Re: libdvdread -- off by 1 errors (not your fault)

Wed Aug 28 2002, James Tappin wrote:
> There would appear to be at least 1 dvd authoring package out there that
> is generating DVDs with off-by-one errors in the lengths of some fields.
> Since the validity of these fields is check by an assert in libdvdread, to
> play said DVDs I have had to comment out the relevant assertions (and
> print the values instead). 
> 
> Line numbers refer to release 0.93.
> 
> The first is at line 950 of ifo_read.c:
>  assert(n > 0 && (n % 4) == 0;
> These DVDs have n=61
> 
Haven't seen this before.  

> the other is at line 184 of nav_read.c:
>  assert(dsi->dsi_gi.zero1 == 0);
> these have dsi->dsi_gi.zero1 == 1
> 
Have seen this, perhaps among others, it's produced by the MacOSX DVD
creation software.  Which seems to becomming ever more popular as 
more people burn their own DVD-R.

> This affects both ogle and xine-dvdnav playback (xine-dvdnav also fails
> another assert in xine-lib that I've not yet tracked down in detail).
> While it's pretty obvious to me that the author of the writing software
> has double-allowed for the null at the  end of a string, this is probably
> a commercial package for windows or mac so I doubt there's much chance of
> exterminating it.
> 
If some one could confirm which program it is and at least report it as
a bug... 

> Is it reasonable that these asserts be downgraded to some kind of warning
> message as the DVD plays correctly in ogle when they are ignored.
> 
Yes, and I've been intending to do this for a while now.  It's not always
as simple as to just remove the assert.  Ofthen it needs to be replaced
by both a check to emit a warnings and some new code to patch up the values
to some thing safe that won't crash or currupt memmory later.

-- 
Håkan Hjort