Update dplyr functions to use non-standard evaluation

[julistanley]

Replaced all instances of depreciated standard evaluation dplyr verbs with their NSE equivalents.

Fixes associated depreciation warnings, so closes #1783. For example,

Before:

> p <- ggplot(txhousing, aes(x = date, y = median, group = city)) +
+   geom_line(alpha = 0.3)
> ggplotly(p) %>% filter(city == "Houston")
Warning message:
`filter_()` is deprecated as of dplyr 0.7.0.
Please use `filter()` instead.
See vignette('programming') for more help

After:

> p <- ggplot(txhousing, aes(x = date, y = median, group = city)) +
+   geom_line(alpha = 0.3)
> ggplotly(p) %>% filter(city == "Houston")

This PR introduces a util function called as_quosures_explicit. If/when issue 1009 in r-lang is resolved, just replace all instances of as_quosures_explicit with rlang::as_quosures.

[julistanley]

Build 4813.1 and 4813.2 passed (devel and oldrel). 4813.3 failed:

══ Failed ══════════════════════════════════════════════════════════════════���═══
── 1. Error: Complex example works (@test-plotly.R#179)  ───────────────────────
Column `.add` can't be modified because it's a grouping variable
Backtrace:
  1. plotly::plot_ly(., x = ~date, y = ~median)
  1. dplyr::group_by(., city)
  1. plotly::add_lines(., alpha = 0.2, name = "Texan Cities", hoverinfo = "none")
  8. dplyr::group_by(., date)
 11. plotly::group_by_.plotly(...)
 13. dplyr:::group_by.data.frame(d, !!!additional_args, .add = add)
 14. dplyr::group_by_prepare(.data, ..., add = add)
 15. dplyr:::add_computed_columns(.data, new_groups)
 17. dplyr:::mutate.tbl_df(.data, !!!mutate_vars)
 18. dplyr:::mutate_impl(.data, dots, caller_env())

Related to this StackOverflow post, maybe? Edit: or maybe just because I missed an add-->.add :o 580cc76.

[cpsievert]

It'd also be great to have a few unit tests to make sure these cases are covered, something like:

expect_warning(
  plot_ly(txhousing) %>% 
    group_by_("city") %>% 
    add_lines(x = ~date, y = ~median),
  NA
)

[julistanley]

Weirdly, I am still getting a warning. Thinking it might have been something weird with my system, I switched to a blank RStudio Cloud instance. (also using lifecycle_verbosity = "warning" but wasn't before--which is why I think I didn't see a warning in the first PR comment)

I clone my fork, then switch to the patch-1 branch, and install all dependencies. Then:

> devtools::load_all(".")
Loading plotly
Visual testing is not enabled.
> options(lifecycle_verbosity = "warning") 
> txhousing %>% plot_ly(x = ~date, y = ~median) %>% group_by(city)
No trace type specified:
  Based on info supplied, a 'scatter' trace seems appropriate.
  Read more about this trace type -> https://plot.ly/r/reference/#scatter
No scatter mode specifed:
  Setting the mode to markers
  Read more about this attribute -> https://plot.ly/r/reference/#scatter-mode
Warning messages:
1: `group_by_()` is deprecated as of dplyr 0.7.0.
Please use `group_by()` instead.
See vignette('programming') for more help 
2: Ignoring 616 observations 

In the terminal, I don't see any instances of group_by_( in the R directory:

rstudio-user@application-2493362-deployment-6489105-2hz46:/cloud/project/plotly/R$ grep -r "group_by_("
plotly_data.R:    #dat <- dplyr::group_by_(dat, crosstalk_key(), add = TRUE)
rstudio-user@application-2493362-deployment-6489105-2hz46:/cloud/project/plotly/R$

There are a few in tests, but I don't think those should interfere:

rstudio-user@application-2493362-deployment-6489105-2hz46:/cloud/project/plotly/tests$ grep -r "group_by_("
testthat/test-plotly-group.R:  dplyr::group_by_("rowname") %>%
testthat/test-ggplot-errorbar.R:    dplyr::group_by_(mtcars, "cyl"),
testthat/test-plotly-data.R:  d <- plotly_data(group_by_(plot_ly(mtcars), c("vs", "am")))

@cpsievert I'm glad that you suggested writing some tests since they caught this. Do you know what might be happening here?

