Re: References to obsolete functions

[Andreas Weber]

Am 27.10.2013 16:19, schrieb Rik:
> On 10/27/2013 01:42 AM, address@hidden wrote:
>> Message: 5
>> Date: Sun, 27 Oct 2013 09:42:15 +0100
>> From: Andreas Weber <address@hidden>
>> To: octave maintainers mailing list <address@hidden>
>> Subject: removed but still mentioned betai and spchol2inv in 3.8
>> Message-ID: <address@hidden>
>> Content-Type: text/plain; charset=ISO-8859-15
>>
>> Dear maintainers, sorry for the late post.
>> Following 2 functions which were removed in 3.8 but still used in the
>> documentation or in error messages. I havend provided a patch yet
>> because I need some help to figure out what the right way is.
>>
>> hg id c6891317ff27
>>
>> betai:
>> ./scripts/statistics/tests/manova.m:37:## (Currently NOT because the
>> fcdf respectively betai code is too bad.)
>>
>> Should we just remove the "betai" part and keep the "bad fcdf code" part?
> 
> I'm not sure how relevant this message is anymore.  The statistical
> functions were overhauled by me for Octave version 3.6 and just saying
> betai is bad doesn't do much for the user.  I would re-phrase to say that
> the values are currently not calculated because the code has not been
> written for that case.

Dear Rik,
I tried to gather the history of manova.m:

The texinfo part, which is not visible to the user
"## Three test statistics (Wilks, Hotelling-Lawley, and Pillai-Bartlett)
## and corresponding approximate p-values are calculated and displayed.
## (Currently NOT because the fcdf respectively betai code is too bad.)"

and the disabled (if (0)) code for Hotelling-Lawley Test and
Pillai-Bartlett has been added with 781c930425fd in 1998

In changeset 6f068e3f3f9c from 2011 f_cdf was replaced with fcdf.

I tried to figure out how manova works and enabled the two other tests
but I can't say if the results are correct. Surely fcdf and betainc made
some improvement like then. Until someone picks this up, how is
something like this:

## The Hotelling-Lawley and Pillai-Bartlett test statistics have been
## added in 1998 but are disabled since then because fcdf and betainc
## weren't fully implemented.  Please feel free to improve this.

>> spchol2inv was removed and chol2inv now also handles sparse matrix:
>> <snip>
>> So a call to chol2inv shows an error "spchol2inv ...."
>> <snip>
>> Should we just remove "spchol2inv" or is it important for the user to
>> remind him, that he called chol2inv with a sparse matrix?
> 
> We should just use chol2inv and get rid of the 'spchol2inv'.  I would do
> this by re-phrasing the error message to name the input that is in error
> and why it is in error.
> 
> The help string for chol2inv is
> 
> Loadable Function:  chol2inv (U)
> 
> So I would use the expression "U must be a square matrix" and the final
> warning message will be "error: chol2inv: U must be a square matrix"

I pushed this changeset
http://hg.savannah.gnu.org/hgweb/octave/rev/eca761671f16