feat: slim down minimal image + docs on rocker-versioned2 mix-and-match

.gitignore

@@ -2,3 +2,4 @@
 .Rhistory
 .RData
 .Ruserdata
+*.html

README.Rmd

@@ -116,6 +116,46 @@ Add :tagname after ghcr.io/geocompx/imagename to get the image you want.
 docker run -e PASSWORD=pw --rm -p 8786:8787 ghcr.io/geocompx/buildbook
 ```
 
+### Building leaner custom images with rocker-versioned2
+
+The images above are opinionated "batteries included" builds. For users who want smaller images or a different package selection, the [rocker-versioned2](https://github.com/rocker-org/rocker-versioned2) repository provides modular `install_*.sh` scripts that can be composed on top of a slim base image.
+
+A reference `minimal-slim/Dockerfile` is included in this repo showing how to build a lean geospatial R image from `rocker/r-ver` without RStudio or heavy CLI tools.
+
+For example, to build a minimal R + geospatial image yourself:
+
+```Dockerfile
+FROM rocker/r-ver:4.4.2
+RUN /rocker_scripts/install_geospatial.sh
+RUN /rocker_scripts/install_rstudio.sh
+```
+
+Or to add only what you need:
+
+```Dockerfile
+FROM rocker/r-ver:4.4.2
+RUN apt-get update && apt-get install -y --no-install-recommends \
+    libgdal-dev libgeos-dev libproj-dev \
+    && rm -rf /var/lib/apt/lists/*
+RUN install2.r sf terra spData
+```
+
+Key [`rocker-versioned2` scripts](https://github.com/rocker-org/rocker-versioned2/tree/master/scripts) to mix and match:
+
+- [`/rocker_scripts/install_geospatial.sh`](https://github.com/rocker-org/rocker-versioned2/blob/master/scripts/install_geospatial.sh) — GDAL, PROJ, GEOS, and the core R geospatial packages (sf, terra, raster, stars, sp, etc.)
+- [`/rocker_scripts/install_rstudio.sh`](https://github.com/rocker-org/rocker-versioned2/blob/master/scripts/install_rstudio.sh) — RStudio Server (uses S6 supervisor, ~300MB)
+- [`/rocker_scripts/install_python.sh`](https://github.com/rocker-org/rocker-versioned2/blob/master/scripts/install_python.sh) — Python 3 + reticulate + venv
+- [`/rocker_scripts/install_tidyverse.sh`](https://github.com/rocker-org/rocker-versioned2/blob/master/scripts/install_tidyverse.sh) — The tidyverse stack
+- [`/rocker_scripts/install_quarto.sh`](https://github.com/rocker-org/rocker-versioned2/blob/master/scripts/install_quarto.sh) — Quarto CLI
+
+#### Suggested improvements upstream (for the Rocker project)
+
+1. **Provide a "geospatial-minimal" variant** in `rocker-versioned2` that omits RStudio Server and the heavy R geospatial package set, giving just the system libraries (GDAL, PROJ, GEOS, etc.) on top of `r-ver`. Users could then `install2.r sf terra` on top of a ~1–2GB base.
+2. **Use a smaller base image** (e.g., `debian:bookworm-slim` or `ubuntu:22.04`) rather than the full `rocker/verse` for images that don't need TeX Live, pandoc, or Jupyter.
+3. **Prune base image layers** — the [`books/rocker` best practices](https://rocker-project.org/use/extending.html) already recommend `--no-install-recommends` and `rm -rf /var/lib/apt/lists/*`, but the base images themselves could be rebuilt with multi-stage builds to discard build tools after package compilation.
+4. **Make stripping of RSPM binary libraries opt-out** rather than opt-in (see [rocker-versioned2#340](https://github.com/rocker-org/rocker-versioned2/issues/340)) — this can save 20–30% on R package image size with no observed runtime cost.
+5. **Offer a "pruned" `rocker/geospatial` tag** that has had LaTeX, pandoc, and Java removed (the largest contributors to image bloat) for users who only need the geospatial C libraries + R bindings.
+
 ## Examples
 
 ### osgeo

README.md

@@ -138,6 +138,77 @@ Add :tagname after ghcr.io/geocompx/imagename to get the image you want.
 docker run -e PASSWORD=pw --rm -p 8786:8787 ghcr.io/geocompx/buildbook
 ```
 
