Using dplyr 1.0.4., plot with controls created by crosstalk and plotly is not interactive any more

[fjodor]

Thanks for looking into this, Romain!

Using remotes::install_github("tidyverse/dplyr"), I get dplyr version 1.0.4.9000. Get the same image that you posted, but the slider and the checkboxes have no effect on the plot, which they do when using dplyr version 1.0.2., or commenting out group_by() .

Originally posted by @fjodor in #5726 (comment)

[fjodor]

Since updating dplyr, a plotly chart linked with a slider and a checkbox via crosstalk does not react to the controls any more. The following code works with dplyr version 1.0.2. In versions 1.0.3 and 1.0.4 I need to comment out the group_by() statement to make it work. Using the current development version of dplyr, 1.0.4.9000, did not help on my system.

[fjodor]

library(plotly)
library(crosstalk)

shared_data <- mtcars %>% 
  SharedData$new()

p <- shared_data %>% 
  plot_ly(x = ~mpg, y = ~hp) %>%
  group_by(am) %>%
  add_markers()

bscols(list(
  filter_slider(id = "mpg", label = "Miles per Gallon",
                sharedData = shared_data, column = "mpg"),
  filter_checkbox(id = "am", label = "Automatic / Manual",
                  sharedData = shared_data, group = ~am)),
  p)

[fjodor]

sessionInfo()
R version 4.0.3 (2020-10-10)
Platform: x86_64-w64-mingw32/x64 (64-bit)
Running under: Windows 10 x64 (build 19042)

Matrix products: default

locale:
[1] LC_COLLATE=German_Germany.1252 LC_CTYPE=German_Germany.1252
[3] LC_MONETARY=German_Germany.1252 LC_NUMERIC=C
[5] LC_TIME=German_Germany.1252

attached base packages:
[1] stats graphics grDevices utils datasets methods base

other attached packages:
[1] **crosstalk_1.1.1 plotly_4.9.3 ggplot2_3.3.3**

loaded via a namespace (and not attached):
[1] Rcpp_1.0.6 pillar_1.4.7 compiler_4.0.3 later_1.1.0.1 tools_4.0.3
[6] digest_0.6.27 jsonlite_1.7.2 lifecycle_0.2.0 tibble_3.0.6 gtable_0.3.0
[11] viridisLite_0.3.0 pkgconfig_2.0.3 rlang_0.4.10 shiny_1.6.0 DBI_1.1.1
[16] yaml_2.2.1 fastmap_1.1.0 withr_2.4.1 **dplyr_1.0.4.9000** httr_1.4.2
[21] generics_0.1.0 vctrs_0.3.6 htmlwidgets_1.5.3 grid_4.0.3 tidyselect_1.1.0
[26] glue_1.4.2 data.table_1.13.6 R6_2.5.0 purrr_0.3.4 tidyr_1.1.2
[31] magrittr_2.0.1 promises_1.2.0.1 scales_1.1.1 ellipsis_0.3.1 htmltools_0.5.1.1
[36] assertthat_0.2.1 xtable_1.8-4 mime_0.9 colorspace_2.0-0 httpuv_1.5.5
[41] lazyeval_0.2.2 munsell_0.5.0 crayon_1.4.1

[romainfrancois]

cc @cpsievert

[cpsievert]

This happens because plotly is essentially assuming that group_by(..., .add = TRUE) will preserve attributes, which seems to no longer be the case when a group(s) get added:

m <- mtcars
attr(m, "foo") <- "a"
m <- dplyr::group_by(m, vs)
attr(m, "foo")
#> [1] "a"
m <- dplyr::group_by(m, cyl, .add = TRUE)
attr(m, "foo")
#> NULL

I can make a fix for this in plotly, but maybe also something you want to address in dplyr?