avoid structure()

[MichaelChirico]

structure() is IME always slower (admittedly on very small absolute scale) than the alternatives, with little tradeoff of readability (if anything I think using setters is more canonical -> more readable). E.g.

library(microbenchmark)

l <- list()

f1 <- function(l) structure(l, class = c("a", "b"), name = "nm")
f2 <- function(l) {
  class(l) <- c("a", "b")
  attr(l, "name") <- "nm"
  l
}

microbenchmark(times = 1e6, f1(l), f2(l))
# Unit: microseconds
#   expr   min    lq     mean median    uq      max neval cld
#  f1(l) 5.432 5.746 7.478789  5.895 6.280 74773.89 1e+06   b
#  f2(l) 1.469 1.616 2.235726  1.681 1.795 69402.46 1e+06  a 

identical(f1(l), f2(l))
# [1] TRUE

[codecov-commenter]

Codecov Report

Merging #2227 (7d28f52) into main (9c49c6e) will increase coverage by 0.00%.
The diff coverage is 100.00%.

❗ Current head 7d28f52 differs from pull request most recent head f6d6810. Consider uploading reports for the commit f6d6810 to get more accurate results

@@           Coverage Diff           @@
##             main    #2227   +/-   ##
=======================================
  Coverage   99.65%   99.65%           
=======================================
  Files         114      114           
  Lines        5173     5175    +2     
=======================================
+ Hits         5155     5157    +2     
  Misses         18       18           

Files	Coverage Δ
R/utils.R	99.13% <100.00%> (+0.01%)	⬆️

[klmr]

This feels very subjective. For what it’s worth, I find the immutable logic using structure() clearer and thus preferable to the reassignment with replacement functions.

I don’t object to providing the lint (after all, many lints are subjective and they are configurable) but I am surprised that this one is being made part of the default list of undesirable functions, since it does not share the objectively objectionable characteristics of the other functions in that list (such as causing global side effects) — instead, it is an entirely subjective preference as far as I can tell.

(And regarding the huge performance difference, this feels worth filing a bug report to https://bugs.r-project.org/ — the difference should not be that big, or exist at all.)