Re: CONFIG_SND_DEBUG

[Takashi Iwai <tiwai@suse.de>]

At Wed, 5 Jun 2013 10:11:12 -0400 (EDT),
Alan Stern wrote:
> 
> On Wed, 5 Jun 2013, Takashi Iwai wrote:
> 
> > > >> Andy/Leann - apparently CONFIG_SND_DEBUG is on by default from
> > > >> upstream, and we explicitly disable it. Is there any reason why we
> > > >> do that?
> > > >
> > > > config SND_DEBUG
> > > >          bool "Debug"
> > > >          help
> > > >            Say Y here to enable ALSA debug code.
> > > >
> > > > It is off by default in upstream, and the really helpful description
> > > > would cirtainly tend to lead to it being disabled.  But if it is helpful
> > > > to your debugging efforts David then I suspect we can enable it in Saucy
> > > > and see what happens.
> > > 
> > > Okay, so then the ball is back in Takashi's area - if we're recommended 
> > > to turn CONFIG_SND_DEBUG on, why is it off by default in the upstream 
> > > Linux kernel?
> > 
> > It's not off as default.  Simply there is no default, just like most
> > of other options.
> > 
> > As already mentioned, if the device is known to work well with the
> > kernel, there is no reason to set it on.  Then it'll saves the memory
> > and code space.  That is, for custom kernels, it's good to be off.
> > But for generic kernels like distro kernel, it'd be better to take a
> > safer side with more safety checks that is built in by that option.
> 
> Given this description, the symbol's name is very misleading.  Instead 
> of being called CONFIG_SND_DEBUG, it should be called 
> CONFIG_SND_SAFETY_CHECKS or something like that.

Very true.  Actually this option has multiple implicit meanings.  It
also enables some debug things, of course, too.

> IMO, distributions are completely justified in turning off any symbol 
> which is named (and described in the help text!) as being used for 
> debugging only.

Heh, I don't blame distros but just recommend to turn it on :)

Anyway, since 3.10, safety checks with snd_BUG_ON() have been already
enabled even without CONFIG_SND_DEBUG.  But this won't give an error
message via WARN() but simply returns the error when built
CONFIG_SND_DEBUG, so it won't help for debugging much, as expected.


And, Alan, please don't work on a patch to correct Kconfig text yet.
I see on my crystalball that you'd send a patch in this very minute ;)

We'll need to work on the debug print things, so the whole kconfig
options need revisited.  Let's sort out what to be changed for
messages, split the sanity check codes and debug codes, and then go on
fixing/improving Kconfig appropriately.


thanks,

Takashi