Re: 3.4.1 release and FLTK

[Michael D Godfrey]

On 04/04/2011 04:35 PM, Rik wrote:

On 04/04/2011 03:35 PM, Michael D. Godfrey wrote:

> I would like to add: https://savannah.gnu.org/bugs/?32980
> 
> This is the fact that OpenGL accepts doubles but converts
> to single.  Thus if some data points are outside the float range they are
> plotted incorrectly.  Also, if axes are specified outside the float range they
> are not handled correctly.  This case is pretty obvious, but the data
> points getting "lost" is not.  No warning or error is issued in these cases.
> 
> Gnuplot handles this correctly.  So, people expect correct double results.
> 
> I would say that until this is fixed the fltk backend cannot be considered
> to be reliable.

I absolutely agree, but I don't think we need this for the 3.4.1 release.
I thought the switchover from gnuplot to FLTK as the default was expected
only before the next major release, i.e., 3.6.  Also, I'm not sure that
this has a simple fix.  Do we just need to add listeners on the
xdata,ydata,zdata channels and warn if values exceed 1e38?

--Rik

Good!  This means that for 3.4.1 we should make clear that there are problems, and
mention this specifically and very clearly.

As to fixing the problem, I do not know enough about the flow from the Octave
call to sending data to OpenGL to say for sure.  One obvious possibility is to
check the data immediately in each plotting function.  At this point an error return
would be the right thing.  There is no point generating a plot if it is incorrect.
Range checks on the axes and data structures should be pretty straightforward
and not a lot of overhead.  They can be made conditional on OpenGL capability
so that it will be simple to make use of full double use if OpenGL ever provides
it, and will not be used if the gnuplot is in use.

Michael