On Sun, 25 Mar 2001, Andrew Josey wrote:
> I forward a view from one participant who sees no change necessary.
> This responds to the proposed solution in section 11
> We have found cases where an application (over which we had no control)
> forks and execs an infinite number of child processes, failing to ever
> reap the zombie with a wait(). Eventually, the entire process table filled
> with zombie processes and all subsequent fork's failed. This problem
> could not have been alleviated if the affect of setting SIGCHLD to SIGIGN
> (causing all of the zombies to disappear) was not inheritable by the
> offending application.
This is a valid concern. However, it should be pointed out that such
applications would also fill up the process table if they were run from an
interactive shell (where they would most likely inherit SIGCHLD==SIG_DFL).
This strikes me as sufficient grounds to file bug reports with suppliers
of such applications, that obviously need to beef up their quality
assurance processes. It is certainly insufficient justification to have
standards "bless" such behaviour at the cost of removing conformance from
the majority of well-behaved applications.
> The change proposes to alter a definition of exec that has stood
> for years. Except for signals that have handlers, the child process
> inherits the signal settings of it's parent. Since this behavior has
> not apparently been a problem in the past (it might have been annoying
> in some circumstances, but no trail of defect reports has been generated
> to compensate for this specification), I do not see why we should change
> this aspect of the standard at this time.
"This behaviour has not apparently been a problem in the past", yes. But,
I can state from personal experience that it is extremely difficult to
bring such problems to the attention to the appropriate standards process.
Is 8 years enough "personal experience"?
There is definitely an almost universally held perception that the
standards could not possibly have gone wrong in this fashion. It should
be recognised that standards are not only specifications for systems. But
they are systems by themselves. Like all systems, standards cannot be
defect-free. And, some defects take a very lloonngg time to surface...
Marc.
+----------------------------------+-----------------------------------+
| Marc Aurele La France | work: 1-780-492-9310 |
| Computing and Network Services | fax: 1-780-492-1729 |
| 352 General Services Building | email: yyy@xxxxxxxxxxx |
| University of Alberta +-----------------------------------+
| Edmonton, Alberta | |
| T6G 2H1 | Standard disclaimers apply |
| CANADA | |
+----------------------------------+-----------------------------------+
XFree86 Core Team member. ATI driver and X server internals.
|