Re: [Pan-devel] SIGSEGV @ socket-impl-gio.cc:502 (abort_flag)

[Heinrich Mueller]

Nathanael Hoyle schrieb:
> I am not specifically familiar with this code, but since no one else 
> has replied... I don't believe that this is duplicate.
>
> Firstly, the if statement block that you commented out occurs within 
> the else clause of the if structure which first checks for 
> _abort_flag.  This means that logically, _abort_flag must *not* be set 
> in order to ever enter the else block, and therefore, if it were a 
> duplicate/redundant check, then _abort_flag would never be set when 
> checked the second time (where you commented it out).  The fact that 
> commenting it out caused a change in behavior for you proves that it 
> is possible for _abort_flag to not be set in the first test, but to be 
> set in the second. Presumably, this is because either/both of 
> do_read() or do_write() may set _abort_flag (in the else block, above 
> your "duplicate" if).  I expect your actual problem lies elsewhere 
> (perhaps within the on_socket_abort() call).
>
> -Nathanael
>
> On Aug 21, 2009, at 4:10 PM, Heinrich Müller wrote:
>
> > OS: 64 bit ubuntu jaunty 9.04, latest version from SVN.
> >
> > code snippet //
> >
> > /if (_abort_flag)
> > {
> >   _listener->on_socket_abort (this);
> > }
> > else if (!(cond & (G_IO_IN | G_IO_OUT)))
> > {
> >   _listener->on_socket_error (this);
> > }
> > else // G_IO_IN or G_IO_OUT
> > {
> >   const DoResult result = (cond & G_IO_IN) ? do_read () : do_write ();
> >   /*  if (_abort_flag)        _listener->on_socket_abort (this);
> >   else*/ if (result == IO_ERR)   _listener->on_socket_error (this);
> >   else if (result == IO_READ)  set_watch_mode (READ_NOW);
> >   else if (result == IO_WRITE) set_watch_mode (WRITE_NOW);
> > }/
> >
> >
> > //
> >
> > isn't the second appearance of abort_flag a dupe? it was already 
> > checked before by the first "if" branch...
> > i commented that and got no errors so far (opposed to random 
> > segfaults before) :D
> >
> >
> > _______________________________________________
> > Pan-devel mailing list
> > address@hidden
> > http://lists.nongnu.org/mailman/listinfo/pan-devel
that's what i believe, too. the else branch wouldn't have the 
_abort_flag set, but the code has a check there.
odd.
gdb says that the call for _listener->on_socket_abort (this) causes a 
SIGSEGV on exactly this line (inside the else branch).

Program received signal SIGSEGV, Segmentation fault.
[Switching to Thread 0x7f61faa28790 (LWP 12446)]
0x000000000052431a in pan::GIOChannelSocket::gio_func (this=0x3281b60,
  channel=0x8e98340, cond=G_IO_IN) at socket-impl-gio.cc:502
502             if (_abort_flag)        _listener->on_socket_abort (this);