gnuplot-beta

empty data-files

[Jacques B. <jac...@on...>]

When you try to plot an empty data-file with "plot", the output terminal 
is not updated, and you get this message:

"no valid data points found in specified file".

Why not.

But if you try to plot 2 data-files, one empty the other not, you get 
the same answer, and that is not OK. There _is_ something to plot. That 
situation may arise if the data-files and the gnuplot script are 
generated automatically.

So I think that in plot2d.c:

int_error(c_token, "no valid data points found in specified file");

should be replaced with:

int_warn(c_token, "no valid data points found in specified file");

Re: empty data-files

[Petr M. <mi...@ph...>]

> But if you try to plot 2 data-files, one empty the other not, you get
> the same answer, and that is not OK. There _is_ something to plot. That
> situation may arise if the data-files and the gnuplot script are
> generated automatically.
>
> So I think that in plot2d.c:
>
> int_error(c_token, "no valid data points found in specified file");
>
> should be replaced with:
>
> int_warn(c_token, "no valid data points found in specified file");

This behviour is fine with me, just there should be an int_error("nothing to
plot") if there was only a single (empty) file to plot. Please update your
patch, add taking care of 3d plots.

What should happen here:
 plot x*x, 'empty.dat'

?? Currently it reports an empty x-range.

---
PM

Re: empty data-files

[Ethan M. <merritt@u.washington.edu>]

On Thursday 13 January 2005 07:53 am, Petr Mikulik wrote:
> >
> > So I think that in plot2d.c:
> >
> > int_error(c_token, "no valid data points found in specified file");
> >
> > should be replaced with:
> >
> > int_warn(c_token, "no valid data points found in specified file");
> 
> This behviour is fine with me, just there should be an int_error("nothing to
> plot") if there was only a single (empty) file to plot. Please update your
> patch, add taking care of 3d plots.

It's more complicated that that.  The first thing that will happen
if you just change int_error into int_warn is likely to be a 
divide-by-zero error as the autoscaling fails.  Beyond that, there
is the question of consistent behaviour of the plot.
For instance, "skipping" a plot  should not change the color
assignments of the remaining plots.

So I think that to do this properly you'd have to mark the
plot empty (or just use the fact that npoints == 0), but carry it
along and test for this explicitly at every stage of the plotting
process. I think that would be straightforward, but it's a lot
more work than just changing one call.


-- 
Ethan A Merritt       merritt@u.washington.edu
Biomolecular Structure Center
Mailstop 357742
University of Washington, Seattle, WA 98195

Re: empty data-files

[Jacques B. <jac...@on...>]

Petr Mikulik <mikulik <at> physics.muni.cz> writes:
> This behviour is fine with me, just there should be an int_error("nothing to
> plot") if there was only a single (empty) file to plot. Please update your
> patch, add taking care of 3d plots.

In that case, you get this:
         warning: no valid data points found in specified file
         all points y value undefined!
Clear enough, isn't it?

> What should happen here:
>  plot x*x, 'empty.dat'

You get this:
         warning: no valid data points found in specified file
         x range is invalid
So you have to use, for example:
plot [-1:1] x*x, 'empty.dat'

The titles x*x and 'empty.dat' are shown, but only x*x is plotted. Not bad.

----------------------------------------------------------------------------
Ethan Merritt <merritt <at> u.washington.edu> writes:
> It's more complicated that that.  The first thing that will happen
> if you just change int_error into int_warn is likely to be a 
> divide-by-zero error as the autoscaling fails.

That doesn't happen. If an empy datafile is plotted alone, you get:
         warning: no valid data points found in specified file 
         all points y value undefined!

> Beyond that, there
> is the question of consistent behaviour of the plot.
> For instance, "skipping" a plot  should not change the color
> assignments of the remaining plots.

There is no change of color.
There is no "skipping" (the title of an empty plot is still there).

----------------------------------------------------------------------------
Hans-Bernhard Broeker <broeker <at> physik.rwth-aachen.de> writes:
> But the plot command, as given, still is wrong.  An entire dataset is 
> missing --- there'll consequently be all kinds of strange behaviour on the 
> plot, which are pretty much guaranteed to make the resulting diagram at 
> least hard to interpret, possibly unusable.

Only the points are missing. The title of each empty dataset is still there. So
the plot command is _not_ wrong.

> If so, then it's a grave bug in the generator that it didn't notice it
> just created a datafile without any data in it.

The bug is not in the generator but in gnuplot. An empty datafile is not wrong.

> That would cause a new problem: gnuplot would now try to generate a plot, 
> even if *none* of the files in the plot contain any data.

No plot is generated in that case. You just get this message:
         all points y value undefined!

And in fact, that behavior doesn't suit me: I would like gnuplot to generate a
plot with no point in it, but with a title and labels (the x and y ranges must
be set in that case of course). I use gnuplot to plot real time data, and I draw
only the data of the last minutes. If in that time there is no data at all, I
want to see the plot all the same (and I certainly don't want gnuplot to stop).

---------------------------------------------------------------------------
SUMMARY: I can't see any drawback in my small patch. But it is no enough. It
should be possible to plot an empty datafile (everything but the points).

Re: empty data-files

[Hans-Bernhard B. <br...@ph...>]

On Sat, 8 Jan 2005, Jacques Bouchard wrote:

> But if you try to plot 2 data-files, one empty the other not, you get 
> the same answer, and that is not OK. There _is_ something to plot. 

But the plot command, as given, still is wrong.  An entire dataset is 
missing --- there'll consequently be all kinds of strange behaviour on the 
plot, which are pretty much guaranteed to make the resulting diagram at 
least hard to interpret, possibly unusable.


> That situation may arise if the data-files and the gnuplot script are
> generated automatically.

If so, then it's a grave bug in the generator that it didn't notice it
just created a datafile without any data in it.

> So I think that in plot2d.c:
> 
> int_error(c_token, "no valid data points found in specified file");
> 
> should be replaced with:
> 
> int_warn(c_token, "no valid data points found in specified file");

That would cause a new problem: gnuplot would now try to generate a plot, 
even if *none* of the files in the plot contain any data.

-- 
Hans-Bernhard Broeker (br...@ph...)
Even if all the snow were burnt, ashes would remain.