Re: [Libcdio-devel] lsn versus lba versus msf

[Robert William Fuller]

On 11/19/2012 03:38 PM, Rocky Bernstein wrote:
> libcdio was developed far before libcddb so it probably got the mistake
> from libcdio. For libcdio's confusion, you probably have to go back to
> vcdimager.

I was hoping for a conspiracy.  That would have been cooler :-)

> But my recollection was that LBA referred to one of two things, I think
> FreeBSD was different than other places, but I'm not sure and would have to
> dig down and we had to pick one.

I see another reason that it might be this way.  The primary (ab)users 
of lba are the cdrdao, nrg, and bin/cue drivers.  I suspect negative 
offsets don't work so well with image files.

I ended up with some code in my ripper which is pretty ugly because I 
misunderstood LBA vs LSN vs MSF.  I fixed it by some ugly hack that I 
commented like this:

    // intentionally call qsc_lba_to_ascii with an lsn

So, my code does the right thing, but it took me half a day to figure 
out how and why (I wrote it in 2008 and did not leave myself good 
documentation.)  In the interest of helping future hackers, it seems 
like we should somehow fix this.

> But that was then and this is now. If someone has a suggestion and and
> there is consensus on it, things can change.

I am unsure how to move forward with this without breaking interfaces, 
but I am thinking about it.  Maybe breaking some interfaces is okay? 
(In particular the ones with lba in the name.  As it turns out, there 
aren't that many.)

Rob

> On Mon, Nov 19, 2012 at 3:31 PM, Robert William Fuller<
> address@hidden>  wrote:
>
> > There is a problem with the libcdio documentation in the section titled,
> > "6.2 block addressing (MSF, LSN, LBA)".  This is accessible on the web at
> > http://www.gnu.org/software/**libcdio/libcdio.html#Sectors<http://www.gnu.org/software/libcdio/libcdio.html#Sectors>
> > .
> >
> > The last part of the section says, "So to convert a LBA into an LSN you
> > just add 150."  To begin with, this is backwards.  The code looks like this:
> >
> >      #define CDIO_PREGAP_SECTORS  150
> >
> >          lba_t
> >          cdio_lba_to_lsn (lba_t lba)
> >          {
> >            if (CDIO_INVALID_LBA     == lba) return CDIO_INVALID_LSN;
> >            return lba - CDIO_PREGAP_SECTORS;
> >          }
> >
> > Clearly, to convert from LBA to LSN you subtract 150, not add 150.
> > Alright, so that is relatively minor.
> >
> > The thing I've struggled with the most over time is that what libcdio
> > calls LSN is what the MMC spec calls LBA.  You can see this most clearly in
> > the function cdio_lsn_to_msf.  That function follows the precise formula in
> > MMC-3 Draft Revision 10g on page 282 for converting from MMC's LBA to MSF.
> >   But, note that the function is called cdio_lsn_to_msf because what libcdio
> > calls an LSN is an LBA in the MMC spec.
> >
> > You can also see this in the MMC commands themselves.  Although they take
> > LSN parameters, there is no conversion from LSN to LBA, even though MMC
> > itself expects LBA.  Note that macros such as CDIO_MMC_SET_READ_LBA are
> > passed unmodified LSNs.  Again, that is because what libcdio calls an LSN
> > is what MMC calls an LBA.
> >
> > So what is my point?  This is confusing to people developing applications.
> >   libcdio's LSN is really MMC's LBA.  What libcdio calls LBA is not what MMC
> > calls LBA.  libcdio's LBA is basically a non-negative version of MMC's LBA
> > (because the program area starts at MMC LBA = -150 and libcdio_LBA =
> > MMC_LBA + 150.)
> >
> > Now curiously enough, libcdio is not the only library suffering from this
> > delusion.  Take libcddb.  It calls things LBAs, but expects what libcdio
> > calls an LBA rather than what MMC calls an LBA.  You can see this in sample
> > code where cdio_get_track_lba is used to call cddb_track_set_frame_offset.
> >   It almost seems like a conspiracy.
> >
> > Is there a way out of the madness?  Or do we just need better
> > documentation?
> >
> > Rob