pkg_deps gets wrong dependencies (for past versions)

[eliotmcintire]

trafficstars

Struggling with several issues in pak recently. This one seems like a simple bug to recreate. The code below requests specific versions of the package BalancedSampling. It shows Rcpp has a minimum version, which appears to be incorrect based on the DESCRIPTION file in the respective archives. The version 1.6.3 does not have Rcpp >= 1.0.12 but pkg_dep says it does.

library(dplyr)

# reprex
pkg <- "BalancedSampling"
ver1 <- "1.6.3"
ver2 <- "2.0.5"

# empty lists to fill
deps <- list()
pakVers <- list()
manual <- list()

deps[[ver1]] <- pak::pkg_deps(paste0(pkg, "@", ver1))
deps[[ver2]] <- pak::pkg_deps(paste0(pkg, "@", ver2))

# pkg_dep gives them the same minimum requirement for ver1 and ver2
pakVers[[ver1]] <- deps[[ver1]] |> filter(ref == "Rcpp") |> select(version) # 1.0.12
pakVers[[ver2]] <- deps[[ver2]] |> filter(ref == "Rcpp") |> select(version) # 1.0.12

# Check with manual downloads
# The two have different minimum requirements for Rcpp

# ver1
setwd(tempdir())
tf <- tempfile(".tar.gz")
fl <- file.path(pkg, "DESCRIPTION")
download.file(file.path(contrib.url(getOption("repos")), 
                        "Archive", pkg, paste0(pkg, "_", ver1, ".tar.gz")), destfile = tf)
untar(tf, files = fl)
manual[[ver1]] <- utils::packageDescription(pkg, lib.loc = ".", field = c("Imports"))

# Check with manual download for ver2
setwd(tempdir())
tf <- tempfile(".tar.gz")
fl <- file.path(pkg, "DESCRIPTION")
download.file(file.path(contrib.url(getOption("repos")), 
                        "Archive", pkg, paste0(pkg, "_", ver2, ".tar.gz")), destfile = tf)
untar(tf, files = fl)
manual[[ver2]] <- utils::packageDescription(pkg, lib.loc = ".", field = c("Imports"))

manual # ver1 has 0.11.1; ver2 has 1.0.12
pakVers   # ver1 has 1.0.12; ver2 has 1.0.12

[gaborcsardi]

Seems correct to me:

❯ pak::pkg_deps("[email protected]")
# A data frame: 3 × 32
  ref     type  direct directpkg status package version license needscompilation
  <chr>   <chr> <lgl>  <lgl>     <chr>  <chr>   <chr>   <chr>   <lgl>
1 Balanc… stan… TRUE   TRUE      OK     Balanc… 1.6.3   GPL (>… TRUE
2 Rcpp    stan… FALSE  FALSE     OK     Rcpp    1.0.12  GPL (>… FALSE
3 Sampli… stan… FALSE  FALSE     OK     Sampli… 1.0.0   GPL (>… FALSE
# ℹ 23 more variables: priority <chr>, md5sum <chr>, sha256 <chr>,
#   filesize <int>, built <chr>, platform <chr>, rversion <chr>,
#   repotype <chr>, repodir <chr>, target <chr>, deps <list>, mirror <chr>,
#   sources <list>, remote <list>, error <list>, metadata <list>,
#   dep_types <list>, params <list>, sysreqs <chr>, cache_status <chr>,
#   lib_status <chr>, old_version <chr>, new_version <chr>

❯ pak::pkg_deps("[email protected]")$deps[[1]]
# A data frame: 3 × 5
  ref             type      package         op    version
  <chr>           <chr>     <chr>           <chr> <chr>
1 Rcpp            Imports   Rcpp            ">="  "0.11.1"
2 SamplingBigData Imports   SamplingBigData ""    ""
3 Rcpp            LinkingTo Rcpp            ""    ""

[gaborcsardi]

# pkg_dep gives them the same minimum requirement for ver1 and ver2

That's not a minimum requirement, pkg_deps() simply resolves all dependencies and lists them in a data frame. It uses the latest versions of the dependencies as much as possible, by default.

[eliotmcintire]

So, the reported version in the column labelled "version" in the returned data.frame is not the required version, it is the "current" version?

The documentation is ambiguous about this:

A data frame with the dependency data
... 
version: package version

Why would it return "dependency data" in the version column that is different than the deps column?

[gaborcsardi]

pkg_deps() looks up the dependencies of a package and returns data about them. It basically does the same as pkg_install(), but stops before the installation.

[gaborcsardi]

AFAICT there is nothing for me to do here, but please reopen if you feel otherwise. Thanks!