guile 1.8.5 compile segfault

guile 1.8.5 compile segfault

[tantalum]

hi

having dependencies
gmp 4.2.2-2
libtool 2.2.6a-1
installed and 64bit linux,
im trying to compile guile 1.8.5, but it fails leaving the message found 
here:
http://pastebin.com/m3bd2e879
"Speicherzugriffsfehler" translates to "segmentation fault"

this happens also with the git version.

im clueless, can i do something?
i really like guile, and like to use the newest version because it fixed 
a bug with g-wrap.

Re: guile 1.8.5 compile segfault

[Neil Jerram]

Hi...

2008/9/19 tantalum <theseph@gmx.de>:
> hi
>
> having dependencies
> gmp 4.2.2-2
> libtool 2.2.6a-1
> installed and 64bit linux,
> im trying to compile guile 1.8.5, but it fails leaving the message found
> here:
> http://pastebin.com/m3bd2e879
> "Speicherzugriffsfehler" translates to "segmentation fault"

A problem very similar to that was fixed by these commits, since the
1.8.5 release:
http://git.savannah.gnu.org/gitweb/?p=guile.git;a=commit;h=9143131b2766d1e29e05d61b5021395b4c93a6bc
http://git.savannah.gnu.org/gitweb/?p=guile.git;a=commit;h=c67a68f4f14bc0bdc11c1dca37eae7f610432a49
http://git.savannah.gnu.org/gitweb/?p=guile.git;a=commit;h=adacd2dac5ad1c80a8faf8933c228533fb82cc80

> this happens also with the git version.

What do you mean, exactly?  (If you mean the head of the
branch_release-1-8 branch, I guess this must be a new problem.)

> im clueless, can i do something?
> i really like guile, and like to use the newest version because it fixed a
> bug with g-wrap.

It depends if this is a new problem, or one that we have fixed - so I
need your answer to the question just above.

Regards,
     Neil

Re: guile 1.8.5 compile segfault

[tantalum]

i meant the version one gets when issuing "git clone 
git://git.sv.gnu.org/guile.git".
now trying this again, it wont complete "autoconf configure.in" (i found 
no further documentation how to compile except a comment in configure.in)
it runs a second then gives:

;;
configure.in:44: error: possibly undefined macro: AM_INIT_AUTOMAKE
      If this token and others are legitimate, please use m4_pattern_allow.
      See the Autoconf documentation.
configure.in:69: error: possibly undefined macro: AC_LIBTOOL_WIN32_DLL
configure.in:83: error: possibly undefined macro: AM_PROG_CC_C_O
configure.in:85: error: possibly undefined macro: AC_LIBTOOL_DLOPEN
configure.in:86: error: possibly undefined macro: AC_PROG_LIBTOOL
configure.in:87: error: possibly undefined macro: AC_CHECK_LIB
configure.in:93: error: possibly undefined macro: AM_CONDITIONAL
configure.in:95: error: possibly undefined macro: AM_PATH_LISPDIR
configure.in:248: error: possibly undefined macro: AC_LIBOBJ
configure.in:837: error: possibly undefined macro: AM_GNU_GETTEXT
;;

Re: guile 1.8.5 compile segfault

[Neil Jerram]

2008/9/20 tantalum <theseph@gmx.de>:
>
> i meant the version one gets when issuing "git clone
> git://git.sv.gnu.org/guile.git".
> now trying this again, it wont complete "autoconf configure.in" (i found no
> further documentation how to compile except a comment in configure.in)
> it runs a second then gives:
>
> ;;
> configure.in:44: error: possibly undefined macro: AM_INIT_AUTOMAKE
>     If this token and others are legitimate, please use m4_pattern_allow.
>     See the Autoconf documentation.
[...]

Guile requires particular versions of the various autotools, which may
not be what you have...

Could you try the latest branch_release-1-8 snapshot at
http://www.ossau.uklinux.net/guile/snapshots ?  You'll be able to
build that with just the standard ./configure and make sequence.

Regards,
       Neil

Re: guile 1.8.5 compile segfault

[tantalum]

ok, i did try to compile from the latest snapshot.
same error, i guess.
'make' produces a shorter message when run a second time.
the linked message is the end of a first 'make' run.
messages:
./configure    http://pastebin.com/m684b4e99
make             http://pastebin.com/m3a251e3a

Re: guile 1.8.5 compile segfault

[Neil Jerram]

2008/9/21 tantalum <theseph@gmx.de>:
> ok, i did try to compile from the latest snapshot.
> same error, i guess.

OK, I'm sorry about that.  The similar problem on other platforms was
caused by incorrect detection of the stack direction (i.e. whether the
stack grows upwards or downwards), so you could try playing with this
to see if that is your problem too...  I would guess that for your
platform (x86_64) the stack actually grows downwards.  If that is
right, in ./configure, try commenting out the 60 lines following (but
not including) where it says

SCM_I_GSC_STACK_GROWS_UP=0

(This is at line 41264 in today's snapshot).

