multiple inside guide box with different position #6210

Yunuuuu · 2024-11-28T16:25:36Z

Try to fix #5712.

This enables users to set legend.position.inside and legend.justification.inside in theme of guide_legend and guide_colourbar.

Here is an example:

ggplot(mpg, aes(displ, hwy, shape = drv, colour = cty, size = year)) +
    geom_point(aes(alpha = cyl)) +
    guides(
        colour = guide_colourbar(
            position = "inside",
            theme = theme(legend.position.inside = c(1, 0))
        ),
        size = guide_legend(position = "top"),
        alpha = guide_legend(
            position = "inside",
            theme = theme(legend.position.inside = c(0, 1))
        ),
        shape = guide_legend(position = "left")
    )

teunbrand

Hi Yunuuuu, thanks for preparing a PR for this issue!

The PR appears to break a visual test for the alignment of legends with the inside position. It is important to us to preserve current behaviour without regressions.

Next, it is hard to tell from the PR what the assembly rules for inside legends are. When legends share legend.position.inside values, they merge (are placed in the same guide-box). It seems that justification also factors into this and I'm not convinced that it should. It'd be great if the logic for this is expressed more clearly, perhaps through code comments.

Lastly, there are a few fussy changes to the code I'd like to propose, but it'd make more sense to go over this after the PR adresses the previous points.

Yunuuuu · 2024-11-29T11:10:29Z

@teunbrand Thank you for the reviewing. I have made the following changes based on your feedback:

Added more comments and explanations throughout the code to improve clarity.
Fixed the test error that was identified.
Merged the guide legends based solely on legend.position.inside, as it seems sufficient to identify the legend groups.

Yunuuuu · 2024-11-29T11:18:01Z

I'll check it

Yunuuuu · 2024-11-29T11:42:57Z

Should we always keep the inside guide box, even if there are no inside guide legends?

teunbrand · 2024-11-29T11:50:05Z

Should we always keep the inside guide box, even if there are no inside guide legends?

An empty inside guide box should give a zeroGrob() for consistency in gtable layout

Yunuuuu · 2024-11-29T12:07:01Z

should fixed

teunbrand

Thanks for getting the PR to work!
My overall feeling about this PR is that it is a little bit on the more invasive side, whereas I think it could be more surgical than it currently is.
If we lift the responsibility of managing the positions into Guides$assemble(), we should have to change less code elsewhere.

R/guide-legend.R

R/guides-.R

teunbrand · 2024-11-29T12:51:52Z

R/guides-.R

+    if (startsWith(position, "inside")) {
+      # Global justification of the complete legend box
+      global_just <- valid.just(calc_element(
+        "legend.justification.inside", theme
+      ))
+      # for inside guide legends, the position was attached in
+      # each grob of the input grobs (which should share the same position)
+      inside_position <- attr(.subset2(grobs, 1L), "inside_position") %||%
+        # fallback to original method of ggplot2 <=3.5.1
+        .subset2(theme, "legend.position.inside") %||% global_just
      global_xjust  <- global_just[1]
      global_yjust  <- global_just[2]
+      x <- inside_position[1]
+      y <- inside_position[2]
+      global_margin <- margin()
+    } else {
+      # Global justification of the complete legend box
+      global_just <- paste0("legend.justification.", position)
+      global_just <- valid.just(calc_element(global_just, theme))
+      x <- global_xjust  <- global_just[1]
+      y <- global_yjust  <- global_just[2]


I think we might not even have to alter the logic of Guides$package_box() if we just manage everything well in Guides$assemble().

Solved, but I haven't found a way to avoid changing Guides$package_box().

I'm happy to explore this if you feel this gets stuck

R/guides-.R

Yunuuuu · 2024-11-29T13:16:57Z

I'll try to make some new changes

Yunuuuu · 2024-11-29T16:41:29Z

Thank you for your thorough review and all the valuable advice.

Yunuuuu · 2024-11-29T17:41:27Z

After writing some examples, I’ve realized that it’s necessary to allow users to set the inside justification for each inside legend. In this way, it would be beneficial for splitting legends based on their justifications, as users may be confused about which justification will be applied if we only split legends with the same inside position into groups.

