signal could hit a delicate point and cause  problems.
> Probably the only way to be really safe would be to add a mechanism  to
> NSRunLoop  to handle this ... such that a signal handler can set  some
> value that the run loop would know how to deal with.  Unless the  run
> loop and signal mechanisms cooperate there is always going to be  a
> chance that truing to manipulate the run loop within a signal  handler
> could cause problems.

Indeed.  The mechanism also needs to take into account that termination
can be aborted so there might be a need for some cleanup also.

>> I'm also considering whether it would be wise to remove the handler
>> during the handling.
> 
> 
> Perhaps blocking SIGTERM rather than just removing the handler would  be
> good?

Yes, that sounds better.

Cheers,
David