Bug#376603: [pkg-fetchmail-maint] Bug#376603: fetchmail: empty pidfile in ~/ causes fetchmail to fail silently

[Justin Pryzby]

On Tue, Jul 04, 2006 at 11:57:13AM +0200, Matthias Andree wrote:
> Justin Pryzby <justinpryzby at users.sourceforge.net> writes:
> 
> > It notices that there's an rc file:
> > |open("/home/pryzbyj/.fetchmailrc", O_RDONLY) = 4
> >
> > WTF?
> > |ioctl(4, SNDCTL_TMR_TIMEBASE or TCGETS, 0xafa04e58) = -1 ENOTTY
> > (Inappropriate ioctl for device)
> 
> glibc at work, probably isatty() or equivalent to find out the default
> buffer size (line vs. fully buffered).
Nice

> > Huh?  Yes, we are a process..
> > |kill(0, SIG_0)                          = 0
> > |close(4)                                = 0
> 
> That's indeed a bug, and it's easy to fix (patch attached, description
> below). The fscanf() in fm_lock_state() returns EOF; which isn't handled
> in <= 6.3.4.

> I find the PID of 0 astonishing however. Is Debian's GCC different than
> SUSE's?
What compiler {,version} and optimization level are you using?

In a test, gcc-3.4 -O0, gcc-4.0 -O3, and gcc-4.0 -O1 results in sscanf
setting integers to zero, even if no valid input was read.

IIRC buffer initialization also changes.

This may, in fact, be a deliberate but otherwise unnecessary change by
the toolchain folks to break in an obvious way things which rely on
unfounded assumptions, rather than allowing them to work with the
author's $cc, but fail obscurely with others'.

Justin