Yunuuuu · 2024-11-29T18:14:54Z

I’m unable to run the tests on my local machine, as there are always some errors. I’m unable to update the test snapshots.

teunbrand

Thanks for all the changes, Yunuuuu!
I apologise but I still have a few more to suggest. I'd be happy to help out trying to find a solution for not editing Guides$package_box() as much as is reasonable.

I’m unable to update the test snapshots.

I'm happy to do this for you if you like.

teunbrand · 2024-12-02T08:05:09Z

R/guides-.R

+        if (is.null(inside_just)) {
+          inside_justs[[i]] <- default_inside_just
+        } else {
+          inside_justs[[i]] <- valid.just(inside_just)
+        }


Suggested change

if (is.null(inside_just)) {

inside_justs[[i]] <- default_inside_just

} else {

inside_justs[[i]] <- valid.just(inside_just)

}

inside_justs[[i]] <- valid.just(inside_just %||% default_inside_just)

teunbrand · 2024-12-02T08:08:44Z

R/guides-.R

+    for (i in seq_along(positions)) {
+      if (identical(positions[i], "inside")) {


Suggested change

for (i in seq_along(positions)) {

if (identical(positions[i], "inside")) {

for (i in seq_along(positions)[positions == "inside"]) {

teunbrand · 2024-12-02T08:17:00Z

R/guides-.R

+    positions <- positions[keep]
+    inside_positions <- inside_positions[keep]
+    inside_justs <- inside_justs[keep]
+    groups <- groups[keep]
+
+    # we group the guide legends
+    locs <- vec_group_loc(groups)
+    indices <- locs$loc
+    grobs <- vec_chop(grobs, indices = indices)
+    names(grobs) <- locs$key
+
+    # for each group, they share the same locations,
+    # so we only extract the first one of `positions` and `inside_positions`
+    first_indice <- lapply(indices, `[[`, 1L)
+    positions <- vec_chop(positions, indices = first_indice)
+    inside_positions <- vec_chop(inside_positions, indices = first_indice)
+    inside_justs <- vec_chop(inside_justs, indices = first_indice)


I think this could be simplified by tracking positions/inside_positions/inside_justs in a data.frame.
vec_group_loc() treats rows that are the same as the same group, and the locs$key should allow you to skip having to subset the first index.

teunbrand · 2024-12-02T08:23:13Z

R/plot-build.R

+  inside_legends <- legends[startsWith(names(legends), "inside")]
+  if (length(inside_legends)) {
+    for (i in seq_along(inside_legends)) {
+      table <- gtable_add_grob(
+        table, inside_legends[[i]], clip = "off",
+        t = place$t, b = place$b, l = place$l, r = place$r,
+        name = paste("guide-box-inside", i, sep = "-")
+      )
+    }
+  } else { # to be consistent with original gtable layout
+    table <- gtable_add_grob(
+      table, zeroGrob(), clip = "off",
+      t = place$t, b = place$b, l = place$l, r = place$r,
+      name = "guide-box-inside"
+    )
+  }


I think it should be possible to avoid this by combining all packaged inside guides into a single grob.

teunbrand · 2024-12-02T08:31:09Z

R/guides-.R

+    positions <- positions %||% vapply(
      params,
-      function(p) p$position[1] %||% default_position,
-      character(1)
+      function(p) p$position[1] %||% "right",
+      character(1), USE.NAMES = FALSE
    )


Can't we assume here that the Guides$draw() method receives a well-formed positions argument, so we don't have to fall back here? You could even change positions = NULL to positions in the arguments to signal that it is a required argument.

teunbrand · 2024-12-02T08:35:18Z

R/guides-.R

+  package_box = function(grobs, position, theme,
+                         inside_position = NULL, inside_just = NULL) {


I still think we could prevent having to edit this method.
We can just make a theme for every iteration that has the correct inside_position and inside_just baked in.

