wxruby users - Dec 2007 - build problems, part III, and runtime problems, part I

After decoding the unresolved symbols from wxruby2.so using c++filt, and
comparing with the symbols provided by wxGTK, it became clear that wxGTK was
using wchar_t whereas wxruby2.so was using char.  So I rebuilt wxGTK,
turning off unicode.    That worked.  No more unresolved symbols.  What
gives?  Should the wxruby code have been built using wchar_t, and if so,
how?

However, now I get the following:

% ruby -rubygems nothing.rb
/home/rwa/linux-packages/ruby-1.8.6/lib/ruby/site_ruby/1.8/i686-linux/wxruby2.so:
[BUG] Segmentation fault

the crash is in rb_hash_aref.  I saw that I was not the only one to run into
this from recent mailing list discussion.

The issue appears to be that swig_ruby_trackings has a bogus value.  In my
case it always happens to be 4.  It appears it is supposed to point to valid
storage.

The relevant function appears to be this one:

SWIGRUNTIME void SWIG_RubyInitializeTrackings(void)

where swig_ruby_trackings is looked up, and if it comes back nil, a new one
is created.  Except that, for reasons I don''t understand, it comes back
4,
not 0.  When this gets dereferenced after a series of calls from
Init_wxRubyConstants, segfault.  Any help?  This seems like a real bug, and
I''m starting to wonder if this package is usable at all.


_______________________________________________
wxruby-users mailing list
wxruby-users@rubyforge.org
http://rubyforge.org/mailman/listinfo/wxruby-users

Robert Anderson wrote:
> So I rebuilt wxGTK, turning off unicode.    That worked.  No more 
> unresolved symbols.  What gives?  Should the wxruby code have been 
> built using wchar_t, and if so, how?
I''m sorry but I  don''t understand this. Unfortunately
I''m away and don''t
have a Linux box here - but, wxruby2 should *only* work with a unicode 
build; all of the conversions from wxString to ruby Strings assume that 
wxGTK has been built with unicode support.
> However, now I get the following:
>
> % ruby -rubygems nothing.rb
>
/home/rwa/linux-packages/ruby-1.8.6/lib/ruby/site_ruby/1.8/i686-linux/wxruby2.so:
> [BUG] Segmentation fault
>
> the crash is in rb_hash_aref.  I saw that I was not the only one to 
> run into this from recent mailing list discussion.
Could you refer me the discussion please?
>
> The issue appears to be that swig_ruby_trackings has a bogus value.  
> In my case it always happens to be 4.  It appears it is supposed to 
> point to valid storage.
>
> The relevant function appears to be this one:
>
> SWIGRUNTIME void SWIG_RubyInitializeTrackings(void)
This is code that''s generated by SWIG, not immediately part of wxRuby. 
What version of SWIG are you using? (`swig -version`) The most recently 
released version of SWIG came out after wxRuby 1.9.2 in November and 
hasn''t yet been tested with it. I''d strongly suggest using
version
1.3.31 for now.
> This seems like a real bug, and I''m starting to wonder if this
package
> is usable at all. 
Clearly not for you, clearly yes for lots of other people. There''s 
something that''s different about your environment that''s
causes serious
bugs; let''s figure out what it is.

alex

On Dec 27, 2007 2:36 PM, Alex Fenton <alex at pressure.to> wrote:
> Robert Anderson wrote:
> > So I rebuilt wxGTK, turning off unicode.    That worked.  No more
> > unresolved symbols.  What gives?  Should the wxruby code have been
> > built using wchar_t, and if so, how?
> I''m sorry but I  don''t understand this. Unfortunately
I''m away and don''t
> have a Linux box here - but, wxruby2 should *only* work with a unicode
> build; all of the conversions from wxString to ruby Strings assume that
> wxGTK has been built with unicode support.

Somehow my build is configured otherwise.  I don''t understand the
mechanisms
for how this is configure, so I don''t know how it got out of whack.

Here''s the evidence:

nm obj/App.o | grep wxStringBase8InitWithEPK | c++filt
         U wxStringBase::InitWith(char const*, unsigned int, unsigned int)

whereas I would assume a unicode build would be using wchar_t instead of
char.

>
> > However, now I get the following:
> >
> > % ruby -rubygems nothing.rb
> > /home/rwa/linux-packages/ruby-1.8.6
> /lib/ruby/site_ruby/1.8/i686-linux/wxruby2.so:
> > [BUG] Segmentation fault
> >
> > the crash is in rb_hash_aref.  I saw that I was not the only one to
> > run into this from recent mailing list discussion.
> Could you refer me the discussion please?

