Re: [tegaki] debian packages

47 views
Skip to first unread message

shikashi

unread,
Sep 18, 2011, 4:07:36 PM9/18/11
to tegak...@googlegroups.com
I know this is an old discussion but, since I couldn't find any other relevant info on the subject, I'm assuming this is still an open issue.

Problem description: http://groups.google.com/group/tegaki-hwr/browse_thread/thread/2c00b1d74cdadae/ba8f279a50855a43 (Basically: tegaki-recognize, as well as other tegaki apps, crash with nothing but an "Illegal instruction" error.)

I had the exact same problem while trying to run tegaki-recognize (in my case as an IBUS component) and on Debian too (squeeze). However, now that I managed to fix it I think the fact that I too was using Debian is just a coincidence.

I downloaded tegaki-recognize from tegaki.org and put it through my debugger. Then the problem was evident: libzinnia.so was using opcodes not supported by the uP (namely sse2 instructions). I followed the link on tegaki.org and downloaded the latest version of zinnia from http://zinnia.sourceforge.net (zinnia-0.06). Compiled the sources and found that nothing changed, the app crashed with the same error.

I went back and skimmed through config.log: It turns out that the configure script was enabling sse2 after "testing" whether gcc (and g++) supported it.

Now, I don't claim I understand configure's voodoo code, but it seems to me that there's really no check being done...

For sse configure tries to compile this:
<<<<<<<<<<<<<<< code start <<<<<<<<<<<<<<<
/* end confdefs.h.  */

int
main ()
{

#if defined(__GNUC__) && defined(__i386__)
__asm__("movups %xmm0, (%esp)");
#else
error Not gcc on x86
#endif

  ;
  return 0;
}
>>>>>>>>>>>>>> code end >>>>>>>>>>>>>>

But it looks like for sse2 and mmx it tries to compile this:
<<<<<<<<<<<<<<< code start <<<<<<<<<<<<<<<
/* end confdefs.h.  */

int
main ()
{

  ;
  return 0;
}
>>>>>>>>>>>>>> code end >>>>>>>>>>>>>>

Furthermore, it seems to me that this would only yield whether gcc _can_ compile into a given ISA extension, but it does not check whether the underlying hardware _supports_ it.

In my case I just hacked configure so that the sse2 check would fail. Tegaki is now running fine.

I guess very (very) few people are still using 9 years old hw and that's why this haven't been noticed earlier.

Grettings.

Even after updating to latest unstable I cannot reproduce.
I advise you do open a bug with Debian ($ reportbug tegaki-recognize). Link to
this discussion and lets hope that the developer can ask the right questions
so we can nail down this issue.

-Christoph

Am Samstag, 29. Mai 2010 schrieb alias:

I used another (in this case only virtual) machine to test.
Used Debian stable, added sid to the sources list and let it install
tegaki and dependecies.
So now the machine is a mix of stable and sid - compared to the other
which is only sid.
Having the same problems (like crashing when tegaki-zinnia-japanese is
installed).
Using 32bit systems just in case it matters.
 
> tegaki-train crashed when doing this:

> starting tegaki-train, click on "+ add", choosing "predifined set -
> hiragana", adding sample for hiraga a, using "train" adding name and
> short name, clicking "ok" it crashes with

I cannot reproduce these messages, neither for tegaki-recognize nor tegaki-
train. I'm also running Debian sid (though currently not fully updated):

Versions of packages tegaki-recognize depends on:
ii  python                        2.5.4-9    An interactive high-level object-
o
ii  python-gtk2                   2.17.0-2   Python bindings for the GTK+
widge
ii  python-tegaki-gtk             0.3.1-1    GTK+ widget Python model for
Tegak

tegaki-recognize recommends no packages.

Versions of packages tegaki-recognize suggests:
ii  tegaki-zinnia-japanese        0.3-1      Japanese handwriting model for
Zin
ii  tegaki-zinnia-simplified-chin 0.3-1      Simplified Chinese handwriting mod

$ aptitude search tegaki
p   ibus-tegaki                     - tegaki engine for IBus                    
i   python-tegaki                   - core Python module of Tegaki              
i   python-tegaki-gtk               - GTK+ widget Python model for Tegaki      
i   python-tegakitools              - command-line tools for Tegaki            
p   scim-tegaki                     - Tegaki input method for SCIM              
i   tegaki-recognize                - handwriting recognition application      
i   tegaki-train                    - train tegaki with your own handwriting    
i   tegaki-zinnia-japanese          - Japanese handwriting model for Zinnia    
i   tegaki-zinnia-simplified-chines - Simplified Chinese handwriting model for Z
i   tegaki-zinnia-traditional-chine - Traditional Chinese handwriting model
for 

Mathieu Blondel

unread,
Sep 19, 2011, 12:40:39 PM9/19/11
to tegak...@googlegroups.com
Nice investigation. Since the problem is in Zinnia, it'd be best to
report the problem upstream.

Mathieu

shikashi

unread,
Sep 20, 2011, 9:55:41 PM9/20/11
to Tegaki Handwriting Recognition Project
I posted it on Zinnia's bug tracker on sourceforge:
http://sourceforge.net/tracker/?atid=1086998&group_id=232651&func=browse

Doesn't seem like there's much traffic, though. Well, my part is done.

Greetings.
Reply all
Reply to author
Forward
0 new messages