Re: stubs for missing functions

[Jaroslav Hajek]

On Tue, Mar 23, 2010 at 11:33 PM, John W. Eaton <address@hidden> wrote:
> We frequently see people sending bug reports or questions on the help
> list about functions that are available in Matlab but missing from
> Octave.  I think we could avoid many of these problems with the
> attached patch.  The idea is that for every core Matlab function that
> we know about that is missing from Octave, we create a simple stub
> function which when run fails with the error:
>
>  error: The FCN function is not yet implemented in Octave.  Please read
>  `http://www.octave.org/missing.html' to find out how you can help to
>  contribute a working version of this function.
>  error: called from:
>  ...
>
> Likewise, if someone asks for help about one of these functions, they
> will get the following help text:
>
>  `FCN' is a function from the file .../FCN.m
>
>   -- Function File:  FCN ()
>       This function is not yet implemented in Octave.  Please read
>       `http://www.octave.org/missing.html' to find out how you can help
>       to contribute a working version of this function.
>
> (In both cases, FCN will be replaced by the actual function name).
>
> The referenced web page, http://www.octave.org/missing.html remains to
> be written.  I intend to explain that there are several possible
> reasons why the function is missing:
>
>  The function may exist in Octave Forge but has not yet been adopted
>  as a part of the core Octave distribution.
>
>  The function may be new to Matlab.  Perhaps the Octave community
>  has no knowledge of it.
>
>  Octave is developed by volunteers who tend to contribute the
>  functions which they need, and so far no one has contributed a
>  version of this function.
>
> My advice will be something like
>
>  1. check Octave Forge, as a similar function may be available there.
>
>  2. consider writing a function and contributing it.
>
>  3. consider paying someone to implement the function for you.
>
>  4. file a feature request in the bug tracker.
>
> Comments?
>
> Are there any objections to committing this patch?
>
> Thanks,
>
> jwe
>
>


Besides Soren's comments, I have a couple more:

1. I'm a bit sceptical that this will help. I suppose most users
realize that the missing functions are not implemented.
2. Several times when writing code compatible with Matlab I used tricks like
if exist ("somematlabfunc")
  ## Use it.
else
  ## Work it around.
end

this will now be impossible.
3. I am almost sure when implementing new functions I will forget to
delete them from the list. (Not that this is serious :)

4. This may give users a false impression that we eventually intend to
implement all of these functions, which I think is not true.
Some are handled by Octave forge instead, others make no sense with
Octave (isstudent etc.).

Here's an alternative suggestion:
Let's create a global pseudo-variable missing_function_handler that,
if set, will be called with the missing identifier name
just prior to raising an error. By default, we'll point it to an
m-file handler. This has several advantages:

1. much less code
2. is configurable (relieved sigh)
3. doesn't spoil tab completion or shadowing
4. we can easily specialize it for certain classes of functions. We
can point users to OctaveForge extensions or suggest alternatives.
(for instance, we can write "textscan is not implemented, try using
textread" or "gsvd is not part of core Octave, see linear-algebra
package on OctaveForge"). And all this can happen in one function.

what do you think?

-- 
RNDr. Jaroslav Hajek, PhD
computing expert & GNU Octave developer
Aeronautical Research and Test Institute (VZLU)
Prague, Czech Republic
url: www.highegg.matfyz.cz