dplyr release 1.0.1 breaks simple ggvis demo: layer_lines and layer_points produce no output.

[timcoote]

I realise that ggvis is in limbo, but I, and presumably others, have some shiny code in end devices that's not easy to update and it's taken a while to confirm that there is an issue and near where to place it.

So I thought that I'd at least document the issue, and potential fix, here. I first found a mention here: https://bit.ly/3k5KO2w.

From an engineering point of view, there may be some sort of regression as the failure is very quiet.

---

ggvis layers stop working. I'm afraid that I couldn't get reprex to work with rstudio-server, so I've put some code below.

install.packages ("ggvis")
library (magrittr)
library (ggvis)

mtcars %>% ggvis (x=~wt, y=~mpg) %>% layer_points()

I'd expect this to produce the "usual" plot. However, there are no points.

However, this does work:

require (devtools)
install_version("dplyr", version = "0.8.5")
# <cr> over which versions to update
# I'm using RStudio Server. So at this point I restart R
library (magrittr)
library (ggvis)

mtcars %>% ggvis (x=~wt, y=~mpg) %>% layer_points()

And the plot with points appears.

I did try to git bisect between 0.8.5 and 1.0.0, but the commit for 0.8.5 didn't seem to be in the clone from github, so I couldn't.

[timcoote]

I did manage to git bisect this from an earlier commit that's in the repo. git bisect log:

# bad: [90b3e4ea164514b57c72954c89dc63ab958eceb0] Bullet errors for summarise() (#4729)
git bisect bad 90b3e4ea164514b57c72954c89dc63ab958eceb0
# bad: [90b3e4ea164514b57c72954c89dc63ab958eceb0] Bullet errors for summarise() (#4729)
git bisect bad 90b3e4ea164514b57c72954c89dc63ab958eceb0
# good: [689518b4eb6adcfdeb5917dd97d41419b3e3d69b] Improve arrange docs
git bisect good 689518b4eb6adcfdeb5917dd97d41419b3e3d69b
# good: [3d45972610542138a790d8c1cd00a5990c0a1304] Upgrade tidyselect (#4720)
git bisect good 3d45972610542138a790d8c1cd00a5990c0a1304
# good: [cab26df184fa7032a16faa602fc529542b508be1] Remove unnused match_vars() function
git bisect good cab26df184fa7032a16faa602fc529542b508be1
# good: [dc8ebd4cb07c61a5e8fb9648a87fbf2f4628897a] Fix broken test (#4743)
git bisect good dc8ebd4cb07c61a5e8fb9648a87fbf2f4628897a
# good: [f7f66821327ce41f9a9a49bfc0f61434039f864f] Tweak error message
git bisect good f7f66821327ce41f9a9a49bfc0f61434039f864f
# bad: [887d40bf95cfe8c526245bd61fcde2c18af65e9d] Set tibble version (#4745)
git bisect bad 887d40bf95cfe8c526245bd61fcde2c18af65e9d
# bad: [125d75d6f76c78c27da45adc258f0b8b6a4620df] Simplify group metadata (#4728)
git bisect bad 125d75d6f76c78c27da45adc258f0b8b6a4620df
# first bad commit: [125d75d6f76c78c27da45adc258f0b8b6a4620df] Simplify group metadata (#4728)

There's quite a bit in that 'first bad' commit.

[lionel-]

Fixed in rstudio/ggvis#485

Install the branch with:

remotes::install_github("lionel-/ggvis@fix-dplyr-1-0-0")

But I think we should also fix this in dplyr. groups() should return NULL when groups are empty, as before.

[timcoote]

Thanks. I finally confirmed that this does work.

It's not really my business, but, personally, I'd make the new test more explicitly based on a boolean value, rather than a function returning a value of 0. If I could test it, I'd propose a change, but I can't - and you should ignore my comment if I'm writing in the face of the project style guide.

I also think that it is a better api to return an empty list if there are no groups, rather than null. R is weakly typed language, so it's not possible to have things like MAYBE types of Haskell to catch errors of not checking for changes in returned type.

[lionel-]

so it's not possible to have things like MAYBE types of Haskell

It is of course possible. There is a whole dynamically typed language funded on this kind of option type: lisp and nil punning.

I agree it is better to return list() in this case, but thought it wasn't worth the breaking change. What I missed is that dplyr would previously return NULL or list() interchangeably, making the return type complex. So the new behaviour was worth the change.