On 29/04/2025 22:54, Dirk Eddelbuettel wrote:
That is not a bad analogy! And it is something one could ask upstream to set. In practice, R upstream is done by a team which has its ways and changing something that has worked for decades (i.e.: no encoded soname) is not all that likely to change.
I wasn't intending to suggest such a thing, only to check my understanding of what r-api is.
| While dh-r is currently missing the equivalent of a shlibs version | (packages built with the *new* version may not work in the *old* version; | newly built packages get a Depends: on at least this version)? The packages I maintain have always added the Depends
...but packages using dh-r don't, which is what I'm proposing to change.
Ok, I will upload a fresh 4.5.0-3 with the Breaks you argued for. Fingers crossed!
Thanks. It has found some potential problems, but much of the point was to find these now and not when it's too late to fix them:
The package set from unstable it's trying in autopkgtest is bigger than I expected, and includes at least 2 packages that currently aren't in testing at all (r-cran-cards and r-cran-dendextend) which means it won't actually migrate that set; I don't yet know why.
r-bioc-stringdb/ppc64el, r-bioc-megadepth/ppc64el - look like random network issues, retried
r-cran-broom.helpers/multiple - is trying to migrate this package as well (unsure why, I didn't intend that), which fails because the version from unstable depends on packages that are not in testing
r-cran-bslib/multiple, r-cran-rmarkdown/multiple - Error in `compile_data(sass_input, options)`: Error: Undefined variable: "$transition-collapse-width". Probably not R 4.5's fault (it has previously passed with R 4.5) but a bug whatever causes it