John Levon wrote:
> On Mon, Mar 22, 2004 at 09:28:01AM +0000, Angus Leeming wrote:
>
>> John Levon wrote:
>> >> Or should the code in child_handler be truly trivial:
>> > Yes.
>>
>> John, I take it that you're (almost) happy with the existing signal
>> handling code in error_handler below. To my mind it needs only a
>> handling_error flag to ensure that it handles multiple signals
>> safely. (The SIGPIPE should be removed if it isn't an error.)
>
> Not really happy actually. If we get a SIGPIPE and a SIGSEGV, we'd
> lose the SIGSEGV afaics.
Hence my statement that we should remove the SIGPIPE. Ie, SIGPIPE
signals should not be handled by error_handler. In fact we don't need
to handle them at all.
> Also the lyxerr outputs still have a chance of not working at all,
> regardless of other signals. The issue is that the signal can come
> along asynchronously wrt state held in library code.
Understood.
> It's not a *big* deal but since you're fiddling with getting
> it right...
So outline a solution strategy. Currently we have:
signal(SIGSEGV, error_handler);
void error_handler(int err_sig)
{
/* print error message */
/* Deinstall the signal handlers */
LyX::cref().emergencyCleanup();
lyxerr << "Bye." << endl;
if (err_sig!= SIGHUP &&
(!GetEnv("LYXDEBUG").empty() || err_sig == SIGSEGV))
lyx::support::abort();
exit(0);
}
I take it that you would prefer:
static sig_atomic_t error_flagged_;
void error_handler(int err_sig)
{
if (!error_flagged_)
error_flagged_ = err_sig;
}
and that the 'real' code handling the error is then handled in the
main program loop. However, I don't see anyway to code that up
without using throw/catch.
--
Angus