+### Building leaner custom images with rocker-versioned2
+
+The images above are opinionated “batteries included” builds. For users
+who want smaller images or a different package selection, the
+[rocker-versioned2](https://github.com/rocker-org/rocker-versioned2)
+repository provides modular `install_*.sh` scripts that can be composed
+on top of a slim base image.
+
+A reference `minimal-slim/Dockerfile` is included in this repo showing
+how to build a lean geospatial R image from `rocker/r-ver` without
+RStudio or heavy CLI tools.
+
+For example, to build a minimal R + geospatial image yourself:
+
+``` dockerfile
+FROM rocker/r-ver:4.4.2
+RUN /rocker_scripts/install_geospatial.sh
+RUN /rocker_scripts/install_rstudio.sh
+```
+
+Or to add only what you need:
+
+``` dockerfile
+FROM rocker/r-ver:4.4.2
+RUN apt-get update && apt-get install -y --no-install-recommends \
+    libgdal-dev libgeos-dev libproj-dev \
+    && rm -rf /var/lib/apt/lists/*
+RUN install2.r sf terra spData
+```
+
+Key [`rocker-versioned2`
+scripts](https://github.com/rocker-org/rocker-versioned2/tree/master/scripts)
+to mix and match:
+
+- [`/rocker_scripts/install_geospatial.sh`](https://github.com/rocker-org/rocker-versioned2/blob/master/scripts/install_geospatial.sh)
+  — GDAL, PROJ, GEOS, and the core R geospatial packages (sf, terra,
+  raster, stars, sp, etc.)
+- [`/rocker_scripts/install_rstudio.sh`](https://github.com/rocker-org/rocker-versioned2/blob/master/scripts/install_rstudio.sh)
+  — RStudio Server (uses S6 supervisor, ~300MB)
+- [`/rocker_scripts/install_python.sh`](https://github.com/rocker-org/rocker-versioned2/blob/master/scripts/install_python.sh)
+  — Python 3 + reticulate + venv
+- [`/rocker_scripts/install_tidyverse.sh`](https://github.com/rocker-org/rocker-versioned2/blob/master/scripts/install_tidyverse.sh)
+  — The tidyverse stack
+- [`/rocker_scripts/install_quarto.sh`](https://github.com/rocker-org/rocker-versioned2/blob/master/scripts/install_quarto.sh)
+  — Quarto CLI
+
+#### Suggested improvements upstream (for the Rocker project)
+
+1.  **Provide a “geospatial-minimal” variant** in `rocker-versioned2`
+    that omits RStudio Server and the heavy R geospatial package set,
+    giving just the system libraries (GDAL, PROJ, GEOS, etc.) on top of
+    `r-ver`. Users could then `install2.r sf terra` on top of a ~1–2GB
+    base.
+2.  **Use a smaller base image** (e.g., `debian:bookworm-slim` or
+    `ubuntu:22.04`) rather than the full `rocker/verse` for images that
+    don’t need TeX Live, pandoc, or Jupyter.
+3.  **Prune base image layers** — the [`books/rocker` best
+    practices](https://rocker-project.org/use/extending.html) already
+    recommend `--no-install-recommends` and
+    `rm -rf /var/lib/apt/lists/*`, but the base images themselves could
+    be rebuilt with multi-stage builds to discard build tools after
+    package compilation.
+4.  **Make stripping of RSPM binary libraries opt-out** rather than
+    opt-in (see
+    [rocker-versioned2#340](https://github.com/rocker-org/rocker-versioned2/issues/340))
+    — this can save 20–30% on R package image size with no observed
+    runtime cost.
+5.  **Offer a “pruned” `rocker/geospatial` tag** that has had LaTeX,
+    pandoc, and Java removed (the largest contributors to image bloat)
+    for users who only need the geospatial C libraries + R bindings.
+
 ## Examples
 
