ggplotly date shows wrong when use geom_bar(stat='count')

I find a problem that, when I use geom_bar(stat='count') to draw a plot and convert to plotly, when pointing to, the label shows date an integer, but when I aggregate the data first and use geom_bar(stat='identity'), the label shows date in proper format. The issue happened after I update to MRO 3.4.1, and try to solve the problem that ggplotly makes graph empty when x axis is date. I update both ggplot2 and plotly to github version that empty problem is gone, but new problem comes.

By the way. Is there a method that control the date format in the label?

devtools::install_github("hadley/ggplot2")
devtools::install_github("ropensci/plotly") 
library(ggplot2) 
library(plotly) 

df2 = data.frame(time = c('2017-01-01 06:00:23', '2017-01-01 10:00:23', 
                          '2017-01-01 18:00:23', '2017-01-01 24:00:23', 
                          '2017-01-02 06:00:23', '2017-01-02 10:00:23', 
                          '2017-01-02 18:00:23', '2017-01-03 24:00:23'), 
                 y = c(1:8)) 

df2.p = ggplot(df2 %>% mutate(date_id = as.Date(time)), aes(x = date_id, y = ..count..)) + 
  geom_bar(stat = 'count') 

ggplotly(df2.p) 

df3 = df2 %>% mutate(date_id = as.Date(time)) %>% 
  group_by(date_id) %>% summarise(
    count = n() 
  ) 
df3.p = ggplot(df3, aes(x = date_id, y = count)) + 
  geom_bar(stat = 'identity') 
ggplotly(df3.p) 

Start with the reproducible R examples in the issue and compare the ggplotly output for geom_bar(stat='count') with the stat='identity' version. Investigate how the date_id aesthetic is represented in each conversion; done means count-based bars display properly formatted date labels, with any date-format behavior clarified.