Re: [Texmacs-dev] GCC-3.2 segfault and compilation fixes

[David Allouche]

On Wed, Jul 09, 2003 at 10:29:36AM +0400, Igor V. Kovalenko wrote:
.
> Sorry I might have missed some important discussion on this subject.

Actually there has not been any discussion on this subject recently.

> gcc (GCC) 3.2.3 20030502 (actually the latest CVS from 3.2 series)
>   Seems to be the same crashing thing
>   (as about Nov, 2002 for TeXmacs 1.0.0.21)
> 
> gcc (GCC) 3.3.1 20030705 (prerelease) (also from gcc-3_3-branch CVS)
>   Seems stable!
>   Also tried --enable-fastalloc=no, seen some decrease in performance
>   (as expected :) and still no crash for the same Help->Welcome test.
> 
> Any volunteer to confirm? If updated GCC fixes the stability problem
> than it may be extremelly hard to determine the exact cause of failure
> with 3.2 GCC.

I am routinely compiling texmacs with all optimizations enabled with
gcc 3.3.1 and I do not have any stability problem. I believe I also
used it a bit with fastalloc disabled without problem, but I have not
made any more formal testing.

It believe the bug was fixed in gcc-3.3.


> BTW I used TeXmacs-1.0.1.16-src.tar.gz from Jun 30, 2003
> with Guile 1.6.4 on latest Gentoo Linux.

TeXmacs-1.0.1.14 -- 1.0.1.16 with guile 1.4 on Debian Testing.

(Note: TeXmacs-1.0.1.17 cannot be compiled on my system)


> Two minor compile fixes were needed:
> ----------------
> diff -ur TeXmacs-1.0.1.16-src-orig/src/Guile/Scheme/object.hpp 
> TeXmacs-1.0.1.16-src/src/Guile/Scheme/object.hpp
> --- TeXmacs-1.0.1.16-src-orig/src/Guile/Scheme/object.hpp     2003-06-30 
> 13:12:32.000000000 +0400
> +++ TeXmacs-1.0.1.16-src/src/Guile/Scheme/object.hpp  2003-07-03 
> 22:34:16.000000000 +0400
> @@ -14,8 +14,9 @@
>  #define OBJECT_H
>  #include "tree.hpp"
>  #include "command.hpp"
> -typedef long SCM;
> -//#include <libguile/tags.h>
> +
> +//typedef long SCM;
> +#include <libguile/tags.h>
> 
>  class object_rep: concrete_struct {
>    SCM handle;

This was introduced in 1.0.1.16 to fix a compilation problem (anyone
can recall on which system the problem occurred? Haaaa, traceability
is wonderful).

If anyone want my opinion, I think typedefing SCM instead of including
libguile/tags.h is a really stupid thing to do.


> diff -ur TeXmacs-1.0.1.16-src-orig/src/common.makefile.in 
> TeXmacs-1.0.1.16-src/src/common.makefile.in
> --- TeXmacs-1.0.1.16-src-orig/src/common.makefile.in  2003-06-30 
> 13:12:41.000000000 +0400
> +++ TeXmacs-1.0.1.16-src/src/common.makefile.in       2003-07-03 
> 22:54:33.000000000 +0400
> @@ -220,7 +220,7 @@
>  $(classes_obj): Objects/%.o: %.cpp
>       $(call cxx_incl,$(classes_incl))
>  $(data_obj): Objects/%.o: %.cpp
> -     $(call cxx_incl,$(data_incl),-I$(tmsrc)/include)
> +     $(call cxx_incl,$(data_incl) -I$(tmsrc)/include)
>  $(edit_obj): Objects/%.o: %.cpp
>       $(call cxx_incl,$(edit_incl))
>  $(guile_obj): Objects/%.o: %.cpp

This is obviously a correct fix for a typo introduced in 1.0.1.16.

-- 
                                                            -- ddaa