fix(setup-r): install resolute binaries on the Ubuntu 26.04 runner

.github/actions/setup-r/action.yml

@@ -10,7 +10,7 @@ inputs:
     description: "R version to install"
     required: false
     # Bump together with the snapshot date below: Posit Package Manager serves
-    # noble binaries per R minor version, so a new R needs a snapshot that has them.
+    # resolute binaries per R minor version, so a new R needs a snapshot that has them.
     default: "4.6.1"
 
 runs:
@@ -49,9 +49,13 @@ runs:
       # install.packages auto-resolves transitive dependencies that a hand-written
       # renv.lock used to miss (munsell, gtable, colorspace, R6, RColorBrewer, ...).
       # Pinning to a dated Posit Package Manager snapshot keeps versions
-      # reproducible without needing a fully-populated lockfile. The `noble` URL
-      # matches the current ubuntu-latest runner image — update if GitHub moves
-      # the runner OS, and bump the snapshot date deliberately.
+      # reproducible without needing a fully-populated lockfile. The `resolute`
+      # URL matches the ubuntu-latest runner image (Ubuntu 26.04 since October
+      # 2026). Change the codename whenever GitHub moves the runner OS: Posit
+      # builds each binary against one release's system libraries, and the
+      # former `noble` (24.04) binaries installed fine on 26.04 but stringi
+      # failed to load there (it links libicuuc.so.74; 26.04 ships ICU 78).
+      # Bump the snapshot date deliberately.
       #
       # Snapshot 2026-10-20 ships ggplot2 4.0.3. The repository has no R lock
       # file, so this line stands in for one: core/library_versions.py reads
@@ -60,7 +64,7 @@ runs:
       # tests/unit/core/test_version_sources.py until this line names the ggplot2
       # version the new snapshot installs (packageVersion("ggplot2") in a run).
       #
-      # Every package below installs from this snapshot as a noble binary for
+      # Every package below installs from this snapshot as a resolute binary for
       # R 4.6.1. ggplot2 4.0 changes rendering defaults; the changes that matter
       # to generated code are listed in prompts/library/ggplot2.md.
       #
@@ -73,7 +77,7 @@ runs:
       run: |
         Rscript -e '
           options(
-            repos = c(CRAN = "https://packagemanager.posit.co/cran/__linux__/noble/2026-10-20"),
+            repos = c(CRAN = "https://packagemanager.posit.co/cran/__linux__/resolute/2026-10-20"),
             Ncpus = parallel::detectCores()
           )
           install.packages(
@@ -85,10 +89,15 @@ runs:
         '
 
     - name: Smoke-test ggplot2 renders a PNG
+      # Loading tidyr and stringr also loads stringi, whose binary links the
+      # distro ICU by soname; str_detect() then calls into it. A Posit binary
+      # built for another Ubuntu release installs without error and fails only
+      # here, at load time, instead of in the middle of a generation run.
       shell: bash
       run: |
         Rscript -e '
-          library(ggplot2); library(ragg)
+          library(ggplot2); library(ragg); library(tidyr); library(stringr)
+          stopifnot(str_detect("anyplot", "plot"))
           tmp <- tempfile(fileext = ".png")
           ggsave(tmp, ggplot(mtcars, aes(wt, mpg)) + geom_point(),
                  device = ragg::agg_png, width = 4, height = 3, dpi = 100)

changelog.d/setup-r-resolute.md

@@ -0,0 +1,12 @@
+### Fixed
+
+- **The R setup installs packages built for the Ubuntu 26.04 runner image.** The
+  setup-r action installed Posit's `noble` (Ubuntu 24.04) binaries. On the 26.04
+  image they install without error, but stringi links ICU 74 while 26.04 ships
+  ICU 78, so loading stringr fails. No catalogue implementation reaches stringi
+  today: all 325 render the same on both builds. A generated implementation that
+  used stringr, or tidyr's `separate()` or `extract()`, would have failed
+  mid-run. The action now installs the `resolute` binaries from the same
+  2026-10-20 snapshot. Its smoke test also loads tidyr and stringr, so a binary
+  built for another Ubuntu release fails the setup step instead of a generation
+  run. (#12263)

tests/unit/core/test_version_sources.py

@@ -135,7 +135,7 @@ def test_uv_lock_with_a_duplicated_block_does_not_resolve(tmp_path: Path) -> Non
 SETUP_R = """
       # Snapshot {note} ships ggplot2 3.5.1. The repository has no R lock file.
       run: |
-        Rscript -e 'options(repos = c(CRAN = "https://packagemanager.posit.co/cran/__linux__/noble/{url}"))'
+        Rscript -e 'options(repos = c(CRAN = "https://packagemanager.posit.co/cran/__linux__/resolute/{url}"))'
 """