Re: tests in log-scale plots

[John W. Eaton]

On  6-Sep-2011, c. wrote:

| 
| On 6 Sep 2011, at 16:37, John W. Eaton wrote:
| 
| > On  6-Sep-2011, Carlo de Falco wrote:
| > 
| > Also, unless there is an unexpected error, your test doesn't catch
| > incorrect results without manual inspection, does it?
| 
| actually it does, octave segfaults when that test is run

Yes, it catches an unexpected error.  I meant that it does not tell
you whether the plot is correct.

| >  I think that if
| > a test can't catch incorrect results, then it should be a demo instead.
| OK
| 
| > It's fine to add a demo.  But tests should only succeed or fail, not
| > produce any output or graphics.
| 
| I agree, I think the correct way to do this is to use 
| 
| figure (1232, "visible", "off");
| 
| and then check for properties of the figure as suggested by Kai

I agree that this is somewhat useful as a test, but it still does not
tell that the plot is rendered correctly.  That's why I think the
demos are still useful.  Perhaps more useful at this point than for
the graphics functions than writing a bunch of automated tests that
show that properties are set correctly.

jwe