Decide how to deal with inheritance of aesthetics

[rpruim]

Currently inheritance causes some layers to crash if the data in those layers isn't help to evaluate all of the aesthetics in the top layer.

This could be fixed by

• moving aesthetics out of the ggplot() call and into the first layer
• adding inherit.aes = FALSE when add = TRUE

For now, users can do inherit.aes = FALSE themselves to avoid issues, but that isn't very hospitable.

[rpruim]

I've implemented a solution along the lines of the first option above.

[rpruim]

One downside of this method is that you don't have the ability to inherit aesthetics. We could add something that would

a. allow the main plot to (optionally?) include aesthetics in the ggplot() call
b. allow added layers to set inherit.aes = FALSE (this should work already).

Just "thinking aloud" here: Perhaps we can auto-detect this by looking at aes_form. If the roles in the aes_form for an add on layer are a subset of the roles in the main plot, we could use inheritance, else not.

The issues would be figuring out what aes_form was in the main plot (use an attribute?) and coming up with an API for specifying modified or additional aesthetics in the add-on layers. For example, suppose wants to use inheritance to avoid repeating the x aesthetic in

gf_point( y1 ~ x, data = some_data) %>%
  gf_point(y2 ~ x)

y2~ isn't an option. ~ y2 would likely work, but seems like a bad idea to promote.

Perhaps we should only allow inheritance when the formula in gf is identical in layer and main plot. That would allow things like

gf_point(y ~ x, data = some_data) %>%
  gf_smoth()

Currently, we need to do gf_smooth(y ~ x)

Or perhaps it is actually best to require the formula to be specified in each layer.