Only call zero_range() and as.numeric() in transformed space

[DavisVaughan]

In case the limits in data space don't support conversion to numeric

---

I've been working on scale support for {clock} types in r-lib/clock#345. These include scales for things like year_month_day<month>, a "year-month" style date vector, and year_week_day<week>, a "year-week" style date vector.

I've got pretty much everything working, but there is a show stopping issue with ScaleContinuous$get_breaks() that I've hopefully fixed here.

It looks like scales::zero_range() is being called on as.numeric(limits), where limits is in the original data space. This is problematic for clock, because calendar types like year_month_day and year_week_day do not support as.numeric(), so the limits can't be converted directly. However, they do support scale$trans$transform() to convert from a clock type in data space to a numeric value in transformed space.

IIUC, I think that the intention of scales::zero_range() is that it should only be called on objects in transformed space anyways. Looking at other uses of it in ggplot2:

• Most importantly, in ScaleContinuous$get_minor_breaks() it is called on limits when they are in transformed space, so this PR basically just makes get_breaks() consistent with that.

  ggplot2/R/scale-.R

  Line 692 in aa2ddbe

  if (zero_range(as.numeric(limits))) {

• In get_breaks() just above where this change happens it is called on domain which is in transformed space

  ggplot2/R/scale-.R

  Line 651 in aa2ddbe

  if (length(domain) == 2 && !zero_range(domain)) {

• In bin_breaks_bins() it is called on x_range which comes through as a $dimension() call, and the dimension is always returned in transformed space

  ggplot2/R/bin.R

  Line 112 in aa2ddbe

  if (zero_range(x_range)) {

Hopefully this is enough evidence to support this change.

I couldn't think of a good test for this, so if anyone has a better idea, feel free to take over.

[thomasp85]

LGTM - @clauswilke can you see any issues with this change?

[teunbrand]

This looks good to me. Would you mind harmonising the news bullet with the latest changes @DavisVaughan or would you prefer me to do that?

[DavisVaughan]

@teunbrand if you could please take over I would really appreciate it! It would probably be much faster for you at this point.

[teunbrand]

Thanks @DavisVaughan for the PR!