Re: [FORGE] struct package segmentation fault

[Olaf Till]

On Sun, Jan 26, 2014 at 12:43:59PM +0100, Olaf Till wrote:
> On Sat, Jan 25, 2014 at 09:34:06PM +0100, Olaf Till wrote:
> > <snip>
> > But according to the backtrace the init
> > function seems to call some functions related to graphics (#3 up to
> > #1), where the crash occurs. The reason why the init function does
> > something graphics-related is probably not in the struct package.
> 
> Checking with gdb myself, the function:
> 
> > #1  0x00007fffe580b723 in base_properties::factory_defaults() ()
> 
> is not even called in my Octave-3.8.0 during installation of the
> released struct-1.0.10.tar.gz.

Further digging through Octave source, AFAIKT the only way
'base_properties::factory_defaults()' is called is by a call to
'root_figure::init_factory_properties()'. And the latter is only
called during static initialization of
'root_figure::factory_properties' (libinterp/corefcn/graphics.cc, line
3221). The latter should only happen once, at start of
Octave. According to your backtrace, the dynamic linker seems to
repeat this static initialization when loading a dynamic function,
which should not happen.

I can only guess, but maybe for some reason a static link of
'cell2fields.oct' to 'liboctinterp.so...' is performed at your
system? Can you tell what were the last lines of the output of
./configure... for your Octave-3.8.0, in particular

  Build static libraries:
  Build shared libraries:

A good hint could also be if you just told the filesize of your
generated 'cell2fields.oct', which should linger somewhere if your
Octave crashed during installation of struct.

And are you sure that the crash occurs only if installing the struct
package, not if installing any other package which contains compiled
functions? As yet I don't see any direct relation to the struct
package.

Olaf

-- 
public key id EAFE0591, e.g. on x-hkp://pool.sks-keyservers.net

signature.asc
Description: Digital signature