Re: Common gripe for unavailable functions when library does not exist

[John W. Eaton]

On 01/15/2016 02:42 PM, Rik wrote:
> 1/15/16
>
> jwe,
>
> How should we handle functions, principally dynamically linked ones, that
> may not exist for a given installation of Octave?
>
> Currently we have at least three strategies.
>
> Strategy 1) #ifdef/code/#else/error (...)/#endif.
>
> See dldfcn/amd.cc for example.
>
> Strategy 2) #ifdef/code/#else/gripe/#endif.
>
> See __glpk__.cc which uses gripe_not_supported.
> See __magick_read__.cc which uses gripe_disabled_feature
>
> Strategy 3) Declare macro that expands to error message and use that in
> place of error or gripe.
>
> See audiodevinfo.cc which has
>
> #define NO_PORTAUDIO_MSG \
>    error ("portaudio not found on your system and thus audio functionality
> is not present"); \
>    (void) args;  /* silence compiler warning "unused parameter" */
>
> It seems like there should be a common response to the same situation which
> says to me that a single gripe_XXX function should be made to handle all of
> these cases.

I agree, using a gripe function to make the error message uniform would 
be good.  I think gripe_disabled_feature is a better attempt than 
gripe_not_supported, at least for these features that rely on external 
>libraries and that might be disabled.

> A second issue is whether we want to suppress the warning messages about
> the unused variable "args" when the DEFUN_DLD macro is used and the library
> is not supported.  It is annoying to see these messages.  On the other
> hand, the obvious way to suppress them leads to code bloat.
>
> #ifdef HAVE_AMD
> DEFUN_DLD (amd, args, nargout,
> #else
> DEFUN_DLD (amd, /* args */, nargout,
> #endif
>
> Or we could have something like what Mike did with a compiler trick to
> silence the warning
>
> (void) args;
>
> but I have to say that even with the comment I was sort of confused as to
> what was going on here.

This is a case where a macro might help.  Then you could just write

  OCTAVE_UNUSED (args);

I already defined an OCTAVE_UNUSED macro (for the unused __attribute__) 
but that could be redefined to work this way instead.

If we do this, could we use a C++-style cast instead?

> The GNU coding standards (http://www.gnu.org/prep/standards/standards.html)
> suggest not going to great lengths:
>
> "Don’t make the program ugly just to placate static analysis tools such as
> lint, clang, and GCC with extra warnings options such as -Wconversion and
> -Wundef. These tools can help find bugs and unclear code, but they can also
> generate so many false alarms that it hurts readability to silence them
> with unnecessary casts, wrappers, and other complications. For example,
> please don’t insert casts to void or calls to do-nothing functions merely
> to pacify a lint checker."
>
> So is the extra #ifdef around the DEFUN_DLD worth it or not?

The warnings are kind of annoying, but I wouldn't work too hard to 
eliminate them.  It's more important to me that Octave compiles without 
warnings when all of the features are available and enabled.  It doesn't 
bother me too much if there are warnings when some features are disabled.

jwe