Define meaning of empty Axis.unit

[henrikt-ma]

This PR tries to address the problem that it is currently not allowed to leave the unit of an Axis empty, which in turn must be interpreted as also the axis specifications x and y of a Plot being mandatory. I don't think it was intended to add the burden of specifying axes for every Plot, and specifying that it means to leave the unit empty opens up for an interesting feature currently not present in the MCP.

This PR is a follow-up to #2579.

[otronarp]

Looks good to me.

[HansOlsson]

For background see my comments in #2633

The following: "When axis bounds are specified by the user, on the other hand, a tool may choose a unit for the variable such that the range of the variable values (expressed in the chosen unit) fit nicely with the range of the unitless axis."
to me suggests that one can give min/max without unit and have the unit adapt to that in some way.

I may be misinterpreting it, but it should at least be clear that min/max must be given in the unit and if the unit is missing or ignored so are min/max.

[henrikt-ma]

Quoting @HansOlsson regarding these lines, so that we get a subthread for discussion:

The following: "When axis bounds are specified by the user, on the other hand, a tool may choose a unit for the variable such that the range of the variable values (expressed in the chosen unit) fit nicely with the range of the unitless axis."
to me suggests that one can give min/max without unit and have the unit adapt to that in some way.

I may be misinterpreting it, but it should at least be clear that min/max must be given in the unit and if the unit is missing or ignored so are min/max.

[henrikt-ma]

to me suggests that one can give min/max without unit and have the unit adapt to that in some way.

No. The of the unitless axis at the end of the sentence is there to make it clear that the axis still doesn't have a unit; the units are always associated with the variables when unit is empty. It is true that it becomes up to the tool to make a clever choice of unit when min and max are given, but unit is empty; a bad choice will result in a poor fit between the variable values and the axis range.

I may be misinterpreting it, but it should at least be clear that min/max must be given in the unit and if the unit is missing or ignored so are min/max.

No. A bad unit is an error, but there is still a suggested way to deal with it. More importantly, min and max do have a well defined meaning when unit is empty, as described above. That min and max are given in unit is expressed by the pseudocode record, expressed in one way in this PR, but more clearly in #2633:

record Axis
  Real min "Axis lower bound, in 'unit'";
  Real max "Axis upper bound, in 'unit'";
  String unit "Unit of axis tick labels";
  String label "Axis label";
end Axis;

[HansOlsson]

My suggestion is that if we don't recognize the unit, it's incorrect, or if it is empty we ignore the min and max as well.

The tools must be able to handle missing unit and min/max - so just fall back to that instead of having advanced logic to handle such problems. It seems simpler and gives more predictable result, and it encourages users to either specify the plots in detail - or not at all.

[henrikt-ma]

I don't see the reason to ignore min and max when unit is empty (there is no complicated logic involved in making use of them). The possibility to have variables of different units on the same axis is quite useful, and you get this mode by using an empty unit. In this mode, it is still perfectly meaningful to specify min and max boundaries for the unitless axis.

Something I think we could clarify, however, is the recommendation for what to do when unit is not recognized: Then, the min and max should be ignored (instead of being applied to the unitless axis) — I'll make a separate comment for this.

[henrikt-ma]

Here's a plot with unitless axis combined with explicit setting of min and max to get 0 vertically centered:

[HansOlsson]

If we want to plot variables in different units with given min/max the most reasonable would be to say that they use the given unit (or displayUnit) of the variables; nothing fancy. If I for the curve above gave min=0, max=0.4 it's not clear if the tool should use "V" or "kV" - and to me it seems as if the specification recommends "kV" (and "kW") since more of the curve is shown in that case.

The other option would be something a lot more advanced (allow different scalings).

[henrikt-ma]

Your interpretation of the recommendation is in line with the intention, and I think we need to stick to mappings to the axis range that can be expressed by the choice of unit.

I would prefer to not rely directly on the variable's displayUnit here, since the displayUnit was probably set with just the variable at hand in mind, not how the range of the variable would match the ranges of some other variable when added to the same unitless plot axis. If you don't like the tool-dependent choice of unit here, the compromise that comes to mind is to extend Curve with (optional) xDisplayUnit and yDisplayUnit, which would default to the displayUnit of the x and y, and then require that xDisplayUnit and yDisplayUnit are used on unitless axes.

My main objection to introducing xDisplayUnit and yDisplayUnit is that it could be confusing that these only come into play when plotted against a unitless axis; the amount of added value compared to added specification text (after adding clarifications to avoid the potential confusion) isn't that great. The other objection is that introducing xDisplayUnit and yDisplayUnit would mean that tools don't get the chance to make an intelligent choice of unit as in the current proposal.

[otronarp]

If we want to plot variables in different units with given min/max the most reasonable would be to say that they use the given unit (or displayUnit) of the variables; nothing fancy.

But the wording doesn't require you to do anything fancy, it only gives you enough leeway to do fancy stuff if you want to.