[r] Controlled downgrade of TileDB-R to ensure matching TileDB Core

[eddelbuettel]

Issue and/or context:

The TileDB-R package builds against TileDB Core, and while we can ensure a minimum version
we cannot ensure an exact version. That can lead to cases as right now where TileDB-R 0.23.0 is
at CRAN with a default of TileDB Core 2.19.0, yet TileDB-SOMA is still set up to build with TileDB
Core 2.18.2. This mismatch is undesirable but a little difficult to avoid for example in continuous
integration where R will always go to the newest package version.

As an experiment, @mojaveazure and I have set up two helper repos. The first ensures we can still build
the previous TileDB-R version at r-universe (which would otherwise pick newest versions from
CRAN and/or GitHub). The second deploys these builds in a particular R repository hosted on
GitHun from which we can install. Together, these provide a simple 'time machine' ability to
the previous binary.

To illustrate, by default we end with these packages (when running on Ubuntu 22.04 as for CI)

  > show_package_versions()
  tiledbsoma:    1.6.0
  tiledb-r:      0.23.0
  tiledb core:   2.19.0
  libtiledbsoma: 2.18.2
  R:             R version 4.3.2 (2023-10-31)
  OS:            Ubuntu 22.04.1 LTS
  >

Whereas once we do a controlled downgrade we get

  > library(tiledbsoma)
  > show_package_versions()
  tiledbsoma:    1.6.0
  tiledb-r:      0.22.0
  tiledb core:   2.18.2
  libtiledbsoma: 2.18.2
  R:             R version 4.3.2 (2023-10-31)
  OS:            Ubuntu 22.04.1 LTS
  >

Changes:

The PR adds a new helper script and invokes it from the r-valgrind action for continuous integration.

Notes for Reviewer:

SC 38681

[shortcut-integration]

This pull request has been linked to Shortcut Story #38681: Add controlled downgrade of TileDB-R (in CI) to ensure matching TileDB Core versions.

[codecov]

Codecov Report

Merging #1994 (11b37ac) into main (bdff402) will decrease coverage by 13.11%.
Report is 1 commits behind head on main.
The diff coverage is n/a.

Additional details and impacted files

@@             Coverage Diff             @@
##             main    #1994       +/-   ##
===========================================
- Coverage   89.07%   75.96%   -13.11%     
===========================================
  Files          34      126       +92     
  Lines        3578    10049     +6471     
  Branches        0      161      +161     
===========================================
+ Hits         3187     7634     +4447     
- Misses        391     2340     +1949     
- Partials        0       75       +75     

Flag	Coverage Δ
libtiledbsoma	69.86% <ø> (?)
python	89.73% <ø> (+0.65%)	⬆️
r	67.38% <ø> (?)

Flags with carried forward coverage won't be shown. Click here to find out more.

Components	Coverage Δ
python_api	89.73% <ø> (+0.65%)	⬆️
libtiledbsoma	53.52% <ø> (∅)

[johnkerl]

On #1993 I found it helpful to include

cat("core_from_tiledb", as.character(core_from_tiledb), "\n")
cat("core_from_soma", core_from_soma, "\n")

[eddelbuettel]

Yes -- I developed this in another repo first (to keep things calmer here) and did similar things there.

[johnkerl]

Controlled downgrade is effective and in place as a repeatable process. I have had zero problems with this since then in CI, and releases are working well in this regard.

• [r/ci] Controlled downgrade for TileDB-R 0.23 #2157 for TileDB-R 0.23
• [r/ci] Controlled downgrade for TileDB-R 0.24 #2295 for TileDB-R 0.24
• We will continue this patter for 0.25 (once 0.26 comes to exist), and so on

[eddelbuettel]

I don't have the energy to debate this endlessly but we had reports of failed installations just days ago. What we set up works if and only if the stars align and tiledb-r is not ahead. I maintain what we can do better.

[johnkerl]

@eddelbuettel these are all for CI, and CI works. When @nguyenv is done with C++-ification and removing our dependency on TileDB-R (and TileDB-Py), this problem will cease to exist.

[eddelbuettel]

We had reports from users being unable to install. That is not CI.