Andrew de Quincey wrote:
On Sunday 12 Sep 2004 17:43, Johannes Stezenbach wrote:
I've seen the source of [name deleted]'s STB software, which has
absolutely nothing to do with Linux. And guess what? They do
device probing, too. Because in real life one piece of software
has to run on a multitude of hardware revisions.
I can't wait until [name deleted] tells you that the new
XY frontend is cheaper that Samsung's, and could you not add
device probing to your USB thingy firmware.
The only thing that is "broken" about kernel I2C autoprobing is
that it is too simple. But hey, it worked for all those years ;-)
Now, for those cases where the simple autoprobing indeed cannot
be made to work, the addition of a NOPROBE flag and doing more
sophisticated probing seems like a straight forward thing to me.
Oh aye, we'll always have to do probing. However, the problems we're having
come from the i2c core probing for every single random i2c device that
happens to have a registered driver loaded. I bet [name deleted]'s STB
software only probes for i2c devices that they know their device might see.
No problems with *that* - it can fix any clashes itself, as it knows what it
expects.
Oh, I thought we were already beyond that point, that it was agreed that
we need the I2C device class field to keep non-DVB stuff away from DVB-I2C buses.
(Really, I thought that issue had been sorted out months ago by patches
posted by Michael Hunold to lkml.)
http://groups.google.com/groups?hl=en&lr=&ie=UTF-8&selm=1Ajn2-5WU-1%40gated-at.bofh.it
I was under the impression that only tda9987 vs. cx22702 was
actually a more difficult to solve problem, currently. And my
thinking was (I could easily be wrong, though), that the probe
functions for those modules need to do something more than just
ping for some device at an address.
As for the changes Holger suggests w.r.t frontend libraries, they're aimed at
cleaning up the frontend code and making it more flexible. Something which
I've wanted to look into for some time.
Sure, I trust you to do the right thing to improve the frontend code.
I just don't like Holger's bashing on kernel I2C code, which IMHO
is unjustified. IMHO there are no unfixable problems.