Sure, here''s a link:

http://rubyforge.org/pipermail/wxruby-users/2007-December/003497.html


>
> >
> > The issue appears to be that swig_ruby_trackings has a bogus value.
> > In my case it always happens to be 4.  It appears it is supposed to
> > point to valid storage.
> >
> > The relevant function appears to be this one:
> >
> > SWIGRUNTIME void SWIG_RubyInitializeTrackings(void)
> This is code that''s generated by SWIG, not immediately part of
wxRuby.
> What version of SWIG are you using? (`swig -version`) The most recently
> released version of SWIG came out after wxRuby 1.9.2 in November and
> hasn''t yet been tested with it. I''d strongly suggest
using version
> 1.3.31 for now.

 % swig -version

SWIG Version 1.3.33

Should I downgrade?
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
http://rubyforge.org/pipermail/wxruby-users/attachments/20071227/ffbca2f1/attachment.html

Robert Anderson wrote:
>
> Somehow my build is configured otherwise.  I don''t understand the 
> mechanisms for how this is configure, so I don''t know how it got
out
> of whack.
wxRuby doesn''t even try to configure this; it simply depends on there 
being a unicode not ANSI build. I am wondering whether SWIG is the issue 
here...
>
> Here''s the evidence:
>
> nm obj/App.o | grep wxStringBase8InitWithEPK | c++filt
>          U wxStringBase::InitWith(char const*, unsigned int, unsigned int)
>
> whereas I would assume a unicode build would be using wchar_t instead 
> of char.
>  
>
So would I; here''s what my build (OS X) says:

Abaddon:wxruby2 alex$ nm obj/App.o | grep wxStringBase8InitWithEPK | c++filt
         U wxStringBase::InitWith(wchar_t const*, unsigned long, 
unsigned long)
>
>
> http://rubyforge.org/pipermail/wxruby-users/2007-December/003497.html
Thanks. Will follow this up.

>
>  % swig -version
>
> SWIG Version 1.3.33
>
> Should I downgrade?
Yes, definitely, to 1.3.31 please. SWIG versions are notoriously 
incompatible even between minor versions, and wxRuby hincludesas 
numerous hacks to work around SWIG bugs.

Then you should be able to do
rake reswig && rake

If this improves things I will update the wiki to make it clearer that 
there is the specific version dependency

Thanks
alex

On Dec 27, 2007 3:12 PM, Alex Fenton <alex at pressure.to> wrote:
> >
> >  % swig -version
> >
> > SWIG Version 1.3.33
> >
> > Should I downgrade?
>
> Yes, definitely, to 1.3.31 please. SWIG versions are notoriously
> incompatible even between minor versions, and wxRuby hincludesas
> numerous hacks to work around SWIG bugs.
>
> Then you should be able to do
> rake reswig && rake
>
> If this improves things I will update the wiki to make it clearer that
> there is the specific version dependency
>
Unfortunately this gives the same result for the 1.9.2 tarball:  segfault in
rb_hash_aref, when the hash has address 0x4.

I tried to build the svn sources, but ended up here:

g++  -c  -I/home/rwa/linux-packages/wxGTK-2.8.7-static/lib/wx/include/gtk2-
ansi-release-static-2.8 -I/home/rwa/linux-packages/wxGTK-2.8.7-static
/include/wx-2.8 -D_FILE_OFFSET_BITS=64 -D_LARGE_FILES -D_LARGEFILE_SOURCE=1
-D__WXGTK__  -g -O2 -Wno-unused-function   -I
/home/rwa/linux-packages/gems/rake-0.8.0/bin -I
/home/rwa/linux-packages/gems/rake-0.8.0/lib -I
/home/rwa/linux-packages/ruby-1.8.6/lib/ruby/site_ruby/1.8 -I
/home/rwa/linux-packages/ruby-1.8.6/lib/ruby/site_ruby/1.8/i686-linux -I
/home/rwa/linux-packages/ruby-1.8.6/lib/ruby/site_ruby -I
/home/rwa/linux-packages/ruby-1.8.6/lib/ruby/1.8 -I
/home/rwa/linux-packages/ruby-1.8.6/lib/ruby/1.8/i686-linux -I . -I
/home/rwa/Distributions/wxruby-svn/doc/lib -o obj/URLDataObject.o
src/URLDataObject.cpp
In file included from src/URLDataObject.cpp:2092:
src/URLDataObject.h:19: base `Swig::Director'' with only non-default
constructor
   in class without a constructor
rake aborted!
Command failed with status (1): [g++  -c
-I/home/rwa/linux-packages/wxGTK-...]
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
http://rubyforge.org/pipermail/wxruby-users/attachments/20071227/7f8f67a3/attachment.html