[linux-dvb] [PATCH] Future of DVB-S2

Hans Werner HWerner4 at gmx.de
Fri Aug 29 18:43:52 CEST 2008


> > Now, to show how simple I think all this could be, here is a PATCH
> implementing what
> > I think is the *minimal* API required to support DVB-S2.
> >
> > Notes:
> >
> > * same API structure, I just added some new enums and variables, nothing
> removed
> > * no changes required to any existing drivers (v4l-dvb still compiles)
> > * no changes required to existing applications (just need to be
> recompiled)
> > * no drivers, but I think the HVR4000 MFE patch could be easily adapted
> >
> > I added the fe_caps2 enum because we're running out of bits in the
> capabilities bitfield.
> > More elegant would be to have separate bitfields for FEC capabilities
> and modulation
> > capabilities but that would require (easy) changes to (a lot of) drivers
> and applications.
> >
> > Why should we not merge something simple like this immediately? This
> could have been done
> > years ago. If it takes several rounds of API upgrades to reach all the
> feature people want then
> > so be it, but a long journey begins with one step.
> 
> This will break binary compatibility with existing apps.  You're right
> -- those apps will work with a recompile, but I believe that as a
> whole, the linux-dvb kernel and userspace developers alike are looking
> to avoid breaking binary compatibility.

Michael,
thank you for your comment.

I understand, but I think binary compatibility *should* be broken in this case. It makes
everything else simpler.

I know that not breaking binary compatibility *can* be done (as in the HVR4000 SFE and
MFE patches) but at what cost?  The resulting code is very odd. Look at multiproto which 
bizarrely implements both the 3.2 and the 3.3 API and a compatibility layer as well, at huge cost
in terms of development time and complexity of understanding. The wrappers used in the MFE
patches are a neat and simple trick, but not something you would release in the kernel.

If you take the position the binary interface cannot *ever* change then you are severely
restricting the changes that can be made and you doom yourself to an API that is no longer
suited to the job. And the complexity kills. As we have seen, it makes the whole process grind to a
halt. 

Recompilation is not a big deal. All distros recompile every application for each release (in fact much more frequently -- updates too), so most users will never even notice.  It is much better to make the right, elegant changes to the API and require a recompilation. It's better for the application developers because they get a sane evolution of the API and can more easily add new features. Anyone who
really cannot recompile existing userspace binaries will also have plenty of other restrictions and
should not be trying to drop a new kernel into a fixed userspace.

I would be interested to hear your opinion on how we can move forward rapidly.

Regards,
Hans

-- 
Release early, release often. Really, you should.

GMX Kostenlose Spiele: Einfach online spielen und Spaß haben mit Pastry Passion!
http://games.entertainment.gmx.net/de/entertainment/games/free/puzzle/6169196



More information about the linux-dvb mailing list