Provide high performance limited functionality equivalent of tibble

[hadley]

i.e. with no evaluation or auto-naming - maybe something like tibbly <- function(...) as_tibble(list(...)) or even new_tibble(list(...)).

Or maybe we should encourage developers to use this idiom some other way?

[krlmlr]

What is the use case, and the performance hit we're seeing with the current implementation? We can move parts to C for better performance. On that note: should we move lst() to rlang?

[hadley]

Use case is that you have a known "safe" data structure, and don't want to waste anytime checking (e.g. #353)

[krlmlr]

Yeah, there is a difference of an order of magnitude, but we're talking about microseconds vs. a millisecond. I wonder if this is really worth the risk and the effort:

library(tibble)
library(rlang)
df <- unclass(nycflights13::flights)

microbenchmark::microbenchmark(
  tibble(!!! df),
  invoke(tibble, df),
  tibble:::new_tibble(df),
  tibble:::new_tibble(df, nrow = 336776L)
)
#> Unit: microseconds
#>                                     expr      min        lq       mean
#>                        tibble(!(!(!df))) 1304.236 1348.9510 1809.99604
#>                       invoke(tibble, df) 1834.505 1950.2320 2208.62348
#>                  tibble:::new_tibble(df)   51.673   58.9760   73.82381
#>  tibble:::new_tibble(df, nrow = 336776L)   51.407   56.8365   64.79954
#>     median       uq       max neval cld
#>  1404.4650 1546.322 25915.788   100   b
#>  2042.5190 2142.852  5741.313   100   b
#>    63.5315   68.260   876.542   100  a 
#>    61.9530   65.871   127.549   100  a

Created on 2018-01-15 by the reprex package (v0.1.1.9000).

[krlmlr]

as_tibble(validate = FALSE) seems to be just fast enough (slowdown x2 with a wide data frame, mostly due to the length check which I'd rather keep). I'll add a reminder to update documentation to mention this:

library(tibble)
library(rlang)
library(dplyr)
#> 
#> Attaching package: 'dplyr'
#> The following objects are masked from 'package:stats':
#> 
#>     filter, lag
#> The following objects are masked from 'package:base':
#> 
#>     intersect, setdiff, setequal, union
library(tidyr)
df <-
  nycflights13::flights %>% 
  count(origin, dest, month) %>%
  unite("relation", origin, dest, sep = "->") %>%
  spread(relation, n) %>% 
  unclass()

length(df)
#> [1] 225

microbenchmark::microbenchmark(
  tibble(!!! df),
  invoke(tibble, df),
  as_tibble(df),
  as_tibble(df, validate = FALSE),
  tibble:::new_tibble(df),
  tibble:::new_tibble(df, nrow = 12L)
)
#> Unit: microseconds
#>                                 expr       min         lq       mean
#>                    tibble(!(!(!df))) 30055.013 32404.4080 34550.9134
#>                   invoke(tibble, df) 34103.947 37124.7775 39586.9928
#>                        as_tibble(df)  1981.951  2141.3525  2528.6651
#>      as_tibble(df, validate = FALSE)   305.573   342.8565   409.1099
#>              tibble:::new_tibble(df)   172.964   195.4785   216.4160
#>  tibble:::new_tibble(df, nrow = 12L)   171.111   189.9820   220.2521
#>      median         uq       max neval  cld
#>  34036.5380 35426.0850 84756.116   100   c 
#>  39005.9890 40569.4315 69548.031   100    d
#>   2240.3870  2455.4070  5812.929   100  b  
#>    369.4780   397.9385  2820.020   100 a   
#>    207.7320   224.9585   440.641   100 a   
#>    204.1255   221.5070   429.062   100 a

Created on 2018-01-15 by the reprex package (v0.1.1.9000).

[krlmlr]

Also, new_tibble() is already exported.

[github-actions]

This old thread has been automatically locked. If you think you have found something related to this, please open a new issue and link to this old issue if necessary.