Work around CRAN metadata inconsistencies for macos binaries

[gaborcsardi]

Cf. yihui/litedown#112

• Use PPM by default?
• Use PPM for macOS as well?
• Implement a workaround in pak?

[danielvartan]

I suggest defaulting upgrade to TRUE in setup-r-dependencies until this is resolved. This appears to bypass the metadata inconsistencies on the macos-latest runner by forcing a fresh resolution.

[gaborcsardi]

@danielvartan That creates other problems, because it'll compile newly released packages from source, and that fails for packages that have system requirements.

[gaborcsardi]

yihui/litedown#112 (comment)

It is unclear if the underlying issue is really fixed or only this instance of the inconsistency. We can wait a bit to see if it crops up again before implementing a tedious workaround in pak.

[tomshafer]

In case it's helpful to know, I have just encountered the same issue with a different package (macos-latest runner). All works fine for Windows and Ubuntu.

...
  ✔ Got xml2 1.5.1 (aarch64-apple-darwin20) (984.25 kB)
  ✔ Got viridisLite 0.4.2 (aarch64-apple-darwin20) (1.30 MB)
  ✔ Got yaml 2.3.12 (aarch64-apple-darwin20) (225.55 kB)
  ✖ Failed to download later 1.4.4 (aarch64-apple-darwin20)
  ✔ Got gh 1.5.0 (aarch64-apple-darwin20) (121.89 kB)
  ✔ Got rstudioapi 0.17.1 (aarch64-apple-darwin20) (317.26 kB)
  ✔ Got lintr 3.3.0-1 (aarch64-apple-darwin20) (1.43 MB)
...

[gaborcsardi]

@tomshafer Can you please link to the failing build? Thanks!

[gaborcsardi]

Right now at https://cloud.r-project.org/bin/macosx/big-sur-arm64/contrib/4.5/ the metadata has

Package: later
Version: 1.4.4

But the package file is already 1.4.5:

later_1.4.5.tgz                                2026-01-08 07:02  751K

On mac.r-project.org the metadata is already correct. On cran.r-project.org it is still wrong.

[tomshafer]

@gaborcsardi Here you go: https://github.com/ElderResearch/eriplots-r/actions/runs/20818795022/job/59801343238