str() is slow for R6 objects containing large vectors

[kevinushey]

library(R6)
Big <- R6Class("Big", public = list(data = rnorm(1E7)))
big <- Big$new()
system.time(str(big))
#> Classes 'Big', 'R6' <Big>
#>   Public:
#>     clone: function (deep = FALSE) 
#>     data: -0.594100136328869 0.104513371816434 0.531537555504934 - ...
#>    user  system elapsed 
#>  19.191   0.417  19.630

See also: rstudio/rstudio#3544

Could R6 subset large vectors before calling str()?

[gaborcsardi]

Seems like str calls format and the format.R6 method is slow.

[wch]

It looks most of the time is spent in this paste():

R6/R/print.R

Line 114 in 1c1f425

else if (is.atomic(obj)) trim(paste(as.character(obj), collapse = " "))

Given that the subsequent trim() call uses the default width of 60 characters, I suppose it would be OK to do something like:

  paste(as.character(head(obj, 60))

Do you guys see anything that I'm overlooking?

[gaborcsardi]

I think head() is a good idea. But it only works for objects with a [ method, so you would need some fallback, maybe just format(x) or the class of the object, like <environment>:

❯ head(environment())
Error in x[seq_len(n)] : object of type 'environment' is not subsettable

❯ head(ps::ps_handle())
Error in x[seq_len(n)] : object of type 'externalptr' is not subsettable

❯ format(environment())
[1] "<environment: R_GlobalEnv>"

❯ format(ps::ps_handle())
[1] "<ps::ps_handle> PID=18088, NAME=R, AT=2018-09-26 20:00:02"

[gaborcsardi]

Btw. there are some possibly related changes in R-3.5.1-patched, it has a workaround for print(runif(1e7)) taking a long time, i.e. it subsets using getOption(max.print) before doing the formatting. Maybe it is worth checking if format was also fixed there, not just print(). OTOH we want to fix older R versions as well...

[wch]

But head() should work for the is.atomic() case, right? For the other cases you listed, it would already print them out in different ways.

[gaborcsardi]

Yes, for is.atomic is should work fine, I think.