Adding layers makes tooltips disappear

Hi there,

I'm experiencing the same issue as @borenstein, referenced here: Adding layer makes tool tips disappear. Someone asked @borenstein to post the issue here but I didn't see a record of it. Here's a shorter reproducible example and summary; please note their solution (comparing data on hover) doesn't work in my use case (not shown here).

I want to create a plot where users can hover over a map and a tooltip for a single layer appears, for example, showing the names of US counties, with state boundaries overlaid on the map. If your mouse is "far" from a state boundary, the tooltip appears. However, if you are "close" (e.g., southeastern edge of map), the tooltip does not appear.

I've tried identifying the traces that contain parts of my plot that I don't want to show tooltips for - as in these stack overflow posts:

• ggplotly - only return tooltip hover text on certain geom objects

• Disable hover information for a specific layer (geom) of plotly

But, it doesn't work.

library(plotly)
library(tidyverse)

### New England: *sometimes* shows tool tips
new_england <- c("maine", "new hampshire", "vermont", "massachusetts", "connecticut", "rhode island")

b <- ggplot2::map_data("state") %>%
  filter(region %in% new_england)

a <- data.frame(group = seq(3085), val = rnorm(3085)) %>%
  left_join(ggplot2::map_data("county"), by = "group") %>%
  filter(region %in% new_england)

p <- ggplot(a, aes(long, lat, group = group)) +
  geom_polygon(aes(fill = val, text = subregion)) +
  geom_path(data = b, color = "black")

ggplotly(p, tooltip="text")

Tagging @MountainCedar so they know I've filed an issue about this.

Start by running the provided ggplotly reproducible example and compare hover behavior near and far from the state boundary. Trace how layered polygon and path traces handle tooltip selection; done means county-name tooltips remain available near boundaries without requiring the proposed workaround. No file or test is named in the issue, so locating the relevant hover code and adding regression coverage will require project familiarity.