change docker setup to install released versions of hubverse packages

[matthewcornell]

From @zkamvar 's review in PR [clean up renv usage to fix incompatibility between stored CRAN repositories and renv version. update README.md files. regenerate renv.lock files #29]:

---

There are a couple of issues that still need to be addressed after this PR:

1. The installations of the GitHub packages from reichlab negates the effort to install released versions of hubverse packages because they always update to the latest version (as the Remotes: field always points to the main branch). This cannot be solved by using renv::install(). The code in the development are considered stable, but it does not match what the models will be validated against.
2. The workflow for building and deploying the docker containers is a manual one and can be automated, but this would require tests to confirm that it works.

Both 1 and 2 are important in the event that the containers need to be rebuilt or updated (e.g. a severe bug in any of the software used or a change to the hub requires newer versions of hubverse packages).

---

[matthewcornell]

from @zkamvar in slack:

install2.r --skipinstalled --error --repo getOption remotes # install remotes for the github installer

# install hubverse packages and all dependencies (includes cran zoltr, but we will update that later)
install2.r --skipinstalled --error \
  --repos https://hubverse-org.r-universe.dev/ \
  --repos getOption \
  hubData \
  hubVis \
  hubEnsembles

# install up-to-date reichlab packages
installGithub.r \
  --update FALSE \
  --repos getOption \
  reichlab/zoltr \
  reichlab/covidData \
  reichlab/idforecastutils

# remove detritus
rm -rf /tmp/downloaded_packages \
    && strip /usr/local/lib/R/site-library/*/libs/*.so

[matthewcornell]

Additional thoughts:

• The above could be run via a model-specific bash script. This would guarantee that we get the released versions of github-based repos. This is the recommended next step, per @zkamvar .
• Alternatively, we could add a per-model DESCRIPTION file and use something like https://desc.r-lib.org load it. Additional_repositories: https://hubverse-org.r-universe.org . This would eliminate step 6 in the README.md
• Finally, the reichlab repos could be moved to their own r-universe repo. This would allow the DESCRIPTION file method to work with Additional_repositories .

[zkamvar]

To put a finer point on the third point:

Finally, the reichlab repos could be moved to their own r-universe repo. This would allow the DESCRIPTION file method to work with Additional_repositories.

This also benefits the script I presented in #30 (comment) as it would reduce the script to this

# install hubverse packages and all dependencies (includes cran zoltr, but we will update that later)
install2.r --skipinstalled --error \
  --repos https://hubverse-org.r-universe.dev/ \
  --repos https://reichlab.r-universe.dev/ \
  --repos getOption \
  hubData \
  hubVis \
  hubEnsembles \
  zoltr \
  covidData \
  idforecastutils

# remove detritus
rm -rf /tmp/downloaded_packages \
    && strip /usr/local/lib/R/site-library/*/libs/*.so

An example DESCRIPTION file would look something like this:

Package: flu_ar2
Title: Flusight AR2 (not a package)
Version: 1.0.0
Authors@R: c(person(,"Infectious Disease Modeling Consortium", c("cph")))
Maintainer: Matt <[email protected]>
Description: This is not a package. It contains dependency requirements for the flu_ar2 model
License: MIT
Imports:
    hubData
    hubVis
    hubEnsembles
    zoltr
    covidData
    idforecastutils
Additional_repositories: https://hubverse-org.r-universe.dev, https://reichlab.r-universe.dev
Encoding: UTF-8