That should force configure to "discover" that the stack grows downwards.

If that works, we can then investigate why the autodetection program
doesn't work on x86_64.

Regards,
       Neil

Re: guile 1.8.5 compile segfault

[tantalum]

it did not work.
i copied and unpacked todays snapshot, and removed the 60 lines.
all text in between "SCM_I_GSC_STACK_GROWS_UP=0"
and
"{ echo "$as_me:$LINENO: checking for float" >&5"

error happens again, with same output as before.


i even set SCM_I_GSC_STACK_GROWS_UP to 1, no change.




tantalum

Re: guile 1.8.5 compile segfault

[Neil Jerram]

2008/9/22 tantalum <theseph@gmx.de>:
> it did not work.
> i copied and unpacked todays snapshot, and removed the 60 lines.
> all text in between "SCM_I_GSC_STACK_GROWS_UP=0"
> and
> "{ echo "$as_me:$LINENO: checking for float" >&5"
>
> error happens again, with same output as before.

OK.... we're doing well here, aren't we? :-)

Can you tell us more about your platform?  I guess I'm not that
familiar with x86_64, what does that mean, and is it similar to either
of the Debian architectures ia64 and amd64 (where Guile is reportedly
building successfully)?  Also what version of GCC are you compiling
with?

Also we could try to look more closely at the crash.  Is there a core
dump?  If so, you could look at by running

 ./pre-inst-guile-env ./libtool --mode=execute gdb libguile/guile core.NNNN

where core.NNNN is the corefile name.  If there isn't already a core
dump, you should be able to do something like "ulimit -c unlimited" to
enable cores, then run make again to cause the problem and generate a
core dump.

Once you're in GDB, type bt to get the stack, and if you can a few
prints to identify the immediate cause of the fault, and let us know
what that gives.

Thanks for keeping at this!  In the meantime, BTW, could you work
around this by using Guile on a less troublesome platform?

Regards,
      Neil

Re: guile 1.8.5 compile segfault

[tantalum]

x86_64 is the same as amd64, just a vendor neutral name.
(http://en.wikipedia.org/wiki/X86-64)

i use gcc 4.3.2-1

2008/9/22 Neil Jerram <neiljerram@googlemail.com>:
 > Once you're in GDB, type bt to get the stack, and if you can a few
 > prints to identify the immediate cause of the fault
im not sure what you mean by prints. the frame command in gdb?

i managed to get a core dump file. heres a backtrace and some "frame x" 
calling:
http://pastebin.com/m3c55053b


 > could you work around this by using Guile on a less troublesome platform?
that is a good idea. i will try to get a setup on another machine i can 
ssh to.

to compile guile 1.8.5 (release or snapshot-0923) on this i386 other 
machine, i need to apply this patch, found via google:
http://pastebin.com/m6226ece1
otherwise it stops leaving its long message, end of which is:
http://pastebin.com/m102992ca

it looks like this error happens at a very similar position as the 
segfault error.

im not too sure if the patch does a good thing, as i ran into strange 
problems with guile-gnome.

Re: guile 1.8.5 compile segfault

[Neil Jerram]

Sorry to take a while to continue this investigation.

2008/9/23 tantalum <theseph@gmx.de>:
>
> i managed to get a core dump file. heres a backtrace and some "frame x"
> calling:
> http://pastebin.com/m3c55053b

This:

(gdb) frame 0
#0  scm_mark_locations (x=0x7fff7f4c8d68, n=18446744073709551561) at
gc-mark.c:435
435           SCM obj = * (SCM *) &x[m];
(gdb) frame 1
#1  0x00002b802b8a66ed in scm_threads_mark_stacks () at threads.c:1390
1390          scm_mark_locations (t->base, t->top - t->base);

really does indicate a problem with SCM_STACK_GROWS_UP.  The relevant
code in scm_threads_mark_stacks () is:

#if SCM_STACK_GROWS_UP
      scm_mark_locations (t->base, t->top - t->base);
#else
      scm_mark_locations (t->top, t->base - t->top);
#endif

so SCM_STACK_GROWS_UP must be defined and non-zero.

The definition of SCM_STACK_GROWS_UP is in libguile/scmconfig.h (which
is itself generated during configuration).  Could you check what you
have there?

> to compile guile 1.8.5 (release or snapshot-0923) on this i386 other
> machine, i need to apply this patch, found via google:
> http://pastebin.com/m6226ece1
> otherwise it stops leaving its long message, end of which is:
> http://pastebin.com/m102992ca

This one is a known problem from using Libtool 2.2.  We currently
build and test Guile with Libtool 1.5.22, and are aware that there are
issues with using 2.2 instead - such as this one.

Regards,
         Neil