[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Exim4 problems



Hallo!

On Sun, 15 May 2011 23:57:00 +0200, Samuel Thibault <samuel.thibault@gnu.org> wrote:
> Svante Signell, le Sun 15 May 2011 23:52:10 +0200, a écrit :
> > On Sun, 2011-05-15 at 23:34 +0200, Samuel Thibault wrote:
> > > Svante Signell, le Sun 15 May 2011 23:20:52 +0200, a écrit :
> > > > Segmentation fault
> > > 
> > > Do you have a core file? Does your shell perhaps has limited core size
> > > limit? (ulimit -a to check)
> > 
> > Sorry I don't find any core file anywhere:
> > ulimit -a
> > socket buffer size       (bytes, -b) unlimited
> > core file size          (blocks, -c) unlimited
> 
> So it's unlimited, but you don't have a core file? What does
> /servers/crash point to?

Last time I looked, the crash server had issues with core file
generation:
<http://www.gnu.org/software/hurd/open_issues/crash_server.html>.


> > I thought development was made with high-level languages nowadays (at
> > least C-level)
> 
> It is. But bugs are quite often when things go wrong ;)

It is quite easy to write code in C that does not conform to the
specification, and as the compiler's optimization passes get better and
better -- but they do rely on conforming code -- more and more bugs are
uncovered.

> > Update on optimization levels: -O1 does not work either :(
> > Other compiler switches are: -fno-strict-aliasing -fvisibility=hidden
> > -D_LARGEFILE_SOURCE -D_FILE_OFFSET_BITS=64 -D_LARGEFILE_SOURCE -Wall

Exim being built with -fno-strict-aliasing is a good sign of what I just
wrote: conforming ``modern'' code would be fine to be built with
-fstrict-aliasing (and its optimization possibilities).

On the other hand, building with -O1 usually does not uncover such
issues, only -O2 (or, more specific, some of -O2's optimization flags).

> > There are a lot of warnings though: maybe some of them can cause
> > problems.
> 
> Possibly.
> 
> It could be very worth trying to build with the exact same compiler and
> libc on linux, to check whether it gets a crash too.

Right.

And, at this point, I would also try building the upstream source
(possibly CVS HEAD/Subversion trunk/Git master branch), and see what's
going on there.


Grüße,
 Thomas

Attachment: pgpB7x69avsYH.pgp
Description: PGP signature


Reply to: