Re: [Texmacs-dev] RE: Texmacs segfault

[Felix Breuer]

Hello! 

I have made some experiments (re)compiling TeXmacs and gcc-3.2 on Gentoo
Linux 1.4. These are the results: 

In the table below "GCC" refers to "the C(XX)FLAGS gcc-3.2 has been
compiled with using gcc-3.2" and "TM" refers to "the C(XX)FLAGS TeXmacs
1.0.0.18 has been compiled with using gcc-3.2". 
"segfaults" means that a TeXmacs that has been compiled with the flags
TM (X) by a gcc-3.2 which itself has been compiled with flags GCC (Y)
segfaults when saving a file on my Gentoo Linux 1.4_rc1 System (guile
1.4-r3). "works" means that it does not segfault. 

C(XX)FLAGS Flags: 

GCC (1) : -O0 -g -ggdb 
GCC (2) : -mmmx -O3 
GCC (3) : -march=pentium4 -mmmx -msse -msse2 -mcpu=pentium4 -O3 

TM (1) : -g -ggdb 
TM (2) : -O3 
TM (3) : -O3 -fexpensive-optimizations -fno-exceptions 


Results: 

        |   TM (1)   |   TM (2)   |   TM (3)   | 
--------+------------+------------+------------+ 
GCC (1) |   works    |   works    | segfaults  | 
--------+------------+------------+------------+ 
GCC (2) |   works    | segfaults  | segfaults  | 
--------+------------+------------+------------+ 
GCC (3) |   works    | segfaults  | segfaults  | 
--------+------------+------------+------------+ 

What is noteworthy about this: gcc-3.2 behaves differently (i.e.
produces different output) when gcc-3.2 itself is optimized. So whether
or not there is some hidden bug in TeXmacs'es code - gcc-3.2's
optimization routines are not entirely trustworthy.

However my entire Gentoo 1.4 System has been built with an optimizing
_and_ optimzed gcc-3.2 and it is *really* stable, so... there is
something strange going on here. (But we knew that right from the start
:)

I am going to write a new TeXmacs-ebuild (for Gentoo) that switches off
optimization for TeXmacs, so that the Gentoo people keep TeXmacs in
their distribution.


Regards,
Felix

On Mon, 2002-10-21 at 22:01, Joris van der Hoeven wrote: 
> 
> > The gentoo guys closed the but with this motivation:
> > 
> > This package has proven to be very unreliable under GCC 3.2, and with
> > any optimizations in general.  As such, it's being moved to an
> > unsupported status. Issues with the package should be taken to the
> > developing party www.texmacs.org.  The best solution to any problems has
> > been, in the past, to disable optimizations. If a guaranteed, tried and
> > tested cross-platform solution to optimization/segfault woes can be
> > found, we will implement it.
> 
> This is crazy; we are not responsible for bugs in the g++ optimizer!
> They could just stick to g++ 2.95.3 if g++ 3.2 is bugged.
> 
> > I'm happy with the static build for now.
> 
> In fact, I think that they did not compile g++ 3.2 well;
> I managed to build an optimized version of TeXmacs with g++ 3.2
> without any problems...
> 
> 
> 
> _______________________________________________
> Texmacs-dev mailing list
> address@hidden
> http://mail.gnu.org/mailman/listinfo/texmacs-dev