gnuplot-beta

timecolumn()

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

Hello, guys.

A while ago, Ethan asked what the secret function timecolumn() was 
about.  Then, more recently, a bug reporter ended up needing it as the 
way to work around a limitation, but it didn't work...  I just patched 
it and added documentation, but some open issues remain.

timecolumn() is sitting between a rock and a hard place because of the
way I implemented multiple separate timefmt specifications a long time 
ago (part of the 'axis patch'), yet another feature that was never 
actually documented.  Each axis can have its own timefmt string.

But parsing of time/date input doesn't really have anything to with axes 
--- it deals with datafile columns.   Taking into account extended using 
specs, the relation between column numbers and axes is many-to-many, so 
the two really have no particular link to each other.

The obviously correct solution would be to pass a timeformat string 
directly as a second argument to timecolumn().  Back when I added 
per-axis timefmt strings, that wasn't possible.  Nowadays, with the 
stringvariables patch integrated, it probably is.

So: anybody interested in implementing support for

	plot 'file' using (timecolumn(1,'%Y-%m-%d %H:%M:%S')):5

or
	tf1='%Y-%m-%d %H:%M:%S'
	plot 'file' using (timecolumn(1,tf1)):5

Re: timecolumn()

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

On Thursday 05 May 2005 07:52 am, Hans-Bernhard Broeker wrote:
> 
> The obviously correct solution would be to pass a timeformat string 
> directly as a second argument to timecolumn().  Back when I added 
> per-axis timefmt strings, that wasn't possible.  Nowadays, with the 
> stringvariables patch integrated, it probably is.

Yes. Passing a string as a 2nd argument (or even a string-valued
function) is easy.  The hard part would have been to retain
backward-compatibility with scripts using only one argument.
Fortunately timecolumn was a secret undocumented function,
so there are no such scripts :-)

But before anyone plunges in, I point out that there is probably
another approach that should be considered.  If the task is
specifically to read time info from the file and associate it
with axis labels, then I think this is almost possible already
by passing a sting-function to the [xyz]xticlabels option.
The following doesn't quite work, but could be made to work:

   timeformat = "some time format"
   plot 'foo' using 4:5:xtic(gprintf(timeformat,column(1)))

The trick would be to extend routine plot_ticlabel_using(int axis)
in datafile.c so that it parses the argument to xtic() as a
general expression rather as an integer constant.  That's something
I meant to do originally, but never got around to it.  It would
have uses beyond timeformats, of course.


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

Re: timecolumn()

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

Ethan Merritt wrote:

> But before anyone plunges in, I point out that there is probably
> another approach that should be considered.  If the task is
> specifically to read time info from the file and associate it
> with axis labels, 

It's not just for labels --- it's for reading timeseries data
in typical human-readable format, i.e. hours, days, and whatnot,
and turning it into continuous data.  timecolumn(n) is supposed to do 
the same thing inside an extended using spec for time/date data as $<n> 
or column(n) already do for ordinary numbers.  The example from that bug 
report I mentioned was time-dependent rescaling of stock prices across 
the change from D-Mark to Euros:

	set xdata time
	set timefmt x '%d/%m/%y'
	to_euro(year,value) = (year < 2002) ? value / 1.95 : value
	plot 'data' using 1:(to_euro(tm_year(timecolumn(1)),$2))

That one works with today's version --- but it's a close shave.