teunbrand · 2024-12-02T08:39:12Z

R/guides-.R

+      directions <- ifelse(
+        positions %in% c("top", "bottom"), 
+        "horizontal", "vertical"
+      )
+    } else {
+      directions <- rep(direction, length(positions))


I'm not understanding this change. The else branch repeats NULL some times, which remains NULL because it has 0-length?

multiple inside guide box with different position

0a69e9f

teunbrand requested changes Nov 29, 2024

View reviewed changes

merge guide legends depend on legend.position.inside only

790fb2e

fix inside legend justification

7e156d0

no inside guide box if no inside guide legends

1d7c5b9

Yunuuuu added 2 commits November 29, 2024 19:54

add guide-box-index when there is no inside legends

c6e90ea

fix inside guide box area

51889d4

teunbrand requested changes Nov 29, 2024

View reviewed changes

Yunuuuu added 7 commits November 29, 2024 23:56

manage position in Guides$assemble()

b7754c8

revert

d7aa7c4

revert

a4f1221

revert

d1f6833

fix inside legend coordinates

c4882c6

fix test error

9854c58

no need to prepare inside legends when empty

1de623f

Yunuuuu added 2 commits November 30, 2024 01:34

allow set the inside justification for each legend

2c1a0d1

test multiple inside legends with different positions

04827f5

Yunuuuu added 2 commits November 30, 2024 01:57

fix R CMD check error

92f57f8

code notes

e381fa3

teunbrand reviewed Dec 2, 2024

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

multiple inside guide box with different position #6210

multiple inside guide box with different position #6210

Yunuuuu commented Nov 28, 2024 •

edited

Loading

teunbrand left a comment •

edited

Loading

Yunuuuu commented Nov 29, 2024

Yunuuuu commented Nov 29, 2024

Yunuuuu commented Nov 29, 2024

teunbrand commented Nov 29, 2024

Yunuuuu commented Nov 29, 2024

teunbrand left a comment

teunbrand Nov 29, 2024

Yunuuuu Nov 29, 2024 •

edited

Loading

teunbrand Dec 2, 2024

Yunuuuu commented Nov 29, 2024

Yunuuuu commented Nov 29, 2024

Yunuuuu commented Nov 29, 2024 •

edited

Loading

Yunuuuu commented Nov 29, 2024

teunbrand left a comment •

edited

Loading

teunbrand Dec 2, 2024

teunbrand Dec 2, 2024

teunbrand Dec 2, 2024 •

edited

Loading

teunbrand Dec 2, 2024

teunbrand Dec 2, 2024

teunbrand Dec 2, 2024

teunbrand Dec 2, 2024

		for (i in seq_along(positions)) {
		if (identical(positions[i], "inside")) {

	for (i in seq_along(positions)) {
	if (identical(positions[i], "inside")) {
	for (i in seq_along(positions)[positions == "inside"]) {

		package_box = function(grobs, position, theme,
		inside_position = NULL, inside_just = NULL) {

multiple inside guide box with different position #6210

Are you sure you want to change the base?

multiple inside guide box with different position #6210

Conversation

Yunuuuu commented Nov 28, 2024 • edited Loading

teunbrand left a comment • edited Loading

Choose a reason for hiding this comment

Yunuuuu commented Nov 29, 2024

Yunuuuu commented Nov 29, 2024

Yunuuuu commented Nov 29, 2024

teunbrand commented Nov 29, 2024

Yunuuuu commented Nov 29, 2024

teunbrand left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Yunuuuu Nov 29, 2024 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Yunuuuu commented Nov 29, 2024

Yunuuuu commented Nov 29, 2024

Yunuuuu commented Nov 29, 2024 • edited Loading

Yunuuuu commented Nov 29, 2024

teunbrand left a comment • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

teunbrand Dec 2, 2024 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Yunuuuu commented Nov 28, 2024 •

edited

Loading

teunbrand left a comment •

edited

Loading

Yunuuuu Nov 29, 2024 •

edited

Loading

Yunuuuu commented Nov 29, 2024 •

edited

Loading

teunbrand left a comment •

edited

Loading

teunbrand Dec 2, 2024 •

edited

Loading