[willgearty]

CoordTransXY$transform() recognizes axis tick labels and axis lines by the shape of their data: all x/y equal to Inf/-Inf, or a single unique x or y value. Layer data with that shape was treated as an axis too, so with coord_trans_xy():

• geom_hline() and geom_vline() weren't drawn,
• a flat geom_line() or a single annotate("text") wasn't drawn,
• a single point was moved to the edge of the panel.

Axis guide data reaching transform() never has a group column (tick labels have x, y, .value, .label; axis lines only x, y), while ggplot2 gives all layer data a group column. So:

• The axis tick label and axis line hacks now only apply to data without a group column. Layer data is always transformed normally. The hacks themselves are unchanged, and their inline comments moved inside each branch to keep the lines short.
• Updates the @details caveat. geom_rect() and similar geoms (geom_tile(), geom_col(), annotate("rect") with infinite limits) already work, because ggplot2 converts them to polygons for non-linear coordinate systems. geom_raster() needs ggplot2 >= 4.0.0, which falls back to drawing rectangles; ggplot2 3.5.x refuses geom_raster() with any non-Cartesian coord. Also removes the stale # TODO: transform corners for geom_rect? comment.

Tests (test-coord_trans_xy.R): a plot with geom_hline(), geom_vline(), a flat geom_line(), a single point and a single text label under shear(2, 0) + rotate(-pi / 3). For each layer, the drawn coordinates are back-transformed and checked against the data values (e.g. y == 1 along the whole hline). It also adds a theme_void() snapshot, for both ggplot2 4.x and pre-4.0, so the snapshot doesn't depend on the pre-4.0 axis fix in #117. Without the code change, all five position checks fail.

Locally, all 23 coord_trans_xy tests pass with ggplot2 4.0.3, including the existing capped-axis, reflect and reverse-scale snapshots, so axis rendering is unchanged. With ggplot2 3.5.2 the new checks pass and the only failures are the outdated pre-4.0 coord_trans_xy snapshots that #117 fixes, the same as on main.

This relies on ggplot2 always giving layer data a group column. That held in both versions I tested (3.5.2 and 4.0.3) and has been ggplot2's behavior for a long time, but it isn't formally documented.

This PR edits the same transform() branches as #117 (which changes the Inf sign checks), and the top of NEWS.md like #116, #117, #120 and #121. Whichever merges later will need a small merge; the two changes combine without conflict in logic.

TODO:

• Add tests
• Update docs
• Add NEWS item

AI disclosure: Claude identified the problem and drafted the fix.

🤖 Generated with Claude Code

https://claude.ai/code/session_01A7zDoJJQSEuNmyB6xR3njB

---

Generated by Claude Code