bioconductor repos are not automatically included

rsconnect uses packrat::.snapshotImpl to record dependencies for a project without having packages installed into packrat/lib. If BiocInstaller is available, it is first loaded via requireNamespace.

https://github.com/rstudio/rsconnect/blob/13a515e5401ca2c7ca990f1aca432db47ecd5d5c/R/bundle.R#L631-L643

Start with the default repository configuration in the RStudio IDE.

getOption("repos")
                       CRAN 
# => "https://cran.rstudio.com/" 
# => attr(,"RStudio")
# => [1] TRUE

Given a Shiny application in a pristine directory (no packrat directory):

library(shiny)
# Biobase depends upon BioGenerics, so we have a top-level and dependent
# package coming from Bioconductor.
library(Biobase)

server <- function(input, output, session) {}
ui <- fluidPage("bio shiny")

shiny::shinyApp(ui = ui, server = server)

Running a snapshot in the containing directory:

packrat::.snapshotImpl(project = ".", snapshot.sources = FALSE, fallback.ok = TRUE, verbose = FALSE)

The packrat.lock file produced has the preamble:

PackratFormat: 1.4
PackratVersion: 0.4.9.13
RVersion: 3.5.1
Repos: CRAN=https://cran.rstudio.com/

This is because the activeRepos function looks in the project library for BiocInstaller and does not allow for projects that have not installed their dependencies (as is the case when used for rsconnect dependency analysis).

https://github.com/rstudio/packrat/blob/4a83b7fcc5c0c92c575c40bbd8a8aba746dd66c3/R/snapshot.R#L352-L366

Workaround is to explicitly configure the repos option to the bioc repos:

options(repos = BiocInstaller::biocinstallRepos())

Start with the activeRepos implementation in R/snapshot.R and the dependency snapshot path referenced in R/bundle.R. Reproduce the provided pristine Shiny application case with Biobase, then inspect the generated packrat.lock repository preamble. Done means Bioconductor repositories are detected and recorded without requiring dependencies to be installed in the project library.