|
From: Hans-Bernhard B. <br...@ph...> - 2004-11-08 15:07:27
|
On Fri, 29 Oct 2004 em...@ru... wrote: > My thought (without having delved deeply into the source code, I must > admit) is to give the user one more type of scale much like logscale > which they can select. You're thinking too narrowly there. What we need here is a broader approach at not just this particular re-mapping idea, but the entire general concept of "translated axes". So far, gnuplot supports two such translations: identity and logarithm. Extending this to general mapping functions has been discussed in the past (I think under a heading like "generalized log axes"), but AFAIK no implementation has ever been attempted. In principle, what is needed is a pair of functions, which are inverses of each other, so gnuplot can map from data coordinates to 'plotted coordinates' for placing things, and also from plotted coordinates back to data coordinates for jobs like auto-ticking and mouse feedback. The mappings have to be monotonous, or all hell would break loose, and for this feature to be more than an isolated special-purpose tool, the mappings have to be specifiable as user-defined functions, e.g. set scale x1 sin(x1), asin(x1) There's a ton of questions to be answered before this can be implemented, mostly revolving about how to code an automatic range fix-up and automatic tick placement algorithm that work somewhat robustly in such a generalized situation. As a first step towards that, we need a re-design of the auto-ticking algorithm that makes them behave reasonably in the face of short-range log axes. For the time being, you'll just have to re-map coordinates on the go and place xtics with 'faked' numbers, as demonstrated by earlier replies. -- Hans-Bernhard Broeker (br...@ph...) Even if all the snow were burnt, ashes would remain. |