devtools::install() is not working if Bioconductor is not available

[a-maldet]

If Bioconductor is not available, then devtools::install() will run into the error caused by this issue r-lib/pak#827
Setting

options(pkg.use_bioconductor = FALSE)
Sys.setenv(PKG_USE_BIOCONDUCTOR = "FALSE")

does not make any difference. Can anybody help? Thanks.

[jennybc]

I have successfully used devtools::install() lately, leading up to the release, for example.

Are you saying devtools::install() categorically never works for you? Or is there more detail you can add?

[BastienFR]

I have the same problem. I have an installation script that I've been running regularly for the last two years on different machine. The script uses devtools::install pointing at a local folder. I've tried it today and now I get this error:

devtools:install("My_folder_name")
--- pak::local_install_deps()--------
Error:
! error in pak subprocess
Caused by error in `download.file(url, tmp, quiet = TRUE)`:
! cannot open URL 'http://bioconductor.org/config.yaml'
---

I work in a close environment without internet.

What is the point of devtools pinging bioconductor? Is there a way to prevent this so the installation work offline? Thanks.

[jennybc]

@gaborcsardi Do you have insight into this? It seems to be another knock-on effect of using pak inside of devtools::install(). Specifically, pak::local_install_deps().

As OP suggests, it sure sounds like r-lib/pak#827, which might be solved by r-lib/pkgcache#134?

[a-maldet]

@jennybc I have been developing R packages for several years now, working at a national statistics department behind a fairly strict proxy with a local CRAN mirror, without admin privileges… I suspect there are other package developers in corporate environments who face similar issues as me. So far, I’ve been able to work around most proxy issues that popped up, but this one is really a deal-breaker, since options(pkg.use_bioconductor = FALSE) doesn’t work and my IT department won’t budge and add Bioconductor to the whitelist. I have debugged devtools::install() and traced the issue back to the call of pak::local_install_deps() in R/install.R line 88).
In my opinion, install() should stick to its primary task and not require a whitelist of non-essential resources.

For me the pull request r-lib/pkgcache#134 does not solve the problem, since the new mirror won't be whitelisted neither and in my opinion the pull request misses the point of the issue. The bug is that options(pkg.use_bioconductor = FALSE) is not working as specified.
My best hope is r-lib/pak#852

Thanks for your help.

[gaborcsardi]

Fixed in dev pak, from tomorrow's nightly builds.

[a-maldet]

@gaborcsardi Thank you very much!