update graph reference

[theengineear]

This should be fixed in the short-term with an update to our current graph-reference system. However, the long term fix is in the plot.ly frontend code.

Nah, we're goin' for it now :)

The goal here is to only validate keys for now.

Todo:

• Remove streaming validation. #281 remove streaming validation, it doesn't exist in plot schema
• add local cacheing of plot schema

[chriddyp]

@theengineear - are you planning on taking this one?

[theengineear]

Yep. This has been waiting for some changes to plot-schema, but I'm actually planning on it for this week. I'll assign myself to make that clearer :)

[theengineear]

@chriddyp @cldougl , thoughts?

I should probably share this to make sure I'm on the right track. I'm giving graph_objs.py a huge overhaul to update the new graph reference stuff. Everything is path based. That means that objects know what they are by knowing roughly where they are in a figure.

One fun add on to this is auto-completion! By overloading attributes/dict keys we can do pretty cool stuff like this now:

scatter = go.Scatter()
scatter.marker.line.color = 'red'
scatter.marker.symbol = 'square'
scatter.opacity = .8
scatter.y = [2, 1, 2]
scatter  # {'marker': {'line': {'color': 'red'}, 'symbol': 'square'}, 'opacity': 0.8, 'type': 'scatter', 'y': [2, 1, 2]}

Of course the old way of instantiating these works. In addition, you get auto-completion for the old version too:

scatter = go.Scatter()
scatter['marker']['line']['color'] = 'red'
scatter['marker']['symbol'] = ' square'
scatter['opacity'] = .8
scatter['y'] = [2, 1, 2]
scatter  # {'marker': {'line': {'color': 'red'}, 'symbol': 'square'}, 'opacity': 0.8, 'type': 'scatter', 'y': [2, 1, 2]}

I think this really tightens up the syntax and I've found it a heck of a lot nicer so far!

Final note, if you turn on greedy completion in IPython, %config IPCompleter.greedy = True, you get full tab completion of all the possible keys. Making chaining attributes through objects super easy to do if you don't know the attributes by heart.

The major downside to using attributes like this is the potential for an attribute and a key to conflict :(. If this happens the . will have to return the bound method, though the [] can still return the overloaded key... not ideal, but non-fatal at least. To my knowledge, we don't have any such collisions and we could even be proactive about it if we wanted an put in a Circle test for it... just a thought. I think the . syntax with auto-completion really improves code, but I'm open to opposition :)

[chriddyp]

scatter = go.Scatter()
scatter.marker.line.color = 'red'

that's wild that that works. wild!

the [], . parallelism is cool, it's just like how pandas references columns. and i love the tab-tab functionality, that's definitely the best way to learn the language.

[cldougl]

I think the . syntax looks really clean and the tab completion will be so helpful!!

[theengineear]

Aight. Object classes are within 3 \o/. I think angularaxis and radialaxis need to make their way into plot schema still.

The remaining one is Font, which requires working in the defs from the plot schema. so close. Font is no longer considered an object in plot schema so we won't be able to validate / tab complete it for now. I'm not super psyched about this, but at this point I'm really trying to get something up for review...