Exploring the object size of `lm()` objects with enclosing environments

lm-hell.r

library(purrr)
library(lobstr)
library(glue)
library(rlang, warn.conflicts = FALSE)

make_an_lm <- function() {
  x <- rep(1L, times = 10000000)
  cat(glue("x is {object.size(x)}B"))
  lm(1 ~ 1)
}

lm_model <- make_an_lm()
#> x is 40000048B

# that's not so bad...
object.size(lm_model)
#> 11336 bytes

# wtf?
obj_size(lm_model)
#> 40,008,504 B

# lm_model is a list...so
str(map(lm_model, obj_size), digits.d = 10)
#> List of 11
#>  $ coefficients : 'lobstr_bytes' num 368
#>  $ residuals    : 'lobstr_bytes' num 280
#>  $ effects      : 'lobstr_bytes' num 368
#>  $ rank         : 'lobstr_bytes' num 56
#>  $ fitted.values: 'lobstr_bytes' num 280
#>  $ assign       : 'lobstr_bytes' num 56
#>  $ qr           : 'lobstr_bytes' num 2096
#>  $ df.residual  : 'lobstr_bytes' num 56
#>  $ call         : 'lobstr_bytes' num 560
#>  $ terms        : 'lobstr_bytes' num 40002832
#>  $ model        : 'lobstr_bytes' num 40003576

# terms? model?
str(lm_model$terms)
#> Classes 'terms', 'formula'  language 1 ~ 1
#>   ..- attr(*, "variables")= language list(1)
#>   ..- attr(*, "factors")= int(0) 
#>   ..- attr(*, "term.labels")= chr(0) 
#>   ..- attr(*, "order")= int(0) 
#>   ..- attr(*, "intercept")= int 1
#>   ..- attr(*, "response")= int 1
#>   ..- attr(*, ".Environment")=<environment: 0x7ff86949f448> 
#>   ..- attr(*, "predvars")= language list(1)
#>   ..- attr(*, "dataClasses")= Named chr "numeric"
#>   .. ..- attr(*, "names")= chr "1"

# attributes?
str(map(attributes(lm_model$terms), obj_size), digits.d = 10)
#> List of 10
#>  $ variables   : 'lobstr_bytes' num 224
#>  $ factors     : 'lobstr_bytes' num 48
#>  $ term.labels : 'lobstr_bytes' num 48
#>  $ order       : 'lobstr_bytes' num 48
#>  $ intercept   : 'lobstr_bytes' num 56
#>  $ response    : 'lobstr_bytes' num 56
#>  $ class       : 'lobstr_bytes' num 176
#>  $ .Environment: 'lobstr_bytes' num 40000216
#>  $ predvars    : 'lobstr_bytes' num 224
#>  $ dataClasses : 'lobstr_bytes' num 336

# environment..........
lm_model_env <- attr(lm_model$terms, ".Environment")
env_names(lm_model_env)
#> [1] "x"

# `x` was carried along!
obj_size(lm_model_env$x)
#> 40,000,048 B

# saves to 40+Mb, or 20+Mb with compression when it "should" only be 0.01 Mb
# (would be 80Mb, but `x` is only saved once, I think)
# saveRDS(lm_model, "~/Desktop/lm.rds", compress = FALSE)