compute() should enforce computation

[mgacc0]

I suspect dplyr::compute() is not enforcing computation.

For example:

df1 <- tibble(
  letter = sample(letters, 1e8, replace = TRUE),
  n = sample(100, 1e8, replace = TRUE)
) %>%
  lazy_dt() %>%
  group_by(letter) %>%
  summarise(n = sum(n))

The call to compute() returns inmediatly (without doing calculations):

{
  tictoc::tic()
  df_m <- df1 %>%
    compute()
  tictoc::toc()
}
# 0 sec elapsed

And to enforce the execution, I'm using (provisionally) this alternative:

df_m_2 <- df1 %>%
  collect() %>%
  lazy_dt()

Both should be true:

expect_s3_class(df_m, "dtplyr_step_first")
# Error: `df_m` inherits from 'dtplyr_step_group'/'dtplyr_step' not 'dtplyr_step_first'.
expect_s3_class(df_m_2, "dtplyr_step_first")

[markfairbanks]

This is mentioned in the documentation here. The point of compute() in dtplyr is to "generate an intermediate assignment in the translation", so this is working as intended.

collect(), as_tibble(), as.data.frame(), and as.data.table() actually enforce the computation.

[mgacc0]

Then, I will continue using

df1 <- df1 %>%
  collect() %>%
  lazy_dt()

to enforce computation.

As you know, this is an optimization to improve performance (when nrows(df1) > 10M): caching df1 if it is going to be reused multiple times.