Add grouping property to traces

[rreusser]

It's become clear that the ids field is insufficient. It's needed for object constancy while animating, but duplicate ids result in disappearing markers. This is the expected behavior; it just precludes overloading the ids field for something like a filter transform, where duplicate data is perfectly reasonable.

This PR:

• adds a grouping property to traces and explicitly notes that it has no inherent effect
• notes in the docs that duplicate ids are invalid or at least have likely-unintended consequences
• adds calcdata tests for casting ids

Feedback on naming and functionality is welcome. I don't like adding attributes, but this is very lightweight, seems necessary, and explicitly notes that it does not have any effect unless hooked up elsewhere by the user (e.g. transforms).

See also: #1027
cc: @chriddyp @etpinard @alexcjohnson

[rreusser]

After talking with @etpinard, I think a slightly more sophisticated/general approach is best.

[rreusser]

@etpinard @bpostlethwaite I guess this PR's relevance has been renewed! I'll check this over, but at a glance, looks like it implements grouping with tests, plot schema and everything……

[etpinard]

adding groupings (I like groups better by the way) to all trace types would be even better.

[rreusser]

Cool. I'll switch the name, extract the attrs, and add it across the board.

[etpinard]

why do we need groupings in gd.calcdata?

[etpinard]

Cool. I'll switch the name, extract the attrs, and add it across the board.

just add it here: https://github.com/plotly/plotly.js/blob/master/src/plots/attributes.js

all trace attributes are based off it - see here.

[rreusser]

@etpinard is there a way to automatically cast these to strings, or do I have to manually loop through the array, if provided, and cast them?

[rreusser]

Oops. Meant to ask for groups, but same question applies to both.

[etpinard]

is there a way to automatically cast these to strings

Not at the moment. Casting within data_array fields is done during the calc step at the moment.

As groups will only be used in filter transforms at first, we could simplify cast groups items in the filter calcTransform handler.

Alternatively, we could add a block to Plots.doCalcdata, casting groups items once and for all. In order to behave correctly during restyle and extendTraces, this would (unfortunately) require adding groups to the recalc-attribute list here.

I have no preference. Your call.

[rreusser]

Actually, I think it's best to leave unspecified… It could be country and you could filter by inclusion in a list, or it could be a value and you could filter numerically.

[rreusser]

As opposed to ids, which must be strings.

[etpinard]

Actually, I think it's best to leave unspecified…

very good point.

[etpinard]

TODO:

• add filter calc transform here
• add Plotly.restyle(gd, 'groups', [/* */]) making filter transforms with filtersrc: 'groups' work as expected here

[etpinard]

Lovely description. Thanks!

[rreusser]

@etpinard Glad to make those changes, but can you clarify why Plotly.restyle(gd, 'groups', [...]) enters the picture? Is the goal to make sure that the transform updates the result when groups are modified?

[etpinard]

Is the goal to make sure that the transform updates the result when groups are modified?

Exactly. From #1028 (comment), if we would have chosen the cast-in-Plots.doCalcdata option, this test case would effectively check that groups was including in the list of attribute to recalc on restyle.

So, might pin down this functionality right now, if we chose to cast groups items at some point.

[etpinard]

Obsolete.