[RkWard-devel] bugs

[Stefan Roediger]

Am Donnerstag, 29. März 2007 20:16 schrieb Thomas Friedrichsmeier:
> Hi,
>
> On Thursday 29 March 2007 18:38, Stefan Roediger wrote:
> > I did some bug hunting on ansari and ansari.exact test and on wilcoxon
> > tests. While going through them I found that error handling is poor
> > (obviously since there is not yet a routine in them). Regarding the
> > output they are "more" correct than before after some new lines but i. e.
> > ansari is something that's needs more work. Thomas would such a change be
> > a bug fix or a new feature? Please test them with the examples from the
> > help or own examples and than with "incorrect" data. In case you think
> > it's a bug please fix it or so. Hope you see what I mean.
>
> well, I think we have two issues here: First is that ansari.test is
> terribly picky about certain circumstances. For instance it will fail, if
> there are ties AND and an exact p value is computes AND the confidence
> interval is computed. Pretty impossible to guard against this, I'd say.
>

"Glad" to hear. So it's up to the user.

> The second issue is how to display the error messages. I agree it's not
> pretty right now, but at least the error does get shown (in a dialog). I'd
> rather not change this in the plugin, right now. We will need to work some
> on error handling after 0.4.7, anyway, see my mail "Usage of try()
> statements" a while ago.

I can remember this but didn't feel that it applies well here as solution 
therefore I said that you should decide how to handle it best. To postpone it 
after the 0.4.7 is best IMHO.

> By now, I think solutions 1b and 2c are the way to 
> go (but still interested in your comments), where warnings during execution
> of a plugin command will be redirected to the output, automatically.
>
> So anything we add in the plugin, now, would be just a temporary hack to
> make error display slightly prettier, and would need to be removed later
> on. Hence, I don't think it's something to change for 0.4.7.

I consent here. Let's keep and think about a much better solution (nobody is 
using it anyway ;) ).

>
> Regards
> Thomas
>
> P.S.: I moved the "paired" parameter to the variables tab instead of the
> options tab. I feel it's a very basic option, and should be on the first
> tab.

Fine.