Edit: Made the instance public, if that helps: https://rstudio.cloud/project/1409602

[julistanley]

Oh cool, here's a much more minimal and illustrative example:

> txhousing %>% group_by(city) %>% print(n=3)
# A tibble: 8,602 x 9
# Groups:   city [46]
  city     year month sales  volume median listings inventory  date
  <chr>   <int> <int> <dbl>   <dbl>  <dbl>    <dbl>     <dbl> <dbl>
1 Abilene  2000     1    72 5380000  71400      701       6.3 2000 
2 Abilene  2000     2    98 6505000  58700      746       6.6 2000.
3 Abilene  2000     3   130 9285000  58100      784       6.8 2000.
# … with 8,599 more rows

> txhousing %>% plot_ly() %>% group_by(city) %>% print(n=3)
Error: (converted from warning) `group_by_()` is deprecated as of dplyr 0.7.0.
Please use `group_by()` instead.
See vignette('programming') for more help

> txhousing %>% plot_ly() %>% plotly_data() %>% group_by(city) %>% print(n=3)
# A tibble: 8,602 x 9
# Groups:   city [46]
  city     year month sales  volume median listings inventory  date
  <chr>   <int> <int> <dbl>   <dbl>  <dbl>    <dbl>     <dbl> <dbl>
1 Abilene  2000     1    72 5380000  71400      701       6.3 2000 
2 Abilene  2000     2    98 6505000  58700      746       6.6 2000.
3 Abilene  2000     3   130 9285000  58100      784       6.8 2000.
# … with 8,599 more rows

What happens to a plot_ly object when it's passed to dplyr::group_by? How is it different than passing to plotly_data() and then to dplyr::group_by()?

Related to dplyr #5083, I think? tidyverse/dplyr#5083

[cpsievert]

The difference in txhousing %>% plot_ly() %>% group_by(city) and txhousing %>% plot_ly() %>% plotly_data() %>% group_by(city) is that the former winds up calling group_by_.plotly() (since the input's class is plotly).

[cpsievert]

Ah, I see why txhousing %>% plot_ly() %>% group_by(city) still throws a warning. It's because group_by(plot_ly()) currently winds up calling group_by_.plotly(), and by that time, the warning has been thrown in the generic group_by_()

> library(dplyr)
> methods(group_by)
[1] group_by.data.frame* group_by.default*   
see '?methods' for accessing help and source code

> getS3method("group_by", "default")
function (.data, ..., add = FALSE, .drop = group_by_drop_default(.data)) 
{
    group_by_(.data, .dots = compat_as_lazy_dots(...), add = add)
}

> group_by_
function (.data, ..., .dots = list(), add = FALSE) 
{
    lazy_deprec("group_by")
    UseMethod("group_by_")
}

This means we'll have to define methods for the non-underscore generic functions found in R/plotly_data.R (i.e., we should be defining/registering group_by.plotly()). Do you feel up to that task @julianstanley?

[julistanley]

Thanks for following-up on this @cpsievert!

I think that makes sense. If I understand correctly, group_by (generally, [dplyr-verb]) methods are defined on data.frame ([dplyr-verb].data.frame ), but default otherwise ([dplyr-verb].default). So Plotly objects trigger [dplyr-verb].default, leading to a warning, but they'll trigger [dplyr-verb].plotly if we define that. So we just need to define each NSE dplyr method for plotly objects ([dplyr-verb].plotly).

Sure thing, this sounds simple enough (famous last words?). I'll give this a shot!

[cpsievert]

Thanks for jump starting this effort @julianstanley, I'll take it over the finish line in #1821

[julistanley]

Sounds good, thanks, @cpsievert!