Re: Strange memory corruption when using vector (works with deque)

[Reinhold Kainhofer]

Am Samstag, 21. November 2009 07:40:50 schrieben Sie:
> 2009/11/21 Han-Wen Nienhuys <address@hidden>:
> > Can you point me to an online version of the suspect commits?
> 
> Is this what you ask for?
> 
> http://git.savannah.gnu.org/gitweb/?p=lilypond.git;a=commit;h=4ba35b32beca4
> 1dce222a95a5e0ed0c959e867dd
> 
> >> On the other hand, changing vector to deque in commit
> >> b238aa006f301fe8c194a1329af75f9db9b3024b will work just fine.
> 
> http://git.savannah.gnu.org/gitweb/?p=lilypond.git;a=commit;h=b238aa006f301
> fe8c194a1329af75f9db9b3024b

Yes, these are the two commits I'm talking about.

The first one is what causes the crash. I don't see anything suspicious about 
it. Head_event_tuple stores everything by value, and the Stream_event pointers 
inside the Audio_element_info are also just stored as values, but the memory 
never allocated, deleted or duplicated.

valgrind shows an invalid read only in 
Tie_performer::stop_translation_timestep, but the memory corruption happens 
much earlier (in particular, after the first push_back call, the size() will 
already report a totally wrong value in 
Tie_performer::acknowledge_audio_element.

Cheers,
Reinhold
-- 
------------------------------------------------------------------
Reinhold Kainhofer, address@hidden, http://reinhold.kainhofer.com/
 * Financial & Actuarial Math., Vienna Univ. of Technology, Austria
 * http://www.fam.tuwien.ac.at/, DVR: 0005886
 * LilyPond, Music typesetting, http://www.lilypond.org