.rang_query_sysreqs is kind of slow with Github packages

[chainsawriot]

https://github.com/chainsawriot/rang/blob/22b30f31f302b75c691c77de713bd1b8461b8f72/R/resolve.R#L338-L341

Maybe a better approach is to separate Github and CRAN packages and do the query, rather than switch to safe when there is just one Github package (usually).

[schochastics]

R> system.time(rang::resolve("schochastics/rtoot"))
   user  system elapsed 
  7.894   0.828  47.624 
R> system.time(rang::resolve("rtoot"))
   user  system elapsed 
  0.233   0.001   1.333 

That's actually quite substantial! Have you identified a bottleneck? I feel like it wasn't that slow a day ago or so. nope was slow before

[chainsawriot]

x <- resolve("schochastics/rtoot", get_sysreqs = FALSE)
system.time(.rang_query_sysreqs(x))
##   user  system elapsed 
##  3.601   0.371  30.717 

y <- resolve("rtoot", get_sysreqs = FALSE)
system.time(.rang_query_sysreqs(y))
##   user  system elapsed 
##  0.069   0.004   0.677 

[chainsawriot]

system.time(.system_requirements_github("schochastics/rtoot", os = "ubuntu-20.04"))
##   user  system elapsed 
##  0.205   0.008   1.339 

@schochastics The exact reason is that all CRAN packages are queried one by one when there is a Github package, rather than in one query as in the all-CRAN case.

[schochastics]

@chainsawriot finally reached that conclusion too now.

I think we could speed it up by grouping the refs?

.query_sysreqs_safe_grps <- function(targets, os = "ubuntu-20.04") {
  grps <- list("cran" = c(), "github" = c())
  refs <- vapply(targets, .parse_pkgref, return_handle = FALSE, character(1))
  for (i in names(grps)) {
    grps[[i]] <- targets[refs == i]
  }

  tryCatch({
    cran_handles <- vapply(grps[["cran"]], .parse_pkgref, character(1), return_handle = TRUE, USE.NAMES = FALSE)
    output <- .system_requirements_cran(handle = cran_handles, os = os)
  }, error = function(e) {
    output <- .query_sysreqs_safe(targets = targets, os = os)
  })
  
  for (pkg in grps[["github"]]) {
    output <- c(output, .system_requirements_gh(.parse_pkgref(pkg), os))
  }
  return(unique(output))
}

targets <- c("github::schochastics/rtoot", "cran::tibble", "cran::vctrs", 
               "cran::glue", "cran::cli", "cran::rlang", "cran::pkgconfig", 
               "cran::pillar", "cran::utf8", "cran::magrittr", "cran::lifecycle", 
               "cran::fansi", "cran::jsonlite", "cran::httr", "cran::R6", "cran::openssl", 
               "cran::askpass", "cran::sys", "cran::mime", "cran::curl", "cran::dplyr", 
               "cran::tidyselect", "cran::withr", "cran::generics", "cran::clipr")

system.time(a <- .query_sysreqs_safe(targets))
   user  system elapsed 
  3.067   0.451  20.978 
system.time(b <- .query_sysreqs_safe_grps(targets))
   user  system elapsed 
  0.406   0.021   2.442 
sort(a)==sort(b)
[1] TRUE TRUE TRUE TRUE TRUE

The function can be easily adapted to include other refs.
Any potential caveats? If not, I could work this in

[chainsawriot]

7bbf7a7

system.time(rang::resolve("schochastics/rtoot"))
##   user  system elapsed 
##  1.181   0.018   9.084 

Pre

system.time(rang::resolve("schochastics/rtoot"))
   user  system elapsed 
  4.419   0.387  38.502 

[schochastics]

perfect :)