Recursively test mfiles in a directory

[forkandwait]

I recently posted a hack the general list, which runs all the tests in a given 
file directory, and Carlo said that it might be good for core (I am 
flattered).  He also suggested I use the code of rundemos.  If anyone is 
interested, here is the almost latest version:

http://thread.gmane.org/gmane.comp.gnu.octave.general/29223/focus=29357

Questions (6 total):

1.  Is it bad to rely on the path mechanism like I do?  For example, rundemo() 
opens the files and reads the text, etc.  If so, why?

2.  Does anyone have presentation suggestions for the output?  I want to keep 
the cell array, since it is easy to process downstream, but I could output 
formatted text output also.  I think just a text table of "filename: passes yes/
no" would be good.

3.  I am inclined to report either passing all tests or a single fail as either 
1, 0 respectively, letting the dev investigate more detail with test(), since 
this is such a high level function.  Thoughts?

4.  Should I really use rundemo() as a model?  Why?

5.  How can I quiet the extraneous output of test()?  I don't want the 
informational messages like "???? unknown test type" etc.  Seems to me these 
should be called with warning() and a keyword (rather than printf()), which 
would allow me to silence them with warning("test", "off") (this is an awesome 
facility, btw).

6.  Is it worth it to make this into a patch?  I can do it, though I will not 
bother learning hg if no one really cares about it.

Anyway