 ### osgeo

minimal-slim/Dockerfile

@@ -0,0 +1,27 @@
+# Slim custom geospatial image built from rocker/r-ver, without RStudio or heavy CLI tools
+ARG R_VER=4.4.2
+FROM rocker/r-ver:${R_VER}
+
+# System dependencies: keep only what sf/terra/spData need at runtime/build time
+# Omitting: gdal-bin, netcdf-bin, postgis, Jupyter, pandoc, TeX, Java
+RUN apt-get update \
+    && DEBIAN_FRONTEND=noninteractive apt-get install -y --no-install-recommends \
+        libgdal-dev \
+        libgeos-dev \
+        libproj-dev \
+        libudunits2-dev \
+        libssl-dev \
+    && rm -rf /var/lib/apt/lists/*
+
+# R packages: geocompx/geocompkg Imports only (sf, terra, spData, units, dplyr, methods, languageserver)
+# Using install2.r from rocker-versioned2 style (lightweight, no extra system deps)
+ARG NCPUS=-1
+RUN install2.r --error --skipmissing --skipinstalled -n "${NCPUS}" \
+        sf terra spData units dplyr methods languageserver \
+    && rm -rf /tmp/downloaded_packages
+
+# Strip binary .so files from RSPM installs
+RUN find /usr/local/lib/R/site-library/*/libs -name '*.so' -exec strip --strip-unneeded {} + 2>/dev/null || true
+
+# Default command
+CMD ["R"]

minimal/Dockerfile

@@ -1,18 +1,27 @@
 ARG ROCKER_GEOSPATIAL_IMAGE=rocker/geospatial:latest
 FROM ${ROCKER_GEOSPATIAL_IMAGE}
 
-ARG QUARTO_VERSION=1.9.37
+# Drop packages that are not needed for R geospatial work
+# (keep libgdal-dev/libgeos-dev/libproj-dev — needed by sf/terra)
+RUN apt-get update \
+    && DEBIAN_FRONTEND=noninteractive apt-get purge -y postgis netcdf-bin \
+    && apt-get autoremove -y \
+    && apt-get clean \
+    && rm -rf /var/lib/apt/lists/*
 
-# Ship editor defaults with the image so downstream Codespaces/devcontainers
-# inherit Quarto support even when they only reference the published image.
-LABEL devcontainer.metadata='[{"customizations":{"vscode":{"extensions":["quarto.quarto","REditorSupport.r","ms-python.python","ms-toolsai.jupyter"]}}}]'
+# Install only the essential geocompkg Imports (not Suggests) to keep the
+# image lean. The full geocompkg Suggests set (60+ packages) is available
+# via the `suggests` image or users can install on demand.
+RUN R -e "if (!requireNamespace('pak', quietly = TRUE)) install.packages('pak', repos = 'https://cloud.r-project.org/'); pak::pak(c('sf','terra','spData','units','dplyr','languageserver','methods'), upgrade = FALSE)"
 
-RUN R -e "if (!requireNamespace('pak', quietly = TRUE)) install.packages('pak', repos = 'https://cloud.r-project.org/'); pak::pak('geocompx/geocompkg', upgrade = TRUE)"
-# Update quarto to latest stable version:
-RUN /rocker_scripts/install_quarto.sh ${QUARTO_VERSION}
-# Set RStudio preferences
-# No inline code:
-RUN echo '{' >> /etc/rstudio/rstudio-prefs.json
-RUN echo '    "rmd_chunk_output_inline": false' >> /etc/rstudio/rstudio-prefs.json
-RUN echo '}' >> /etc/rstudio/rstudio-prefs.json
+# Strip binary shared libraries from RSPM packages to reduce final image size
+# See https://github.com/rocker-org/rocker-versioned2/issues/340
+RUN strip /usr/local/lib/R/site-library/*/libs/*.so
 
+# Set RStudio preferences — no inline code
+RUN echo '{' > /etc/rstudio/rstudio-prefs.json \
+    && echo '    "rmd_chunk_output_inline": false' >> /etc/rstudio/rstudio-prefs.json \
+    && echo '}' >> /etc/rstudio/rstudio-prefs.json
+
+# Clean up apt caches
+RUN rm -rf /var/lib/apt/lists/*