Changing facets loses mirroring

[infotroph]

As currently implemented, mirror_ticks(p) works by subclassing p$facet, but meanwhile the current implementation of ggplot's + operator handles p + facet_* by overwriting p$facet entirely. This means refaceting a previously-mirrored plot will produce an unmirrored plot.

Current workarounds: Adopt a style where you call mirror_ticks last thing before plotting, or else wrap additional mirror_ticks calls around any downstream facet changes:

p = mirror_ticks(ggplot(...) + ... )
p = p + ...
...
 p = mirror_ticks(p+facet_wrap(...))

Possible more complete fixes:

• Mask + with a version that adds mirroring back on after rewriting a facet?

• Subclass the facet_* functions? Major API change, but might allow a nicely gg-ish interface:

  p + facet_wrap_ticks(~fac, sides="trbl")
  p + facet_grid_ticks(f1~f2, scales="free_y", sides="lr", allPanels=TRUE)
  p + facet_null_ticks() #...might want a better name for the one-panel case, tho

  Would be easier if/when ggplot facets are rewritten with extensibility in mind; unofficial word from Hadley (via Twitter 2016-02-02) is that changing facets to ggproto is on his "eventually" list.

[infotroph]

The subclassing option might be easier than I thought. The following is sufficient to obtain a mirrored facet_wrap from a previously unmirrored plot:

facet_tickwrap = function(...){
    f=facet_wrap(...)
    class(f)=c("ggTicks", class(f))
    f
}
p + facet_tickwrap(~fac)

Need to account for all the corner cases of allPanels and of null vs wrap vs grid facets, but those all seem doable.