pak does not use binaries

[janni-hr]

On Posit Workbench, pak does not use binaries even if binaries should be available in theory (PPM setup). We use a private instance of PPM, but I also get this behaviour for the public PPM:

> pak::system_r_platform()
[1] "x86_64-pc-linux-gnu-ubuntu-22.04"
> getOption("repos")
                                                         CRAN 
"https://packagemanager.posit.co/cran/__linux__/jammy/latest" 
> pak::meta_list("glue")[["sources"]]
[[1]]
[1] "https://packagemanager.posit.co/cran/__linux__/jammy/latest/src/contrib/glue_1.8.1.tar.gz"
> pak::meta_list("fs")[["platform"]]
[1] "source"
> pak::cache_clean()
> pak::pkg_install("glue")
✔ Loading metadata database ... done
                                        
→ Will install 1 package.
→ Will download 1 package with unknown size.
+ glue   1.8.1 [bld][cmp][dl]
  
ℹ Getting 1 pkg with unknown size
✔ Got glue 1.8.1 (source) (173.80 kB)                                
✔ Downloaded 1 package (173.80 kB) in 1.2s                           
ℹ Building glue 1.8.1
✔ Built glue 1.8.1 (388ms)                                  
✔ Installed glue 1.8.1  (1.2s)                                
✔ 1 pkg: added 1, dld 1 (173.80 kB) [7.5s]   

pak_siterep():

> pak::pak_sitrep()
* pak version:
- 0.9.4
* Version information:
- pak platform: x86_64-pc-linux-gnu (current: x86_64-pc-linux-gnu, compatible)
- pak repository: - (local install?)
* Optional packages installed:
- pillar
* Library path:
- /nfs-data/rhome/rgi/R/x86_64-pc-linux-gnu-library/4.5
- /opt/R/4.5.2/lib/R/library
* pak is installed at /nfs-data/rhome/rgi/R/x86_64-pc-linux-gnu-library/4.5/pak.
* Dependency versions:
- callr      3.7.6     
- cli        3.6.6     
- curl       5.2.3     
- desc       1.4.3     
- filelock   1.0.3     
- jsonlite   2.0.0     
- keyring    1.4.1     
- lpSolve    5.6.23    
- pkgbuild   1.4.8     
- pkgcache   2.2.5.9000
- pkgdepends 0.9.1.9000
- pkgsearch  3.1.5     
- processx   3.8.7     
- ps         1.9.2     
- R6         2.5.1     
- yaml       2.3.12    
- zip        2.3.3     
* Dependencies can be loaded

[gaborcsardi]

I can't reproduce this:

FROM ghcr.io/r-lib/rig/ubuntu-22.04
RUN R -q -e 'getOption("repos")'
RUN R -q -e 'pak::pkg_install("glue")'

❯ docker build --progress plain --platform linux/amd64 .
#0 building with "rancher-desktop" instance using docker driver

#1 [internal] load build definition from Dockerfile
#1 transferring dockerfile: 145B done
#1 DONE 0.0s

#2 [internal] load metadata for ghcr.io/r-lib/rig/ubuntu-22.04:latest
#2 DONE 0.0s

#3 [internal] load .dockerignore
#3 transferring context: 2B done
#3 DONE 0.0s

#4 [1/3] FROM ghcr.io/r-lib/rig/ubuntu-22.04:latest@sha256:09eeab66070b17b4d086930c3592484816c661fec3fb84c6425fe141670fd78b
#4 resolve ghcr.io/r-lib/rig/ubuntu-22.04:latest@sha256:09eeab66070b17b4d086930c3592484816c661fec3fb84c6425fe141670fd78b 0.0s done
#4 DONE 0.0s

#4 [1/3] FROM ghcr.io/r-lib/rig/ubuntu-22.04:latest@sha256:09eeab66070b17b4d086930c3592484816c661fec3fb84c6425fe141670fd78b
#4 DONE 0.1s

#5 [2/3] RUN R -q -e 'getOption("repos")'
#5 1.615 > getOption("repos")
#5 1.617                                                           P3M
#5 1.618 "https://packagemanager.posit.co/cran/__linux__/jammy/latest"
#5 1.618                                                          CRAN
#5 1.618                                 "https://cloud.r-project.org"
#5 1.618 >
#5 DONE 1.7s

#6 [3/3] RUN R -q -e 'pak::pkg_install("glue")'
#6 1.695 > pak::pkg_install("glue")
#6 8.066
#6 13.02 ✔ Updated metadata database: 6.09 MB in 10 files.
#6 13.03
#6 13.04 ℹ Updating metadata database
#6 64.10 ✔ Updating metadata database ... done
#6 64.11
#6 65.46
#6 65.50 → Will install 1 package.
#6 65.59 → Will download 1 package with unknown size.
#6 65.61 + glue   1.8.1 [dl]
#6 65.62
#6 66.26 ℹ Getting 1 pkg with unknown size
#6 67.49 ✔ Got glue 1.8.1 (x86_64-pc-linux-gnu-ubuntu-22.04) (173.80 kB)
#6 68.17 ✔ Installed glue 1.8.1  (261ms)
#6 68.23 ✔ 1 pkg: added 1, dld 1 (173.80 kB) [1m 6.4s]
#6 68.24 >
#6 DONE 68.5s

[janni-hr]

Thank you. If I configure the repos the same way as you do, it works as expected for the public PPM. However, it does not work for our private instance. I’ve looked into it a bit further and I think it has to do with the URL of our PPM. It has the form

{base}/cran-repo/__linux__/jammy/latest

pkgcache:::is_ppm_linux_repo_url() returns FALSE for our URL, because of the "(?<repo>cran|all)/" in pkgcache:::re_ppm_linux(). And this causes pkgcache:::ppm_binary_url() to return NA.

Could this be the problem? And if yes, is there a way to work around it?

[gaborcsardi]

Yes, that's probably the problem. I can modify the regex, this is probably enough:

grepl(
  "^(?<base>.*/)(?<repo>[^/]+)/__linux__/(?<distro>[a-zA-Z0-9]+)/(?<version>latest|[-0-9]+)$", 
  "{base}/cran-repo/__linux__/jammy/latest", 
  perl = TRUE
)

#> [1] TRUE

I.e. to allow any URL before the __linux__ part.

[gaborcsardi]

Should be fixed in tomorrows nightly devel build, if you can install them on your system. See https://pak.r-lib.org/reference/install.html#nightly-builds

[janni-hr]

It works now as expected with the night devel build. Thank you!

[gaborcsardi]

This is now released in pak 0.9.5.