ViewScale's break_positions() now returns pre-transformed positions

[cpsievert]

Consider the following plot

# Create some noisy exponentially-distributed data
dat <- data.frame(
  xval = c(0.26932812,-0.05341404,0.36977717,0.91504712,0.46329006,0.37956526, 0.93290644,0.75558976,0.67633497,0.48655293,0.79478162,0.55109982, 0.51681398,0.81073512,0.49406579,0.93919618,0.90472008,0.98732256, 0.94379876,0.95790909,0.54614241,1.13356941,1.13299144,1.18159277, 1.16428407,1.22955005,1.21030897,1.23314811,1.53822718,1.53674330, 1.80020468,1.40774011,1.74573515,1.26651625,2.06607711,1.50237263, 1.38480531,1.83625381,2.35275649,1.99004291,2.80396442,2.20863240, 2.42998876,2.12801180,2.26290348,2.38185989,2.14936036,2.66587947, 2.64586596,2.44240603,2.39266452,3.11831215,2.70258927,2.65529134, 2.65634690,2.95984290,2.71058076,2.87919480,3.07739358,2.66841935, 3.10792706,3.17134285,3.98070271,3.55497279,3.36831009,3.31390892, 3.32753965,2.86981968,3.22741000,3.78806438,3.74434536,3.56928928, 3.83783177,3.24485807,4.05766233,4.13619455,4.26888054,3.47546258, 3.93045819,3.77620080,4.66676431,3.88059240,4.54694485,4.03915767, 4.25556093,4.39251819,4.42692029,4.23262929,4.44890758,4.84981161, 4.51104252,4.33004508,5.06350705,4.89714069,4.21599077,4.55457578, 5.04044393,4.89111297,5.03105215,4.64113164), 
  yval = c(1.177512e+01,7.303113e+00,6.109053e+00,2.545169e+01,3.366341e+01,1.042255e+01,2.703767e+01,1.178223e+01,4.495965e+01,1.614609e+01,4.003015e+01,1.038442e+02,4.024992e+01,4.163942e+01,9.108197e+01,3.116299e+01,2.558871e+02,7.482977e+01,2.502789e+01,5.923683e+01,3.967814e+01,9.207318e+01,1.298618e+02,1.138197e+02,1.804303e+02,3.363494e+02,3.197204e+02,4.968737e+02,1.783433e+02,4.765546e+02,4.486885e+02,6.736079e+02,4.289288e+02,3.433946e+02,5.658634e+02,4.667053e+02,5.257803e+02,3.401038e+02,6.131335e+02,5.928647e+02,7.838524e+02,7.987915e+02,3.348470e+03,1.704767e+03,1.264169e+03,2.690011e+03,2.738240e+03,1.663862e+03,5.377442e+03,3.883820e+03,6.673624e+03,1.857346e+03,6.683962e+03,1.213027e+03,1.742885e+03,2.146094e+03,4.597174e+03,4.357154e+03,8.413851e+03,8.194194e+03,7.076611e+03,1.554628e+04,6.984783e+03,1.027392e+04,1.158795e+04,9.193111e+03,3.226748e+04,3.955445e+04,2.978953e+04,1.926420e+04,7.610544e+04,2.129694e+04,1.438764e+04,7.908876e+04,2.676003e+04,1.791758e+05,3.978871e+04,9.411120e+04,4.486940e+04,1.270526e+05,1.587331e+05,1.616173e+05,3.351522e+05,3.001782e+05,2.527824e+05,2.745851e+05,3.446376e+05,1.544497e+05,1.318314e+05,8.334336e+05,2.464391e+05,8.694818e+05,2.747323e+05,6.373497e+05,2.918690e+05,9.505114e+05,7.835278e+05,3.775567e+05,1.795523e+06,1.568159e+06)
)

p <- ggplot(dat, aes(xval, yval)) +
  geom_point() + 
  coord_trans(y = "log2")

If I build it and look at the $break_positions() of the y-axis, they don't seem to consider the coordinate transformation (prior to 3.4.0, they did):

b <- ggplot_build(p)
b$layout$panel_params[[1]]$y$break_positions()
#> [1]        NA 0.2967544 0.5935098 0.8902652

Thankfully, the y.range field has the post-transformed values. However, it'd be nice if break_positions() yielded the same result, especially considering that y.major was silently dropped in recent releases(#3436)

b$layout$panel_params[[1]]$y.major
#> [1] 0.9122862 0.9623324 0.9916076

[cpsievert]

Perhaps part of the problem is that panel_params has lost it's transformation along the way somehow?

p$coordinates$trans$y
#> Transformer: log-2 [1e-100, Inf]

b$layout$panel_params[[1]]$y$scale$trans
#> Transformer: identity [-Inf, Inf]

[yutannihilation]

Since this is about coord_trans() and ViewScale, #3972 might be the cause. If I understand correctly, before v3.4.0, CoordTrans had Scale objects, not ViewScale ones. There should be no change on ViewScale's break_positions().

I'm not immediately sure if my guess is right or wrong, and what's the supposed behavior here. Let me think...

[teunbrand]

The preferred interface for accessing key values in guides is through get_guide_data(), so that (hopefully) accessing the internals to extract tick mark locations won't be needed anymore.