On Thu, Jan 29, 2009 at 10:57:03AM +0100, Jean Delvare wrote:
> Hi Andre,
>
> On Thu, 29 Jan 2009 10:09:45 +0100, Andre Prendel wrote:
> > sensors_error() returns NULL if errnum is not valid. Returning
> > "Unknown error" in this case, makes output clearer than:
> >
> > foo: (null)
> >
> > and prevents error output/logging from segmentation faults.
> >
> > What do you think?
> >
> > Andre
> >
> > Signed-off-by: Andre Prendel <andre_prendel at gmx.de>
> > ---
> >
> > --- lm-sensors-dev/lib/error.c 2009-01-26 17:43:43.000000000 +0100
> > +++ my-lm-sensors/lib/error.c 2009-01-28 22:49:53.000000000 +0100
> > @@ -33,7 +33,7 @@ void (*sensors_fatal_error) (const char
> > sensors_default_fatal_error;
> >
> > static const char *errorlist[] = {
> > - /* Invalid error code */ NULL,
> > + /* Invalid error code */ "Unknown error",
> > /* SENSORS_ERR_WILDCARDS */ "Wildcard found in chip name",
> > /* SENSORS_ERR_NO_ENTRY */ "No such subfeature known",
> > /* SENSORS_ERR_ACCESS_R */ "Can't read",
>
> Out of curiosity: did you actually hit this problem in real life?
No, I didn't. It's just theoretical.
> I was about to say that the caller was supposed to check for NULL, but
> apparently neither sensors nor sensord nor even libsensors does, so...
IMO error reporting should be very stable. If the caller checks
return value carefully it's fine too. Fprintf() isn't a problem. But I
don't know what about vsnprintf(), syslog().
> I guess you're right, having an error string instead of NULL would be
> better. Actually that's what libsensors 2.x did, I can't remember why I
> changed that in 3.x:
> http://www.lm-sensors.org/changeset/4854
>
> Probably I thought it was not supposed to happen, but I guess it may
> actually happen if mixing a recent application with an older version of
> libsensors.
>
> So I am pretty much inclined to apply your patch.
Vous avez le choix. :)
Andre
> --
> Jean Delvare
