Subject: Re: Signal catching changes broke Matlab
To: None <firstname.lastname@example.org>
From: Christos Zoulas <email@example.com>
Date: 09/27/2004 17:35:17
In article <20040927171244.ECDA0492308@s102-n054.tele2.cz>,
Jaromir Dolecek <jdolecek@NetBSD.org> wrote:
>> Mark Davies tracked down the problem with Matlab. It's caused
>> by change in rev. 1.190 of sys/kern/kern_sig.
>> Problem is, how would be fix this properly? Just backing
>> the change off is an option. Perhaps only do the resetting
>> if the signals are not shared (i.e. p->p_sigacts.sa_refcnt == 1) ?
>I think the change should just be backed off - apps should
>know best why they mask SIGSEGV, and doing magic behind
>app's back like resetting the signal handler is IMHO not right.
>So, unless someone thinks different, I'd remove the change by
>end of the week and request pullup do 2.0 branch (now 2.0RC1).
It is not the masking of SIGSEGV... It is a more subtle thing,
because linux behaves the same way. Anyway the change should
not be backed out because it will make SEGV processes spin again,
but we should investigate why side-effect it is causing.