libi18nlangtag crash on SunOS 5.11

[Richard PALO]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Le 19/12/14 16:02, Eike Rathke a écrit :
> Hi Richard,
> 
> On Friday, 2014-12-19 14:18:36 +0100, Richard PALO wrote:
> 
>>> warn:vcl:101172:1:vcl/generic/fontmanager/fontconfig.cxx:868: In glyph fallback 
>>> throwing away the language property of en because the detected script for '0x9f
>>> 3' is Bengali and that language doesn't make sense. Autodetecting instead.
> 
> That warning is only indirectly related because something may trigger
> initialization of liblangtag for an unknown locale value, I've seen it
> without any crash. Adapted subject..
> 
>> My locale is fr_FR.UTF-8 on SunOS 5.11 under pkgsrc.
>>
>> it's coring with:
>> 	sigmask = 0xffffbefc,0xffffffff,0x000001ff
>> 	cursig = SIGSEGV
>>
>> Coredump indicates:
>>> 08045fb8 libi18nlangtag.so`LiblantagDataRef::setup+0x4c(fe5b5bdc, fe5b5bdc, 
>>> fe5b5a1c, 804605c)
>>> 08046128 libi18nlangtag.so`LanguageTagImpl::canonicalize+0x784(e34a900, 0, 
>>> fe59b58b, fe5b28bc)
> 
> That is the actual top of the backtrace?
> There isn't much in i18nlangtag/source/languagetag/languagetag.cxx
> LiblantagDataRef::setup() that could go wrong. The only thing that comes
> to mind is that LiblantagDataRef is a static instance initialized with
> 
> namespace {
> struct theDataRef : public rtl::Static< LiblantagDataRef, theDataRef > {};
> }
> 
> which for some reason may not work on SunOS / your compiler combination.
> As probably no one is using that here, you'd have to break in
> LanguageTagImpl::canonicalize() line 1204 where
> theDataRef::get().incRef() is called (or in the
> LiblantagDataRef::incRef() inline) and investigate what actually happens
> the very first time when the incRef() calls setup()
> 
> Other than that maybe the lt_db_initialize() call misbehaves, but since
> that is not in the backtrace..
> 
> However, only a step-through could provide insights.
> 
>   Eike
> 

Okay, thanks.  I'm building now an upgrade to 4.3.5.2 and will attempt to see more in 
the suggested areas. (I too noticed I'd be wasting my time in fontconfig, hence my post:)

For info, build is with gcc4.9.2 and the back trace comes from 'mdb' on the core file 
with c++ demangling turned on.

cheers
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1
Comment: Using GnuPG with Thunderbird - http://www.enigmail.net/

iQEcBAEBAgAGBQJUlEHYAAoJECAB22fHtp27V90H+gIPADtnj4ZPmEIyT00SLjJL
xob8DpRcllcNzk3nd9blPqC0EMF5cIz86d4iz/Ryeek+HEEYkwDQrNAdnNTSizzK
o1XG3cX+wpvE5j4upvivs+Wa4cTNYT5jwAmJIjRtNGFG50oYWvYouvQzWZyO0COP
jT2ZLszVIwH+BFafztsOJD6SbZl+d/CJk1PAGZaVmNzSYaZTGCflzbMRm0QFsm8Z
55n1HAL3ImI5P6r2yzOFlTRqXaw5nPsmDJCPHLGrtuc+F3OXQZS0II/v65zw1JFr
DvQRaIdXevVloYyvFJZtPmbhjEWRhsf584ejrkgoOnvqBKp1R46LEIrrtWdC1Dw=
=se6a
-----END PGP SIGNATURE-----