Re: Plotting date on xaxis

["Eric Schulte" <schulte.eric@gmail.com>]

Eric S Fraga <ucecesf@ucl.ac.uk> writes:

> On Tue, 23 Feb 2010 17:05:45 -0700, "Eric Schulte" <schulte.eric@gmail.com> wrote:
>> 
>> Eric S Fraga <ucecesf@ucl.ac.uk> writes:
>> 
>> [...]
>> >
>> > I have narrowed down what causes the problem although I don't
>> > understand why (I haven't had any time to look at the elisp code).
>> > Basically, the problem with Johan's initial table was the extra empty
>> > row at the end:
>> >
>> 
>> Ah,
>> 
>> Thanks for isolating the issue.
>> 
>> It looks like the empty last row is tricking org-plot.el into thinking
>> that the entire Date column contains text labels, so it tries to use the
>> values of that column to label the data with xticlabels [1].  This is
>> useful for histograms such as the "Citas" graph in [2].
>> 
>> So all in all I'd say this is fine behavior for something like org-plot
>> which tries to guess the user's intentions.
>
> Well, good intentions is one thing, wrong behaviour is another?  If
> the user has specified "ind:1 deps:(2)", even if org-plot identifies
> column 1 as having labels for the tic marks, the generated parameter
> to the plot command should be, at best:
>
>   ... using 1:2:xticlabels(1) ...
>

I just applied a patch which results in the above plotting command, and
yields the following gnuplot error when the table contains an empty line

 "illegal day of month"

which to me seems entirely appropriate.

All of the plots form the org-plot page on worg still work with this
patch (which amounts to org-plot's test suite), so I relatively
confident that it shouldn't break any existing functionality.

One problem with org-plot's attempt to guess the best action in each
situation is that the code will get increasingly ugly as these corner
cases continue to arise, and as that happens it becomes increasingly
difficult to know a-priori if a patch will cause unforeseen problems.

Best -- Eric

>
> which is actually nonsensical as well but not as bad as what is
> currently generated:
>
>   ... using 1:xticlabels(1)
>
> which completely ignores the "deps:(2)" option!
>
> In any case, I'm not suggesting in any way that this is a critical
> bug!  Simply removing the empty line allows org-plot to work
> perfectly.